Verizon and Ericsson Advance AI-Native RAN Operations with Nationwide EIC Rollout

Posted on by Alan Weissberger

Executive Summary:

Verizon and Ericsson are expanding their longstanding collaboration to introduce AI-native automation into Radio Access Network (RAN) operations. The companies plan a nationwide deployment of Ericsson Intelligent Controller (EIC) [1.], a software platform for running RAN applications (rApps) co-developed by Ericsson and Verizon.

The initiative is intended to improve mobile network performance, automate network optimization and advance Verizon’s longer-term objective of achieving Level 4 (L4) network autonomy. The companies have demonstrated EIC and associated rApps in a live network environment, using AI-based cell shaping and traffic steering to optimize Massive Multiple-Input Multiple-Output (Massive MIMO) antenna operation.

According to Ericsson and Verizon, field deployment of the Advanced Antenna System (AAS) Cell Shaper rApp produced throughput gains of up to 8.9% in downlink data radio bearer (DRB) throughput and up to 8% in uplink user equipment (UE) throughput.

The collaboration has three principal objectives:

  • RAN autonomy: Building on open, programmable interfaces to automate network operations and optimization.
  • AI-based optimization: Deploying rApps to improve radio performance, traffic distribution and operational efficiency.
  • Operator-controlled innovation: Enabling Verizon to develop and deploy its own rApps alongside Ericsson applications rather than relying exclusively on vendor-developed network optimization software.

Note 1. “Ericsson Intelligent Controller (EIC) is a new product part of the Ericsson Intelligent Automation Platform (EIAP) which is Ericsson’s Open RAN aligned service management and orchestration (SMO) solution. EIAP provides multivendor RAN automation, management, and orchestration of Cloud and Open Radio access networks. Ericsson differentiator is the extended support to purpose-built radio access networks.”  SOURCE: Ericsson

Image generated by ChatGPT which was also used for research and writing this article.

………………………………………………………………………………………………………………………………………………………………………………………………………….

AI-Native RAN Automation and Massive MIMO Optimization:

A central component of the collaboration is the application of AI to Massive MIMO antenna systems.  Massive MIMO uses large antenna arrays at 5G base stations to transmit and receive multiple spatially separated data streams. Through beamforming, the radio system can concentrate transmitted energy toward individual users or groups of users rather than radiating power uniformly in all directions. Combined with spatial multiplexing, this can improve spectral efficiency, increase network capacity and extend coverage.

However, optimizing Massive MIMO performance across a large mobile network is complex. Radio propagation conditions, user locations, traffic demand, interference, antenna configuration and mobility patterns change continuously. A configuration that improves performance in one cell or at a particular time of day may not be optimal under different network conditions.

Ericsson’s Advanced Antenna System (AAS) Cell Shaper rApp applies AI-based optimization to antenna operation and cell coverage. Its purpose is to adjust radio coverage characteristics and traffic distribution to better match changing network conditions.

Cell shaping and traffic steering are particularly important in densely populated areas where neighboring cells may experience significantly different traffic loads. Intelligent adjustment of antenna coverage and traffic distribution can help balance demand, reduce congestion and improve utilization of available radio resources.

These reported throughput improvements demonstrate the potential to improve the utilization of existing radio infrastructure through software-based optimization. However, the companies have not disclosed the geographic scope, test duration, baseline conditions, radio configurations or measurement methodology underlying the maximum reported gains. Consequently, the results should not be interpreted as nationwide average improvements or as evidence that equivalent gains will be achieved throughout Verizon’s network.

Automating Fixed Wireless Access Mobility:

Another application demonstrated by Verizon addresses mobility problems affecting Fixed Wireless Access (FWA) devices transitioning between LTE and 5G Standalone (SA) networks.

Verizon implemented a Connected Mode Inter-Radio Access Technology (Inter-RAT) handover methodology to address failed transitions from LTE to 5G SA for FWA devices. According to the companies, the automated methodology eliminates the need for manual intervention when these handover failures occur.

In this scenario, Inter-RAT mobility involves transitioning a device between different radio access technologies: 4G LTE and 5G NR.

Conventional 5G Non-Standalone (NSA) deployments use an LTE anchor, with 5G NR added as a secondary radio access technology (5G NSA is essentially 4G LTE with a 5GNR RAN interface). It relies on the  LTE/Evolved Packet Core (EPC) architecture. By contrast, 5G SA uses a 5G NR radio access network connected directly to a 5G Core (5GC), without requiring an LTE anchor.  Only 5G SA can realize the 5G features and functions defined by 3GPP’s 5G Architecture specs.

–>A successful 5G NSA to 5G SA transition therefore requires appropriate mobility procedures, network configuration and coordination between the LTE and 5G SA environments.

This application illustrates how programmable RAN automation can address specific operational problems through targeted software applications rather than relying exclusively on conventional network-management procedures and manual troubleshooting.  It also demonstrates that RAN automation can address mobility management, network reliability and operational efficiency beyond radio-throughput optimization

Expanding the rApp Ecosystem and Network Autonomy:

Verizon plans to deploy EIC nationally during 2026 and introduce additional rApps developed both internally and by Ericsson. Planned applications include:

  • Cell Anomaly Detector: Identifies abnormal cell behavior that may affect network performance.
  • Root Cause Explainer: Helps identify potential causes of network problems, supporting more efficient troubleshooting and service restoration.
  • Anomaly General Optimizer: Supports network optimization in response to detected anomalies.

The names of these applications suggest different stages of the network assurance and optimization lifecycle. Anomaly detection identifies departures from expected network behavior; root-cause analysis helps determine potential underlying problems; and optimization applications can initiate corrective actions.

When integrated into a coordinated automation framework, these functions can form a closed-loop operational process:

  1. Monitor: Collect network performance measurements, alarms, configuration data and other operational information.
  2. Detect: Identify abnormal conditions or performance degradation using analytics and AI/ML models.
  3. Diagnose: Analyze network data to identify potential root causes.
  4. Optimize: Determine appropriate corrective actions based on network policies and operating constraints.
  5. Actuate: Apply approved changes to network configuration or operating parameters.
  6. Verify: Monitor subsequent network performance to determine whether the corrective action achieved the intended result.

This feedback process is fundamental to autonomous network operations. It allows network optimization to continue without requiring human intervention for every individual adjustment, while retaining appropriate operator oversight and control.

The companies have not provided detailed technical specifications for the three planned applications, so their precise functions, interfaces and degrees of automation remain to be established.

Ericsson Intelligent Automation Platform and Open Interfaces:

The Ericsson Intelligent Controller is a component of the Ericsson Intelligent Automation Platform (EIAP), Ericsson’s open network management and automation environment designed for multivendor, multi-technology networks.

Key platform capabilities include:

  • Support for 4G and 5G Radio Access Networks and Ericsson core networks.
  • A Non-RT RIC execution environment for rApps.
  • Standardized interfaces, including the O-RAN R1 interface.
  • Application lifecycle management and deployment.
  • Network data collection, analytics and AI/ML workflow capabilities.
  • Policy management and coordination of network optimization activities.
  • Controlled application execution and network actuation.

Ericsson reports that EIAP’s ecosystem includes more than 100 available rApps from over 100 ecosystem members, supporting different stages of the network automation lifecycle.

The ability for Verizon to develop its own rApps is particularly important. It gives the operator an opportunity to create network-specific optimization functions based on its traffic patterns, spectrum holdings, radio configurations and operational requirements, while also drawing on Ericsson’s broader rApp ecosystem.

An open application environment can reduce dependence on proprietary, vendor-specific optimization functions and encourage third-party application development.

However, the availability of open interfaces and a broad rApp ecosystem should not be confused with a fully interoperable multivendor RAN deployment. Ericsson describes EIAP as supporting multivendor networks, including purpose-built radio access networks. This means that deploying EIC and using O-RAN-aligned interfaces does not, by itself, establish that Verizon is replacing its existing RAN equipment with Open RAN equipment.

The distinction is important: Open RAN automation interfaces and an open application ecosystem can be deployed alongside conventional, purpose-built radio infrastructure.

The Road to Level 4 Network Autonomy:

The nationwide EIC deployment represents another step toward more autonomous RAN operations. Verizon’s longer-term objective is Level 4 (L4) network autonomy, in which network functions can perform increasingly complex operational and optimization tasks with minimal human intervention.

The TM Forum’s Autonomous Networks framework defines L4 as a highly autonomous network capability involving predictive decision-making and active closed-loop management of service-driven and customer-experience-driven networks, supported by AI modeling and continuous learning.

Achieving L4 autonomy requires considerably more than deploying AI applications. It requires:

  • Reliable, continuous monitoring of network conditions and performance.
  • Accurate AI/ML models that can operate under changing radio and traffic conditions.
  • Closed-loop optimization with measurable performance verification.
  • Coordination among multiple applications and network domains.
  • Policy enforcement, security controls and safeguards against unintended network changes.
  • Appropriate human oversight and the ability to intervene when automated actions produce unexpected results.

The distinction between automating individual network functions and achieving end-to-end autonomous network operations is important. An operator may successfully deploy several autonomous rApps without having achieved L4 autonomy across its entire RAN.

Furthermore, the TM Forum autonomy levels are assessed against specific network scenarios and operational domains. Achieving L4 in one function or network domain does not necessarily mean that the entire nationwide network has reached that level.

Verizon’s collaboration with Ericsson provides a programmable software platform and an expanding application ecosystem to support that transition. Whether it ultimately delivers L4 autonomy will depend on the scope of automation, the reliability of closed-loop operations and the integration of these capabilities across Verizon’s network.

Conclusions:

The Verizon-Ericsson initiative illustrates how AI-based software applications, open interfaces and closed-loop control can improve the operation of existing 4G and 5G radio access networks.  The reported Massive MIMO throughput improvements and automated LTE-to-5G SA FWA handover demonstrate practical applications beyond AI experimentation. The EIC deployment also provides Verizon with a platform for developing proprietary rApps while leveraging Ericsson’s broader application ecosystem.

The larger challenge will be scaling these capabilities across a nationwide network while maintaining interoperability, reliability, security and measurable improvements in customer experience and network economics.  Ultimately, the success of AI-native RAN automation should be judged by sustained improvements in network performance, spectral efficiency, energy consumption and operating costs—not simply by the number of AI applications deployed.

……………………………………………………………………………………………

References:

Verizon and Ericsson Partner to Transform AI-Native Autonomous Networks — November 3, 2026

https://www.ericsson.com/en/portfolio/cross-portfolio/rapp-directory

Autonomous Networks Mission and Levels Taxonomy — TM Forum

About Ericsson Intelligent Controller

About Ericsson Intelligent Automation Platform 

AI-RAN and Agentic AI get real: Ericsson, Nokia, Verizon & other operators enter into a new network automation era

Private 5G’s Business Model Shifts: Ericsson’s Move Signals a Market Reset

Ericsson signs MoU’s with Murata and Sumitomo to progress next-gen mobile network infrastructure

4.8 GHz to 4.9 GHz frequency band uses & Verizon Wireless experimental license for testing ISAC with Ericsson

Ericsson leads SK Telecom AI RAN vs. NVIDIA’s GPU centric AI RAN Alliance

Dell’Oro: Private Wireless Revenue Growth Prospects and Challenges

 

 

Infineon PSoC MCUs Expand TinyML into a Broader Range of Edge AI Applications

Posted on by Alan Weissberger

Disclaimer: ChatGPT helped research and write this article.

…………………………………………………………………………………………………………………….

Introduction:

Artificial intelligence (AI) running on embedded devices is evolving beyond the relatively narrow range of machine-learning applications traditionally associated with TinyML [1.], rather than Large Language Models (LLMs).  Newer microcontrollers (MCUs) are beginning to support a broader class of Edge AI workloads, including speech recognition, natural-language processing, computer vision, optical character recognition (OCR), selected translation applications and increasingly sophisticated voice interfaces. Many selected AI workloads can now be executed locally on embedded MCUs, reducing or eliminating the need for cloud processing for those particular functions.

Note 1. Traditional TinyML (Tiny Machine Learning) focuses on relatively small, task-specific machine-learning models running on resource-constrained embedded devices. The newer development is the ability to supplement those workloads with compact language models and more sophisticated speech and vision inference.

…………………………………………………………………………………………………………………………………………………………………………………

Infineon’s PSoC™ Edge family of microcontrollers illustrates this transition. By combining heterogeneous Arm processors, dedicated neural-network accelerators, substantial memory resources, graphics capabilities and embedded AI software, these MCUs provide a platform for intelligent devices capable of performing more complex AI processing locally.

Infineon states: “The relevant question is not whether an MCU can replicate an 8B LLM, but whether it can deliver useful AI experiences within the power, cost, and latency constraints of edge products. For many real-world edge applications, PSOC™ Edge is not merely capable, it is often a more appropriate solution than a higher-(processing) power MPU-based architecture.”

Let’s examine that assertion and consider Edge AI applications that can be supported by the Infineon PSoC MCUs.

Edge AI Applications supported by Infineon PSoC MCUs:

Infineon’s PSoC Edge MCUs support TinyML, which consists of low-latency, deterministic AI models that process sensor, audio, and vision data locally. The hardware is designed for applications requiring battery-powered execution, sub-1 milliwatt wake states, and zero cloud dependency.

From Infineon: PSoC Edge AI Applications include:

 Audio & Voice Processing:

  • Speech-to-Intent: Converting spoken commands directly into system actions (e.g., “Turn on the fan”) without needing to transcribe speech into text first.
  • Acoustic Event Detection: Identifying specific environmental sounds such as glass breaking, baby crying, gunshots, or alarms.
  • Beamforming & Noise Reduction: Using neural networks to filter out background noise in communication devices and smart wearables.

Low-Power Vision (TinyVision):

  • Presence & Gesture Detection: Recognizing if a human is in front of a device or interpreting basic hand movements for touchless control.
  • Object & Face Counting: Tracking things like people entering a room or basic inventory sorting on a conveyor belt.
  • Defect Classification: Performing basic visual quality inspections in automated manufacturing lines.

Smart Sensing & Predictive Maintenance:

  • Vibration and Anomaly Detection: Monitoring industrial motors, pumps, or consumer appliances to predict mechanical failure before it happens.
  • Sensor Fusion: Combining data from multiple sensors (e.g., temperature, pressure, humidity, and gas sensors) to evaluate environmental conditions or air quality index (AQI) changes in real time.
  • Biometric Tracking: Processing health metrics locally in fitness trackers, such as advanced heart-rate variability and sleep stage classification.

The table below illustrates how the PSoC Edge E84 matches these workloads by splitting tasks between its distinct processing domains:

Workload Type Active Hardware Component Typical Power/Performance State
Always-On Passive Listening Cortex-M33 + NNLite microNPU Ultra-low power, sub-1 mW state waiting for wake words or threshold anomalies.
Active Machine Learning Inference Cortex-M55 + Arm Ethos-U55 NPU High performance, using 128 MACs/cycle execution to compute vision or audio matrices instantly.

Source (for the information in this section and the table): Infineon

………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………

Infineon PSOC Edge MCUs bring more capable AI processing to embedded devices, supporting applications ranging from speech recognition and compact language models to computer vision and intelligent human-machine interfaces.  Image generated by ChatGPT.

……………………………………………………………………………………………………………………………………………………………………………….

PSoC Edge: A Heterogeneous Architecture for Embedded AI:

The PSoC Edge family is designed to combine high-performance AI inference with low-power, always-on processing. Its multi-domain architecture enables different processors and neural accelerators to handle workloads according to their performance and power requirements.

The PSOC Edge E84, used in the EWNA demonstration, incorporates:

  • Arm Cortex-M55: A 400 MHz processor with Helium DSP extensions for signal processing and machine-learning workloads.
  • Arm Ethos-U55-128 NPU: A dedicated neural processing unit for accelerating neural-network inference.
  • Arm Cortex-M33: A separate, lower-power processor operating at up to 200 MHz.
  • Infineon NNLite: A dedicated neural-network accelerator associated with the Cortex-M33 domain for low-power, always-on AI applications.

The E84 also provides up to 6 MB of system SRAM, 512 KB of resistive RAM (RRAM), external-memory interfaces and a low-power 2.5D graphics processor.  Its display controller supports MIPI-DSI/DBI interfaces and resolutions up to 1024 × 768. Multiple digital audio interfaces support multi-microphone applications, including far-field voice interaction.  Connectivity includes USB, 10/100 Ethernet, CAN-FD, SPI, UART, I2C, I3C and I2S.

This heterogeneous architecture provides two important processing domains:

  • High-performance domain: The Cortex-M55 and Ethos-U55 combination handles more computationally demanding AI workloads, including full voice inference and more complex neural-network applications.
  • Low-power domain: The Cortex-M33 and NNLite combination supports continuous, energy-efficient processing, including wake-word detection and other always-on machine-learning functions.

This arrangement is particularly useful for intelligent voice interfaces. The lower-power domain can monitor for an activation phrase without requiring the higher-performance processing resources to remain continuously active. Once activated, more demanding speech-recognition and language-model workloads can use the higher-performance domain.

The result is a flexible MCU architecture that combines conventional embedded processing with multiple levels of AI acceleration.

PSOC Edge family: Different configurations for different AI workloads:

Not all PSOC Edge devices offer identical AI and HMI capabilities. Infineon’s product comparison identifies the following differences:

MCU AI acceleration Voice AI Vision AI Graphics
E84 NNLite + Ethos-U55-128 Full voice inference Presence, gesture, motion Yes
E83 NNLite + Ethos-U55-128 Full voice inference Presence, gesture, motion No
E82 NNLite Wake-word detection Not listed Yes
E81 NNLite Wake-word detection Not listed No

Source: Infineon data sheets

The E84 offers the broadest combination of the capabilities listed above, including full voice inference, vision functions and graphics.  This differentiation is important when evaluating the family for embedded AI designs. The appropriate MCU depends on the workload, processing requirements, Human Machine Interface (HMI) complexity, memory resources and power budget.

………………………………………………………………………………………………………………………………………………………………

Infineon Demo At Embedded World North America (EWNA) 2026:

At EWNA 2026 [September 22–24, 2026 in Anaheim CA], Infineon demonstrated an on-device Edge AI application using the PSOC™ Edge E84. The demonstration integrated a 6.5-million-parameter Nexus Ember task-oriented language model, with a 512-token context window, and a 16-million-parameter FastConformer streaming speech-to-text engine, together with graphics support. The application demonstrated local speech recognition and language-model inference without requiring cloud processing for those functions.

Of particular interest, Infineon supplied detailed measurements of the implementation, including model size, memory utilization, language-model generation speed, speech-recognition latency and power consumption. These provide useful engineering data for evaluating the feasibility of executing compact language models and multiple AI workloads on MCU-class hardware.

Its reported measurements—18 generated tokens per second, a 0.36-second language-model time to first token, approximately 17 MB peak memory usage for the 16M voice-control pipeline excluding graphics buffers, and 92.6 mW active SoC power—provide useful evidence of what is achievable with a compact MCU-based AI implementation.

The broader application opportunities include advanced speech recognition, interactive voice interfaces, task-oriented language models, selected translation workloads, computer vision, gesture recognition, OCR and industrial monitoring.

These MCUs will not replace application processors (see Addendum below), embedded Linux systems or cloud-based AI infrastructure for large language models and computationally intensive multimodal applications. Their significance is different: an expanding range of useful AI functions can now be performed locally on increasingly capable microcontroller platforms.  For embedded-system designers, that can mean reduced dependence on cloud services, improved data privacy, lower communication requirements, faster interaction and potentially lower system cost and power consumption.

Paradigm Shift: Why Localized AI Execution Matters (Source: Google Gemini):

Moving complex speech and language models down to the MCU layer—where appropriate—solves three critical engineering dilemmas inherent in modern IoT deployments:
  1. Deterministic Latency & Reliability: By removing reliance on cellular, Wi-Fi, or cloud backends, critical applications suffer zero network dropouts (1:14). Latency drops to a deterministic sub-second window, which is vital for industrial machinery interfaces or responsive consumer electronics (0:11).
  2. Ironclad Data Privacy: Because processing happens entirely on the local silicon chip, sensitive voice data, visual captures, and text streams never leave the device boundary (0:25). This inherently complies with privacy frameworks and eliminates consumer data exposure risks.
  3. Drastic Power Savings: Transitioning an RF radio into high-throughput transmit mode to stream audio or video payload data to the cloud consumes significant energy. Offloading neural processing locally to a highly optimized NPU cuts total system energy consumption drastically, extending battery life and making “always-on” edge devices sustainable

Conclusions:

The important development is that Edge AI is no longer limited to narrowly defined TinyML inference. Infineon’s PSOC Edge family demonstrates how microcontrollers can increasingly support intelligent, voice-enabled, vision-aware and language-model-driven embedded applications.

The combination of heterogeneous processing, neural-network acceleration, graphics and memory resources enables a broader range of applications, including conversational interfaces, speech recognition, computer vision, OCR, selected translation workloads and industrial monitoring. The E84 demonstration provides concrete evidence that compact language models and streaming speech recognition can operate together locally, with relatively low power consumption.

Although MCU-based Edge AI will not replace more powerful application processors or cloud-based AI for demanding workloads, it can reduce cloud dependence, improve privacy and responsiveness, and lower system costs for many embedded applications. The opportunity is to bring useful AI capabilities directly into devices that traditionally relied on much simpler embedded processing.

……………………………………………………………………………………………………………………………………………………………………..

Addendum:

Google Gemini suggests using dedicated Application Processors (MPUs) or Edge Accelerators for open ended AI at the edge.

To support more demanding, general-purpose generative AI at the edge applications than the Infineon PSoC Edge MCUs, designers will typically need to move beyond MCU architectures to application processors (MPUs) or dedicated AI accelerators with substantially greater memory capacity and bandwidth. Microcontrollers lack the extreme memory bandwidth and capacity (such as High Bandwidth Memory or large external LPDDR buses) required to execute unconstrained, open-ended generative text and vision models.

Overview of Edge AI Hardware Tiers:

Processor Category Example Hardware Generative AI Capabilities
High-End Edge NPU Modules NVIDIA Jetson Orin Nano / NX Runs 1B to 3B parameter models locally with fast token generation. Uses shared unified LPDDR5 VRAM to fit small LLMs (e.g., Llama 3.2 3B or Qwen 2.5 3B) entirely on the device.
Application Processors NXP i.MX 95 / STMicroelectronics STM32MP2 Can handle highly optimized, tiny text models or text-to-speech tasks, but constrained by memory bandwidth. Ideal for structured, deterministic multi-modal smart interfaces rather than open-ended text gen.
Dedicated AI Accelerators Hailo-8 / Raspberry Pi AI Kit Excellent for offloading complex vision and neural network computations if paired with a capable host CPU. Designed for highly power-efficient mathematical processing (up to 26 TOPS) but lacks on-chip external DRAM for massive LLM weight pools.
Product Official vendor hyperlink
NVIDIA Jetson Orin Nano / NX NVIDIA Jetson Modules
NXP i.MX 95 NXP i.MX 95 Applications Processor
STMicroelectronics STM32MP2 STMicroelectronics STM32MP2 Series
Hailo-8 Hailo AI Products
Raspberry Pi AI Kit Raspberry Pi AI Kit

Note: The original Raspberry Pi AI Kit uses the Hailo-8L (13 TOPS), not the Hailo-8 (26 TOPS) in the first “Overview of Edge AI Hardware Tiers.. Raspberry Pi’s official documentation confirms this distinction.

……………………………………………………………………………………………………………………………………………………………………………………………………………………………………………..

Architectural Deep Dive: Why These Distinctions Matter:

The Memory Wall Constraint: Single-stream token generation for language models is heavily throttled by memory bandwidth rather than raw TOPS compute. This is why High-End NPU Modules with high-throughput unified memory architectures excel at localized LLM tasks over general-purpose application processors.

Coprocessor Offloading: Dedicated accelerators like the Hailo-8 require a powerful host CPU (like an ARM Cortex-A) because the accelerator manages the parallelized weight matrices for neural layers while the host handles system OS routines, memory routing, and basic context token tracking.

……………………………………………………………………………………………………………………………………………………………………………………………………………………..

References:

https://www.infineon.com/products/microcontroller/32-bit-psoc-arm-cortex/32-bit-psoc-edge-arm

https://www.infineon.com/products/microcontroller/32-bit-psoc-arm-cortex/32-bit-psoc-edge-arm/psoc-edge-e84

Blaize and Winmate Forge Strategic Partnership to Accelerate Edge AI Integration in Ruggedized Systems

Analysis: Edge AI and Qualcomm’s AI Program for Innovators 2026 – APAC for startups to lead in AI innovation

Is the “far edge” a bridge to far to cross for AI inferencing? What about “Distributed AI Grids”?

T-Mobile’s Strategy of Deploying Multi-Vendor AI-RAN on “Kinetic Edge” using AWS Pillars

Will “AI at the Edge” transform telecom or be yet another telco monetization failure?

Custom AI Silicon: OpenAI, Anthropic, Hyperscalers, Huawei and the Battle with Nvidia, AMD and Broadcom

Google’s TPU Business Outpaces Rivals as Hyperscalers Accelerate Custom AI Silicon Strategies

AWS to deploy AI inference chips from Cerebras in its data centers; Anapurna Labs/Amazon in-house AI silicon products

ATIS Proposes Wireless Physical-Layer Foundation Models for AI-Native 6G Networks

Posted on by Alan Weissberger

Disclaimer: ChatGPT was used extensively to read and summarize the 25 page ATIS Whitepaper described in this article.

………………………………………………………………………………………………………………………………………………………………………….

Introduction:

The Alliance for Telecommunications Industry Solutions (ATIS) has introduced Wireless Physical Layer Foundation Models (WPFMs), a proposed approach to applying artificial intelligence directly to radio-frequency (RF) signals and wireless channel measurements. The objective is to develop reusable AI models that learn characteristics of the radio environment and apply that knowledge to multiple physical-layer communications, sensing, and signal-processing functions.

The ATIS white paper, Wireless Physical Layer Foundation Models for Advanced Communication, Sensing, and Neuro-Symbolic Decision-Making, describes how these models could improve wireless signal interpretation, channel estimation, interference detection, beam management, localization, and joint communications and sensing (JCAS).  It is a promising  direction for AI in telecommunications because it moves beyond network management, orchestration, and application-layer intelligence. Instead, WPFMs seek to apply foundation-model techniques to the underlying radio signals and propagation characteristics that determine wireless network performance.

“Future wireless networks will need to do more than respond to changing conditions. They will need to understand their environments and make increasingly complex decisions,” said ATIS President and CEO Susan Miller. “This research explores how a shared foundation for AI could help wireless networks operate more efficiently, adapt more effectively, and make decisions that can be better understood and trusted.” 

“Learning from large amounts of unlabeled wireless signal data means Wireless Physical Layer Foundation Models need less labelled data for each new task,” said Lead Researcher for the WPFM Jaron Fontaine, imec – IDLab Ghent – Ghent University. “Combined with neuro-symbolic reasoning, they can help networks reason about changing conditions and make decisions that are easier to explain.” 

Source: ATIS Whitepaper

The white paper highlights the potential for WPFMs to support more integrated and intelligent wireless systems that can move beyond individual AI capabilities as the industry advances toward 6G. It also identifies opportunities for further research and industry collaboration.  This ATIS proposal is particularly relevant to the evolution of 5G-Advanced (3GPP Releases 18-20) [1.] and future 6G wireless networks under the ITU-R IMT-2030 framework.  That’s where increasingly complex spectrum environments, higher frequencies, massive antenna arrays, mobility, and integrated sensing will create additional demands on physical-layer signal processing.

Note 1. The latest ITU-R M.2150-3 recommendation (IMT 2020 RIT/SRITs), approved in February 2026, officially incorporates the initial 5G-Advanced baseline from 3GPP Release 18.

…………………………………………………………………………………………………………………………………………………………………………………………………………………………

From Task-Specific Wireless AI to Foundation Models:

Today’s wireless AI implementations typically employ specialized machine-learning models for individual tasks. Examples include modulation recognition, spectrum sensing, interference detection, channel-state information (CSI) compression, beam management, localization, and RF sensing.

Although deep learning has improved the ability to recognize complex signal patterns, these models often remain dependent on particular datasets, hardware platforms, operating conditions, and wireless environments. A model trained for one application may not generalize effectively to another.

ATIS proposes using a foundation-model architecture to address this fragmentation. WPFMs would be pre-trained on large, diverse collections of wireless RF data, primarily using self-supervised learning. This allows models to learn underlying signal and channel characteristics from unlabeled data, reducing the need for extensive manually labeled datasets for each downstream application.  It’s important to note that  some of the functions depicted are proposed future capabilities, not demonstrated implementations. Figure 1. below provides a useful overview of the WPFM concept and its relationship to wireless physical-layer intelligence.

The resulting representations could subsequently be fine-tuned or adapted for different wireless tasks.  The RF data used for training and inference can include:

  • Raw in-phase and quadrature (I/Q) samples representing complex wireless signals.
  • Channel-state information (CSI) and channel impulse response (CIR) measurements.
  • Spectrograms and other time-frequency representations.
  • Doppler signatures and Fast Fourier Transform (FFT)-based features.
  • Contextual metadata, including sampling rate, center frequency, antenna identity, timing, and hardware platform.

 

 

 

 

Source: ATIS Whitepaper

…………………………………………………………………………………………………………………………………………………………………………………………………………………….

The white paper describes a multimodal architecture that maps different RF data types into a common latent representation. That’s shown in the illustration below.

Transformer-based models are one candidate architecture because of their ability to capture relationships across time, frequency, and spatial dimensions.  An important design consideration is preserving the physical meaning of RF observations. Timing, frequency shifts, antenna configurations, and propagation characteristics cannot simply be treated as interchangeable numerical inputs. The model must retain sufficient information about the underlying radio conditions to support accurate interpretation.

Source:  ATIS Whitepaper

…………………………………………………………………………………………………………………………………………………………………………………………………………………………

Potential Applications in Wireless Networks:

ATIS identifies several areas where a shared RF representation could support multiple wireless applications.

Radio resource management: WPFMs could provide reusable information about channel quality, mobility, interference, and network loading. This information could help optimize physical resource block (PRB) allocation, scheduling, and handover decisions. 

Spectrum management and interference detection: By learning RF characteristics across different bands, waveforms, and wireless technologies, WPFMs could help identify unfamiliar signals, detect interference, and improve spectrum awareness.

Channel prediction and localization: Shared representations of propagation conditions and channel behavior could support channel prediction, positioning, and localization across changing environments.

Joint communication and sensing (JCAS): WPFMs could help extract information about motion, occupancy, object presence, and environmental conditions while simultaneously supporting wireless communications. The paper also discusses applications involving passive RF sensing.

Industrial IoT and connected vehicles: Potential applications include detecting anomalies in industrial wireless environments, supporting cooperative vehicle perception, and improving wireless performance under changing propagation conditions.

These are proposed applications, not demonstrated commercial capabilities. ATIS explicitly describes the performance improvements in its use-case tables as illustrative evaluation targets rather than experimentally validated results.

…………………………………………………………………………………………………………………………………………………………………………………………………………………………

Massive MIMO, Phased-Array Beamforming and Beam Management:

Massive multiple-input multiple-output (MIMO) systems use large antenna arrays to serve multiple users and improve spectral efficiency through spatial multiplexing and directional transmission. Massive MIMO is already deployed in sub-6 GHz cellular networks, while at millimeter-wave and higher frequencies, electronically steered phased-array antennas are particularly important because they concentrate RF energy into narrow beams to compensate for higher propagation losses.

Managing these beams requires continual estimation of channel conditions, selection of appropriate beam directions, and adaptation to user mobility, blockage, and changing multipath propagation. Conventional beam-training and beam-search procedures can introduce signaling overhead and latency, particularly in highly dynamic environments.

A Wireless Physical Layer Foundation Model could learn relationships among channel-state information (CSI), channel impulse responses (CIR), Doppler measurements, antenna configurations, and observed RF signal characteristics. These learned representations could help predict beam quality, identify alternative propagation paths, and recommend beam adjustments before an existing connection deteriorates.

For massive MIMO, this intelligence could also assist with antenna-array configuration, spatial channel characterization, and user-specific beam selection. Potential benefits include reduced beam-training overhead, more efficient spatial resource utilization, improved link reliability, and faster adaptation to changing radio conditions.

These applications are especially relevant to 5G-Advanced and future 6G systems employing large antenna arrays, higher-frequency spectrum, and increasingly sophisticated beamforming techniques. However, the practical value of WPFMs will depend on demonstrating improvements in beam-selection accuracy, processing latency, and overall radio performance compared with conventional channel estimation and beam-management algorithms.

…………………………………………………………………………………………………………………………………………………………………………………………………………………………

O-RAN: A Potential Deployment Architecture:

Although O-RAN [2.] has not delivered on its original promise of broad, plug-and-play multi-vendor RAN interoperability, its disaggregated architecture remains a potentially relevant environment for WPFMs.   That’s because its disaggregated architecture distributes RAN intelligence across functions operating at different timescales.

ATIS proposes that shared RF representations could be made available to O-RAN’s RAN Intelligent Controller (RIC) applications, including:

  • Non-real-time RIC rApps: Applications operating on longer timescales could use WPFM-derived information for network optimization, planning, and policy-related functions.
  • Near-real-time RIC xApps: Applications could use shared RF information to support more adaptive RAN optimization and resource-management decisions.

Note 2. O-RAN refers to the O-RAN ALLIANCE’s architecture and specifications for OpenRAN.

The potential benefit is that different RAN applications, including those developed by different vendors, could work from a common representation of radio conditions rather than each independently collecting, processing, and interpreting similar RF data.  However, the white paper does not define a standardized WPFM interface or establish how these representations would be exchanged among multi-vendor RAN components. Such interoperability remains an important research and development requirement.

…………………………………………………………………………………………………………………………………………………………………………………………………………………………

Neuro-Symbolic AI: Adding Reasoning and Operational Constraints:

One of the more interesting aspects of the ATIS proposal is its integration of WPFMs with neuro-symbolic AI.

A neural foundation model can recognize RF patterns, estimate network conditions, identify anomalies, and predict changes in the radio environment. However, pattern recognition alone does not establish whether a proposed network action is appropriate under operator policies, spectrum regulations, protocol requirements, or physical constraints.

Neuro-symbolic AI combines neural-network learning with explicit rules, knowledge representations, and logical reasoning.

In the proposed architecture, a WPFM provides the perceptual foundation. A separate reasoning layer evaluates its outputs against operational policies and constraints before a network-control function executes an action.

For example, a WPFM might identify deteriorating channel conditions and propose a beam change or resource-block reallocation. The neuro-symbolic reasoning layer could then evaluate the proposal against interference conditions, quality-of-service requirements, spectrum constraints, and network-slicing priorities.

This separation is important: the WPFM is not intended to independently control the network. Instead, its learned representations and predictions become inputs to a broader perception-reasoning-action process.

ATIS also discusses a future agentic architecture in which multiple AI agents and tools coordinate network decisions. However, the development of such neuro-symbolic and agentic architectures is explicitly identified as future work beyond the scope of the current white paper.

…………………………………………………………………………………………………………………………………………………………………………………………………………………………

Major Technical Challenges Remain:

Despite the potential advantages, substantial engineering work is necessary before WPFMs can become practical components of operational wireless networks.

Real-time performance and computational requirements: Wireless physical-layer functions often operate under stringent latency constraints. ATIS illustrates a possible model hierarchy ranging from large models with tens of millions of parameters for offline pre-training to lightweight models intended for edge and device-level operation. Its illustrative latency targets range from approximately 100 milliseconds for large models to less than one millisecond for ultra-lightweight models. These are proposed deployment trade-offs, not measured performance results.

Generalization across wireless environments: A model must operate reliably across different frequency bands, propagation conditions, antenna configurations, hardware platforms, mobility conditions, and terrestrial and non-terrestrial networks. Performance in one environment does not guarantee comparable results elsewhere.

RF datasets and training: Developing sufficiently large and diverse datasets is difficult. Data quality, hardware diversity, synchronization, metadata consistency, and representation of uncommon but operationally important RF conditions all affect model performance. Synthetic data, ray tracing, and digital twins may supplement real-world RF measurements, but simulation-to-reality transfer must be demonstrated.

Neuro-symbolic integration: Translating high-dimensional RF representations into explicit knowledge and rules without losing important signal information is a significant challenge. The combined neural and symbolic processing must also satisfy wireless latency and throughput requirements.

Multi-vendor interoperability: Independently developed WPFMs could generate incompatible internal representations or embeddings. Standardized interfaces, semantic alignment, model provenance, data governance, and interoperability testing will be necessary if multiple vendors are to share and use these models.

Validation and benchmarking: The industry currently lacks standardized multi-task benchmarks for WPFMs. Demonstrating improvements against conventional task-specific models under realistic RF conditions will be essential to establishing their practical value.

…………………………………………………………………………………………………………………………………………………………………………………………………………………………

Conclusions:

ATIS’s Wireless Physical Layer Foundation Model proposal represents a promising direction for AI in telecommunications. Rather than adding another collection of task-specific AI applications to wireless networks, it seeks to establish a reusable understanding of the physical radio environment that can support communications, sensing, localization, and RAN optimization.

The proposed combination of shared RF representations and neuro-symbolic reasoning could eventually provide a foundation for more adaptive and explainable wireless-network operations.

However, the distinction between a proposed architecture and a deployable technology is important. ATIS has outlined a research framework and potential applications; it has not announced a commercially available WPFM, demonstrated operational network deployment, or established a new 3GPP wireless standard.

The key questions are whether these models can generalize across real-world RF environments, meet stringent physical-layer latency and computational requirements, interoperate across vendors, and demonstrate measurable improvements over existing specialized AI models.

For the telecommunications industry, the significance of WPFMs will ultimately depend not on the foundation-model concept itself, but on whether a shared AI understanding of the radio environment can deliver measurable improvements in wireless network performance, efficiency, and operational reliability.

……………………………………………………………………………………………………………………………………………………………………………………………………………………..

References:

ATIS Introduces Wireless Physical Layer Foundation Models for Next-Generation Wireless Intelligence

 Wireless Physical Layer Foundation Models for Advanced Communication, Sensing, and Neuro-Symbolic Decision-Making

ITU-R M.2150-3: https://www.itu.int/rec/R-REC-M.2150-3-202602-I/en

AI-Native 6G RAN in Practice: Research and Validation Insights from 6G-MIRAI-HARMONY

T-Mobile’s Strategy of Deploying Multi-Vendor AI-RAN on “Kinetic Edge” using AWS Pillars

AI-RAN Push: NVIDIA, Nokia, and Operators Drive Towards Distributed Compute Platforms

Deutsche Telekom: Device-to-Cloud AI Traffic Won’t Force RAN Upgrades or Increase CAPEX

Analysis: Nvidia–MediaTek Deepen AI Platform Collaboration -What it Means for AI RAN & DCI

Ericsson leads SK Telecom AI RAN vs. NVIDIA’s GPU centric AI RAN Alliance

Analysis: Nokia’s new AI-RAN platform and Standalone AI-RAN node with Nvidia GPUs

Dell’Oro: AI RAN revenue forecast: $35B from 2026-to-2030; 3 types of AI RAN explained

Dell’Oro: AI RAN to account for 1/3 of RAN market by 2029; AI RAN Alliance membership increases but few telcos have joined

AI-Era Cloud Network Transformation: A Reference Architecture and Implementation Roadmap

AI-RAN and Agentic AI get real: Ericsson, Nokia, Verizon & other operators enter into a new network automation era

Federated Wireless Spectrum AI: Advancing CBRS Efficiency Through AI-Driven RAN Optimization

Analysis: Nvidia’s rumored new 6G AI-RAN – likely features/functions and industry impact

AT&T and Ericsson boost Cloud RAN performance with AI-native software running on Intel Xeon 6 SoC

What Might Qualify as an Equivalent Replacement for a Copper Based Telephone Landline?

Posted on by Alan Weissberger

Introduction:

Retiring a copper telephone network is not simply a change of transmission medium. A migration can change where power is supplied, how calls are established, how connected equipment communicates, and which components must remain operational during an emergency.  The objective is equivalence of required service outcomes—not replication of every characteristic of the old network. This article proposes a framework for evaluating those outcomes; it does not describe an existing regulatory certification.

The issue is becoming concrete for California customers. In its May 20, 2026 announcement, AT&T committed $19 billion through 2030 to fiber and wireless investment and described plans to discontinue specified copper-based services in parts of California on or after June 1, 2027. The company also committed to continued access to voice and 911 service. These are announced plans and commitments, not evidence of completed migration outcomes (more below).

The engineering question is more demanding than whether a replacement produces dial tone. A meaningful assessment should examine the functions customers depend on: conversational performance, emergency calling, power resilience, connected-device compatibility, and recovery after failure.

Separate the Access Connection From the Telephone Service:

“Fiber,” “wireless,” and “digital voice” describe different aspects of a communications system. A migration assessment should identify both the connection serving the premises and the voice service operating over it.  Ofcom describes the UK’s digital-landline transition as a move to Voice over Internet Protocol, or VoIP. Its guidance also explains that customers do not necessarily need to purchase a high-speed broadband package to retain telephone service. The service requirement and the broadband product offered to a customer are therefore not interchangeable.

For evaluation purposes, three questions should remain separate:

  • Is the access connection operational?

  • Can the telephone service establish and receive calls?

  • Does the complete installation deliver the functions the customer requires?

An answer to the first question does not settle the other two. The proposed assessment should follow the entire service path, rather than treating the access technology as a proxy for telephone-service performance.

Power Resilience Requires a Complete Installation Test:

Traditional exchange-powered telephone service could keep a basic corded telephone working during a household power failure. Broadband-based voice introduces a different dependency: equipment at the customer’s premises needs electricity. Ofcom warns that digital landlines require a resilience solution if they are to remain usable during a power cut.

That distinction should not become a claim that copper service was immune to failure. It identifies a particular capability that a migration may need to preserve: telephone access when household mains power is unavailable.

The proposed acceptance procedure should identify every premises component required for calling and establish which backup arrangement powers each one. The test should then disconnect household mains power and verify the installed service—not merely inspect a battery indicator.

The resulting record should state:

  • Which equipment receives backup power.

  • Whether incoming and outgoing calls remain possible.

  • The tested standby duration and calling duty cycle.

  • The battery condition and replacement responsibility.

  • What happens when mains power returns.

A premises test also has a boundary: it cannot establish how long the provider’s external network will remain available during a regional outage. Installation-level results and provider-level resilience evidence should be documented separately.

Evaluate Conversation, Not Just Call Establishment:

The framework should distinguish successful call setup from an acceptable conversation.

A practical test program would evaluate incoming and outgoing calls under representative conditions, including simultaneous household broadband use where applicable. It should record call completion, audible interruptions, conversational delay, unexpected disconnections, and performance during relevant busy periods.

Rather than prescribing one universal numerical threshold, the migration plan should identify the applicable performance objectives and their basis. The acceptance record should make clear what was measured, under which conditions, and what constituted a passing result.

This approach avoids declaring one technology inherently superior. It asks whether the deployed service meets an explicit requirement.

Wireless Replacements Need Site-Specific Evidence:

AT&T’s California announcement identifies AT&T Phone – Advanced and AT&T Business Voice among its replacement offerings. The company states that these products provide battery backup and support existing telephones and various connected devices. Those statements should be presented as provider claims, not independent compatibility findings for every installation.about.att

The assessment should identify the product’s actual access arrangement rather than grouping every wireless-enabled voice product under one technical description.

For a wireless installation, the proposed checks include:

  • Reception at the equipment’s installed location.

  • Call completion and conversational performance at relevant busy periods.

  • Operation with household mains power disconnected.

  • Recovery after a loss of connectivity.

  • Successful operation of identified connected devices.

  • Any alternative communications path included in the service design.

A coverage statement and an installation acceptance result answer different questions. The proposed framework requires evidence about the installed service.

A Telephone Socket Does Not Establish Device Compatibility:

Some telephone connections support applications beyond conversation. Ofcom specifically advises customers to check equipment such as telecare devices and burglar alarms before switching to digital landlines.

The technical distinction is important: a signal that remains intelligible to a person may not be reproduced accurately enough for a machine. RFC 4733 explains that low-bit-rate speech codecs cannot be assumed to reproduce signaling tones reliably for automatic recognition. It defines mechanisms for transporting telephone events, including keypad digits, separately from ordinary encoded speech.

IETF RFC 4733 also discusses distinct approaches for fax, voice-band data, and modem communications, including T.38, V.152, and V.150.1. These mechanisms address different signal types; their existence is not a universal compatibility guarantee.  Acceptance should therefore test the application’s complete transaction. For an alarm, that would mean verifying that the event reaches the monitoring center, identifies the correct account or location, and receives the expected acknowledgment.  If migration requires replacing the connected device, the replacement should be evaluated as a new end-to-end system, including fault reporting and recovery—not simply described as preservation of the old telephone interface.

Emergency Access Needs Its Own Acceptance Plan:

The proposed framework should treat emergency access as a separate service requirement, with documented arrangements for:

  • Call origination and routing.

  • Service-location information.

  • Callback capability.

  • Operation during a premises power failure.

  • Recovery after equipment restart or access failure.

Verification should use provider-supported procedures and an agreed process with the relevant emergency organization. The framework does not call for unsolicited test calls to emergency numbers.

Any live emergency-calling validation must follow an agreed procedure with the provider and relevant emergency organization—not an unsolicited test call.

Network-side resilience also deserves attention. U.S. rules for covered 911 service providers address physical diversity, operational integrity, and monitoring. These obligations concern specified emergency-network facilities; they should not be confused with a blanket guarantee that every residential connection survives a prolonged outage.  Evidence should also be matched to its scope. A successful installation test does not demonstrate regional emergency-network availability, just as a provider’s network-resilience statement does not demonstrate that a particular household has functioning backup power.

Fiber-based voice: access capacity is not the same as voice assurance:

ITU-T Recommendation G.9807.1 specifies XGS-PON, a 10-Gigabit-capable symmetric passive optical network. It is an access-network specification—not a certification that a particular retail telephone service reproduces every function of an analog line.

The voice service has additional layers. IETF RFC 3261 defines the Session Initiation Protocol, or SIP, used to establish, modify, and terminate sessions. Voice quality also depends on the media path and signal processing, not merely successful session establishment.rfc-editor+1

This creates an important distinction between three tests:

  • The access connection is operational.

  • A telephone call can be established.

  • The resulting conversation meets an acceptable quality and reliability target.

ITU-T G.114 addresses the effect of one-way transmission delay. G.107’s E-model provides a planning framework for considering transmission impairments, including delay and codec-related effects. ITU technical guidance also explains the jitter-buffer trade-off: buffering can accommodate variations in packet arrival time, but adds delay.itu+1

A proposed migration test should therefore measure conversational performance under realistic simultaneous broadband use—not only with an otherwise idle connection. The relevant evidence would include delay, packet loss, conversational interruptions, call completion, and performance at busy periods.

The objective is not to declare IP voice inherently inferior or superior. It is to establish whether the deployed service meets its stated performance objectives.

Fixed wireless (FWA): validate the installation, not just the coverage map:

Fixed-wireless voice is another migration option. AT&T’s May 2026 California announcement describes a transition toward fiber and wireless services, including AT&T Phone – Advanced and AT&T Business Voice. The company states that these products include battery backup and support existing telephones and various connected devices. Those are provider claims, not independent compatibility results for every installation.

For a fixed-wireless replacement, a useful acceptance procedure would examine:

  • Reception at the actual installed equipment location.

  • Call completion and audio quality during busy periods.

  • Battery operation with household electricity disconnected.

  • Recovery after loss of radio connectivity.

  • The behavior of connected alarms and monitoring equipment.

  • Whether an alternative communication path is available when the primary service fails.

These are proposed engineering checks, not assertions that a particular wireless product fails them.

“Wireless coverage exists” and “this installation provides an adequate replacement” are different propositions. The second requires site-specific evidence.

Legacy devices: intelligible speech does not prove compatibility:

An analog telephone line may carry more than conversation. Ofcom specifically warns customers to check equipment such as telecare devices and burglar alarms before migration.

The technical difficulty is that a signal understandable to a person is not necessarily reproduced accurately enough for a machine.

RFC 4733 explains why separate transport of telephone events is useful: low-bit-rate speech codecs cannot be assumed to reproduce signaling tones accurately enough for automatic recognition. The RFC defines RTP payload formats for DTMF digits and other telephone events and tones.

Relevant standards address different aspects of the problem:

Standard Relevance to migration
IETF RFC 4733 Transport of DTMF digits and other telephone events and tones in RTP.
ITU-T T.38 Specialized transport for Group 3 fax over IP networks.
ITU-T V.152 Procedures for supporting voice-band data over IP networks.
ITU-T V.150.1 Procedures for modem relay across IP networks.

These mechanisms are distinct; they are not a universal compatibility guarantee. RFC 4733 identifies T.38, V.152, and V.150.1 as alternative approaches for particular signal types. ITU guidance further explains that voice-band data may require changes such as disabling voice-activity detection and comfort-noise generation during data transfer.itu+1

Consequently, an adapter with a familiar telephone socket should not be treated as proof that every attached application will work.

A stronger migration procedure would test the complete transaction: an alarm reaches its monitoring center, receives the expected acknowledgment, and reports the correct account or location. Where replacement equipment is needed, the installation should be evaluated as a new end-to-end system rather than as transparent preservation of the old interface.

Backup-power requirements: distinguish a minimum from a guarantee:

Ofcom’s guidance calls for a solution enabling access to emergency organizations for at least one hour during a premises power cut. Providers should offer the solution free to customers dependent on their landline for emergency calling. Ofcom’s consumer guidance recognizes that a mobile solution is appropriate only where there is usable mobile coverage.

The United States requires particular care when discussing historical rules. The retrieved text of 47 CFR §9.20 describes requirements to offer eight-hour and twenty-four-hour standby backup options for covered residential services that are not line-powered. However, that same text contains an explicit September 1, 2025 sunset provision. It should not be cited without qualification as an ongoing October 2026 requirement.  Even as an engineering benchmark, standby duration is not identical to continuous talk time, network availability, or guaranteed operation throughout a regional emergency.  A replacement-service specification should state the assumed duty cycle, battery condition, maintenance responsibility, and equipment included in the backup arrangement.

Image Generated by Perplexity.ai which was also used for research contained in this article.

……………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………

Industry activity makes this a high priority issue:

AT&T’s May 20, 2026 California announcement commits $19 billion through 2030 to fiber and wireless investment. It also describes plans to discontinue specified copper-based services in parts of California on or after June 1, 2027, while committing that customers will retain access to voice and 911 service.

“Californians depend on and expect fast, reliable connectivity every day and AT&T is committed to delivering it,” said Susan Santana, state president of AT&T California. “With recent, forward-looking federal network modernization policies, we’re able to make our largest-ever California investment commitment ($19 billion) as we transition away from copper networks. As the largest fiber builder in the state, this expands future-ready fiber and wireless services at affordable prices and modernizes copper networks. By investing in newer, more reliable technology, we can deliver better service, reach more customers, improve our communities, and support California’s economy for years to come.”

–>These are AT&T announced plans and commitments, not completed migration outcomes.

In the UK, Ofcom’s digital-landline guidance places particular emphasis on customers dependent on telephone service and on equipment connected to the line. The transition is therefore not solely an infrastructure replacement program; it also requires identifying customer needs and resolving application-level dependencies.ofcom+1

The common challenge is translating broad continuity commitments into installation-level evidence.

A proposed replacement-service acceptance framework:

No single standard cited above defines a universal substitute for every telephone landline. A practical assessment could instead use a documented service profile:

Dimension Proposed acceptance evidence
Basic calling Successful incoming and outgoing calls, ringing, required calling features, and number retention where applicable.
Conversational performance Results against stated objectives under representative operating conditions.
Premises power resilience Demonstrated operation of all necessary equipment for an agreed duration and calling duty cycle.
Emergency access Documented verification of routing, location handling, and callback arrangements through an approved procedure.
Connected applications Successful end-to-end transactions for identified alarms, monitoring systems, fax devices, or other required equipment.
Recovery and restoration Observed restart and reconnection behavior, plus documented fault-reporting and repair arrangements.
Costs and responsibilities Clear allocation of equipment, battery replacement, maintenance, compatibility remediation, and support obligations.

https://www.ofcom.org.uk/siteassets/resources/documents/phones-telecoms-and-internet/information-for-industry/future-of-fixed-telephone-services/future-fixed-telephone-services.pdf

https://www.ecfr.gov/current/title-47/chapter-I/subchapter-A/part-9/subpart-H/section-9.20

https://www.ecfr.gov/current/title-47/chapter-I/subchapter-A/part-9/subpart-H

https://www.itu.int/dms_pubrec/itu-t/rec/g/T-REC-G.9807.1-202010-S!Amd2!TOC-HTM-E.htm

https://www.rfc-editor.org/info/rfc4733/

https://about.att.com/story/2026/att-makes-ca-commitment.html

AT&T’s Fiber Expansion Strengthens Its Converged Wireless and Home Broadband Strategy

S&P Global Market Intelligence Surveys: Fiber Deployments in U.S. and Europe + AI Infrastructure Causes Market Shift

T-Mobile expands FTTH footprint via 50-50 JVs with Oak Hill Capital and Wren House

2026 Fiber Connect Keynote: “The Future of Fiber Optics: AI and the Quantum”

More KPN customers use fiber vs copper for broadband services in Nederlands

U.S. fiber rollouts now pass ~52% of homes and businesses but are still far behind HFC

 

 

Intel and Broadcom Demonstrate IEEE 802.11bn/Wi-Fi 8 Interoperability

Posted on by Alan Weissberger

Disclaimer:  ChatGPT was used for research in this article, but the author had to frequently checkpoint its AI work for omissions and accuracy.

…………………………………………………………………………………………………………………………………………………

Introduction:

Wi-Fi 8, based on the IEEE 802.11bn amendment, reached an important development milestone with a public cross-vendor interoperability demonstration involving prototype equipment from Intel and Broadcom. Conducted in London on October 27, the demonstration showed that a Broadcom-based Wi-Fi 8 access point could interoperate with an Intel Wi-Fi 8 client while demonstrating improvements in spectrum utilization, aggregate throughput and latency.

The demonstration is significant because it moves Wi-Fi 8 beyond individual vendor prototypes toward cross-vendor interoperability. It also illustrates the technology’s central objective: improving reliability, predictable performance and efficient spectrum utilization rather than simply increasing maximum theoretical data rates.  However, this was an engineering demonstration, not formal Wi-Fi Alliance certification in advance of the IEEE 802.11bn standard which has not been finalized.

IEEE 802.11bn: The Foundation for Wi-Fi 8:

Wi-Fi 8 is the industry designation for the next-generation wireless LAN technology being developed through the IEEE 802.11 Working Group’s P802.11bn amendment, formally titled Enhancements for Ultra High Reliability (UHR).   Unlike previous generations that emphasized increasing maximum theoretical data rates, IEEE 802.11bn focuses on improving actual network performance under challenging operating conditions. These include dense deployments, overlapping access points, interference, mobility, edge-of-coverage conditions and networks supporting large numbers of simultaneously connected devices.

The amendment covers enhancements to both the physical layer (PHY) and medium access control (MAC) layer. Its objectives include:

  • Higher reliability and more consistent throughput.
  • Reduced latency and latency variation.
  • Improved spectrum utilization and coexistence.
  • Better performance in dense wireless deployments.
  • Improved coordination among multiple access points.
  • Greater energy efficiency.

Wi-Fi 8 retains important capabilities introduced with Wi-Fi 7, including operation in the 2.4, 5 and 6 GHz unlicensed bands, channel bandwidths up to 320 MHz, 4096-QAM modulation and Multi-Link Operation (MLO).  The key distinction is that Wi-Fi 8 is intended to improve the consistency and reliability of wireless connectivity rather than simply raise maximum advertised speeds.

In other words, UHR is about delivering better real-world wireless performance, particularly when networks are congested or radio conditions are less than ideal.

IEEE 802.11bn Task Group and Standards Status:

The IEEE 802.11 Working Group has been responsible for developing wireless LAN standards, since it was created in September 1990.  The IEEE 802.11 Task Group bn (IEEE 802.11bn) is specifically responsible for the technical development of the Wi-FI 8 standard which will provide enhancements for an Ultra High Reliability capability in an IEEE 802.11 Wireless Local Area Network (WLAN).  IEEE 802.11bn (Wi-Fi 8) is still under development and is far from a completed, approved standard. The IEEE 802.11 Working Group continues to develop the specification through Task Group bn. Although the project has progressed to draft development and working-group balloting, it has not completed the IEEE standards approval process. Technical features and requirements may therefore change before final approval. The current target timeline calls for completion in 2028, but that schedule remains subject to change.

The IEEE 802.11bn task group includes participants from across the wireless industry, including chipset manufacturers, equipment vendors, technology companies and research organizations. These participants contribute technical proposals, evaluate alternative approaches and develop the amendment through the IEEE standards process.

IEEE’s published project information indicates that the P802.11bn Draft 2.0 Working Group letter ballot passed in September 2026. The ballot attracted 560 returned votes from a pool of 705 voters, with 417 approving and 128 disapproving. The approval rate was 76.5%, just above the required 75% threshold.  The task group subsequently began resolving comments and developing the next draft.

Final IEEE 802.11 Working Group approval is currently projected for March 2028, followed by additional IEEE approval processes. These dates remain subject to change.  Consequently, Wi-Fi 8 remains a work in progress. Prototype equipment can implement selected draft features, but those implementations should not be confused with compliance to a finalized standard.

Intel and Broadcom Conduct Cross-Vendor Interoperability Testing:

The October 27th demonstration was reported by Wi-Fi NOW and involved Intel and Broadcom engineers.  It was conducted at a Wi-Fi Alliance members meeting in London, collocated with the Dynamic Spectrum Alliance (DSA) Global Summit.  The demonstration used a Broadcom-based Wi-Fi 8 access point interoperating with an Intel Wi-Fi 8 client device. Existing Wi-Fi 7 equipment was also used for performance comparisons.

The CMX500 Signaling Tester, now has WiFi8 Test Capabilities   Photo Credit: Rohde & Schwarz

………………………………………………………………………………………………………………………………………..

It is important to distinguish the responsibilities of the organizations involved:

  • IEEE 802.11 Working Group, Task Group bn: Developing the IEEE 802.11bn technical standard.
  • Intel and Broadcom: Conducted the cross-vendor interoperability demonstration using their respective prototype devices.
  • Rohde & Schwarz: Demonstrated Wi-Fi 8 RF signaling test capabilities with Broadcom using its CMX500 one-box signaling tester. The platform supports RF testing of Wi-Fi 8 devices, including physical-layer measurements such as distributed Resource Unit (dRU) and User EVM Quality Metric (UEQM). This is RF conformance and performance testing, distinct from cross-vendor interoperability demonstrations and Wi-Fi Alliance certification.
  • Wi-Fi Alliance: Provides an industry framework for Wi-Fi interoperability and product certification. Its members’ meeting hosted the demonstration, but the demonstration itself was not a Wi-Fi Alliance certification event.
  • Dynamic Spectrum Alliance: Organized the colocated Global Summit. It was not responsible for developing IEEE 802.11bn or conducting the Intel-Broadcom test.

This distinction matters. Cross-vendor interoperability demonstrations help identify implementation problems and establish that equipment from different manufacturers can work together. Formal Wi-Fi Alliance certification is a separate process involving defined specifications and testing requirements.

The October 27th demonstration was an encouraging interoperability milestone, not proof of universal compatibility across Wi-Fi 8 products or features.

Non-Primary Channel Access (NPCA): Improving Spectrum Utilization:

The demonstration focused on Non-Primary Channel Access (NPCA), one of the MAC-layer enhancements being developed for Wi-Fi 8. NPCA is designed to improve spectrum utilization when multiple access points compete for access to the same wide channel.

Consider a dense wireless deployment in which two access points operate within the same 160 MHz channel. One access point may transmit using only part of that channel bandwidth, while another access point must wait for the primary channel to become available. This can leave portions of the wider channel temporarily underutilized.

NPCA enables a Wi-Fi 8 device, under specified operating conditions, to access an available non-primary portion of the channel rather than unnecessarily waiting for the entire channel to become available. The objective is to improve channel utilization, reduce contention and increase aggregate network throughput without requiring additional spectrum. This is particularly relevant to enterprise WLANs, high-density residential environments, public venues and other deployments where multiple access points share limited unlicensed spectrum.  NPCA is an example of the broader UHR philosophy: improving the efficiency and reliability of existing wireless resources instead of relying exclusively on wider channels or higher-order modulation.

Demonstrated Throughput and Latency Improvements:

Intel and Broadcom constructed a live demonstration to illustrate NPCA’s potential benefits.

The initial configuration showed Wi-Fi 7 network performance without additional interference. A second access point was then activated on the same channel, increasing contention and reducing throughput as the two networks competed for spectrum access.

The demonstration subsequently enabled Wi-Fi 8 NPCA functionality.

According to the reported results, enabling NPCA produced:

  • Approximately 30% higher aggregate throughput.
  • Up to 50% lower latency compared with the baseline scenario.

These improvements were achieved using existing spectrum resources rather than additional channel allocations.

These numbers require careful interpretation. They are results from a specific demonstration configuration, not independently verified performance improvements applicable to all Wi-Fi 8 deployments.

Actual results will depend on channel conditions, network configuration, access-point density, device capabilities, interference and traffic patterns. The reported improvements should therefore be viewed as evidence of the potential benefits of NPCA, not as guaranteed performance gains for commercial Wi-Fi 8 networks.

Intel also prepared a demonstration of additional intermediate Modulation and Coding Scheme (MCS) levels. These were not presented during the main event because of time constraints.

Additional MCS levels can provide finer adjustments to transmission rates as radio conditions change, potentially improving spectrum efficiency and link performance.

Other Organizations Advancing Wi-Fi 8 Testing:

Cross-vendor interoperability is only one part of the testing required before commercial Wi-Fi 8 products become widely available.

Several organizations are contributing to different aspects of Wi-Fi 8 development and validation:

  • Rohde & Schwarz and Broadcom

In February 2026, Rohde & Schwarz announced Wi-Fi 8 RF signaling test capabilities for its CMX500 one-box signaling tester.

Working with Broadcom, Rohde & Schwarz demonstrated testing of a prototype Wi-Fi 8 device at the physical layer. Its testing capabilities address new IEEE 802.11bn features, including distributed Resource Units (dRU) and unequal modulation (UEQM).

These tests are important because cross-vendor interoperability alone does not establish that a device meets all required RF performance specifications. RF signaling tests provide a means to evaluate device behavior and implementation of PHY-layer capabilities under controlled conditions.

  • Wi-Fi Alliance

The Wi-Fi Alliance plays a separate industry role in interoperability testing and certification of Wi-Fi products.

Its certification programs help ensure that products from different manufacturers conform to applicable Wi-Fi specifications and can operate together.

As IEEE 802.11bn progresses toward completion, the Wi-Fi Alliance will have an important role in developing the associated Wi-Fi 8 certification program and testing requirements.

The distinction between IEEE standardization and Wi-Fi Alliance certification is fundamental: IEEE develops the underlying technical standard, while the Wi-Fi Alliance develops industry certification programs to promote interoperability and product compatibility.

  • Intel and Broadcom

The October demonstration is particularly significant because it moved beyond testing an individual manufacturer’s prototype against its own equipment.

Interoperability between a Broadcom access point and an Intel client demonstrates progress toward cross-vendor operation, an essential requirement for a successful Wi-Fi ecosystem.

Intel has indicated that additional interoperability testing, including a planned industry plugfest around the third quarter of 2027, is expected before its Wi-Fi 8 client products reach commercial introduction.

What Comes Next?

Wi-Fi 8 still faces substantial engineering and standardization work.

The IEEE 802.11bn amendment must complete its draft development, comment resolution and approval processes. Chipset manufacturers and equipment vendors must continue validating PHY and MAC functionality, RF performance and interoperability.

Additional cross-vendor plugfests will be important for identifying implementation differences and testing more combinations of access points, client devices and operating conditions.

The Wi-Fi Alliance will also need to establish its certification program and associated testing procedures. A successful Intel-Broadcom demonstration does not replace that process.

The October demonstration is an encouraging indication that the ecosystem is moving beyond individual prototype development toward cross-vendor interoperability. Nevertheless, a single successful demonstration does not establish full interoperability across the many device configurations, operating conditions and features contemplated by IEEE 802.11bn.

Conclusions:

Wi-Fi 8 represents an important evolution in wireless LAN technology. Its emphasis on Ultra High Reliability, improved spectrum utilization, lower latency and more consistent performance addresses practical shortcomings that cannot necessarily be solved by simply increasing channel bandwidth or peak data rates.

The Intel-Broadcom demonstration of IEEE 802.11bn-based equipment, particularly its use of Non-Primary Channel Access, provides an early indication of how these improvements may be achieved.  This demo marks a development milestone, but the reported measurement results are not yet finalized Wi-Fi 8 performance numbers.

Three separate milestones will determine progress toward commercial Wi-Fi 8 deployment:

  1. IEEE standardization: Completion and approval of the P802.11bn UHR amendment.
  2. Cross-vendor interoperability: Continued testing by chipset and equipment manufacturers, including Intel and Broadcom, to establish that independently developed implementations work together.
  3. Industry certification: Development of Wi-Fi 8 certification and testing programs by the Wi-Fi Alliance.

The ultimate measure of Wi-Fi 8’s success will not be its theoretical maximum speed, but whether it delivers more reliable, predictable and efficient wireless connectivity in real-world deployments.

References:

https://standards.ieee.org/ieee/802.11bn/11393/

https://www.ieee802.org/11/Reports/tgbn_update.htm

https://www.ieee802.org/11/Reports/802.11_Timelines.htm

https://ieee802.org/11/email/stds-802-11/msg09578.html

https://www.rohde-schwarz.com/us/about-us/news-press/all-news/rohde-schwarz-and-broadcom-showcase-first-wi-fi-8-rf-signaling-tests-paving-way-for-next-gen-connectivity_229356-1610900.html

https://www.rohde-schwarz.com/us/solutions/wireless-communications-testing/wireless-standards/wlan-wi-fi/wi-fi-8-testing/wi-fi-8-testing_258504.html

Broadcom with Samsung Electronics: Integrated 5G and Wi-Fi 8 FWA Platform

Analysis: Broadcom’s end-to-end 50G PON Edge AI portfolio with WiFi 8 support

Qualcomm FastConnect 7800 combining WiFi 7 and Bluetooth in single chip

 

 

Can Digital Twins Deliver Real Energy Savings for AI Data Center Cooling?

Posted on by Alan Weissberger

Disclaimer: ChatGPT was used extensively for research and writing this article.

………………………………………………………………………………………………………………….

Introduction:

Data center cooling systems consume significant auxiliary energy, yet optimization studies rarely quantify the gap between theoretically optimal and operationally deployable control strategies. The rapid expansion of AI infrastructure is creating a difficult engineering problem: how to remove increasing amounts of heat from densely packed accelerators without consuming an excessive share of the electricity needed to run them. Digital twins—virtual models of physical systems—are increasingly proposed as a way to optimize cooling. But can they consistently deliver meaningful energy savings in operating data centers, or will their benefits fall short of simulations and vendor expectations?

A new collaboration between Synopsys and the National University of Singapore (NUS) aims to help answer that question. Announced on October 29th, the project will combine engineering simulation, AI-assisted design, reduced-order modeling and physical testing through the Sustainable Tropical Data Center Testbed Phase 2.0 (STDCT 2.0). Its initial research concepts include two-phase direct-to-chip liquid cooling for high-heat AI accelerators and digital twins for predictive thermal management.

The proposed work is relevant well beyond Singapore. Data centers operating in hot, humid climates face demanding cooling conditions, while AI workloads are pushing rack power densities higher. NUS has identified cooling and ventilation as major contributors to data-center energy consumption in tropical environments. However, the collaboration is a research and validation effort—not evidence that the proposed digital twins have already delivered commercial-scale savings.

Brandon Wang, Strategic Management, Vice President, Technology Strategy & New Horizons, Synopsys, said:

“At Synopsys, we look beyond today’s opportunities to the new horizons that can strengthen our technology long term. Through our Technology Intelligence Center, our team connects emerging technology shifts with Synopsys capabilities and leading research ecosystems to turn promising ideas into strategic opportunities. Our collaboration with STDCT 2.0 reflects that commitment by bringing our silicon-to-systems engineering capabilities into a real-world environment for sustainable AI infrastructure.”

Source:  Synopsys announcement via HPCwire.

What a digital twin can and cannot do:

A digital twin combines a model of a physical system with operational data to estimate its current state, predict its behavior and help determine how it should be operated. For data-center cooling, that can mean modeling coolant flow, temperatures, heat exchangers, pumps and the interaction between IT equipment and facility cooling systems.

The potential benefit is straightforward. Cooling equipment is often operated conservatively to maintain safe temperatures under changing workloads and environmental conditions. A sufficiently accurate model could identify opportunities to adjust coolant flow, supply temperatures and other control settings while maintaining thermal limits.

But a digital twin does not inherently save energy. It provides predictions and recommendations; actual savings depend on the accuracy of those predictions, the available control mechanisms and whether the operating system can safely implement the recommended changes.

There is also an important distinction between using a digital twin to evaluate a proposed design and using it to optimize a live data center. A model that predicts the behavior of a system under selected test conditions may not remain accurate as rack configurations, workloads, ambient temperatures and cooling hardware change.

The modeling challenge: accuracy under changing conditions:

Cooling a modern AI data center is a multiscale engineering problem. Heat originates at individual chips, moves through server cooling components and coolant loops, and ultimately must be rejected through facility-level equipment. The system’s behavior depends on interactions among thermal loads, fluid dynamics, control systems and the surrounding environment.

High-fidelity computational fluid dynamics and multiphysics simulations can help engineers understand these interactions. But detailed simulations can be computationally expensive, making them impractical for every real-time operational decision.

This is where reduced-order models (ROMs) become important. They approximate a more complex physical model using a computationally simpler representation. A digital twin can use such a model to estimate temperatures and evaluate operating changes more quickly.

The trade-off is that simplification can sacrifice accuracy outside the conditions used to develop and validate the model. AI-assisted models introduce additional questions: How well do they generalize to previously unseen conditions? How are prediction errors detected? What happens when sensors drift, telemetry is missing or an operating regime changes abruptly?

Synopsys and NUS propose to connect high-fidelity simulation with measurements from a physical testbed before developing reduced-order models. This is the right general direction. The critical test, however, will be whether model accuracy remains adequate across a broad range of operating conditions—not simply whether a model matches a limited set of laboratory measurements.

Evidence that optimization can work—but with constraints:

Research on liquid-cooled supercomputers illustrates both the opportunity and the need for caution.  A 2026 study, Digital Twin-Based Cooling System Optimization for Data Center, examined liquid-cooling infrastructure at the Frontier exascale supercomputer. The researchers developed a surrogate model and validated it against a year of operational data. Their modeled optimization scenarios reported energy savings of 20.4% for flow-only optimization, 30.1% when jointly optimizing coolant flow and supply temperature, and 27.8% when practical actuator ramp-rate constraints were included.

These are encouraging results, but they are specific to the system, baseline operating conditions and optimization methods studied. They should not be interpreted as a forecast of savings for a typical enterprise or hyperscale AI data center. Nor do model-based estimates automatically establish that the same savings can be sustained after deployment.

The result also illustrates a central engineering point: the theoretical optimum is not necessarily the operational optimum. Cooling controls must respect equipment limits, thermal safety margins, response times and reliability requirements.  Source: Research paper.

Image Generated by ChatGPT

………………………………………………………………………………………………………………………………………………………………………………………………

Five tests for a credible cooling digital twin explained:

Before data-center operators accept claims of substantial digital-twin-driven energy savings, they should ask for evidence on five fronts.

1. Independent validation of Model Accuracy. Does the model predict measured temperatures, flow rates and energy consumption accurately across different workloads, ambient conditions and cooling configurations? Validation should use data not employed to tune the model.

2. Net energy savings. Are savings measured across the complete cooling system, including pumps, coolant distribution units, heat rejection and other auxiliary equipment? Reducing one component’s power consumption is not enough if another component consumes more energy as a result.

3. Thermal safety and reliability. Can the controller respond safely to sudden workload changes, component failures, sensor errors and loss of cooling capacity? Energy optimization must not increase the risk of thermal excursions or equipment shutdowns.

4. Water and environmental impact. Does the proposed operating strategy reduce electricity consumption at the expense of greater water use or other environmental costs? The answer depends on the cooling architecture and local conditions, so both energy and water should be measured.

5. Sustained, real-world performance. Are the reported savings maintained over weeks or months, including seasonal changes and varying AI workloads? A controlled experiment is useful, but sustained operational evidence is more persuasive.

–>The comparison must also use an appropriate baseline. If a digital twin is compared with an unnecessarily conservative cooling configuration, the reported savings may overstate the benefit available in a well-tuned facility. Operators should compare it with existing best-practice controls under equivalent workloads and environmental conditions.

Why the Synopsys–NUS collaboration matters:

The STDCT 2.0 testbed offers a potentially useful environment for evaluating cooling technologies under tropical operating conditions. The collaboration’s proposed sequence—high-fidelity simulation, physical validation, reduced-order modeling and digital-twin integration—provides a framework for testing whether virtual models can support reliable operational improvements.

The planned investigation of two-phase direct-to-chip cooling is also significant. As AI accelerator heat flux rises, removing heat close to its source can become increasingly important. Yet the performance of a cooling technology cannot be judged by heat-transfer capability alone. Pressure drop, pumping requirements, fluid management, maintenance, reliability and heat rejection all affect the system-level result.

The same standard should apply to digital twins. A sophisticated model is not an outcome in itself. Its value lies in whether it helps operators make better decisions than existing monitoring and control systems—and whether those decisions produce measurable improvements without compromising reliability.

Conclusions: 

The key take-away is to measure the savings, not the sophistication of the proposed solution for data center cooling.  Digital twins could become valuable tools for managing AI data-center cooling, particularly as liquid-cooling systems grow more complex and operating conditions become less predictable. However, their effectiveness depends on the quality of the underlying physics, the representativeness of validation data, the reliability of sensors and controls, and the ability to maintain accuracy as the physical system changes.

The Synopsys–NUS project is worth watching because it proposes to connect simulation with physical testing rather than relying on modeling alone. Its significance will ultimately depend on the results: independently validated model accuracy, sustained reductions in total cooling energy, acceptable water use and reliable operation under real workloads.

For data-center operators, the decisive question is not whether a digital twin can predict cooling behavior. It is whether using one saves more energy than existing controls—consistently, safely and at an operational cost that makes deployment worthwhile.

……………………………………………………………………………………………………………………………………………………………..

References:

https://www.hpcwire.com/off-the-wire/synopsys-joins-singapore-consortium-to-improve-tropical-data-center-cooling/

https://cde.nus.edu.sg/stdct/

https://arxiv.org/abs/2410.05133

https://arxiv.org/abs/2605.15516

https://arxiv.org/abs/2610.33148

https://arxiv.org/abs/2603.01198

https://www.sciencedirect.com/science/article/pii/S0306261924012236

https://www.sciencedirect.com/science/article/pii/S030626192501791X

https://doi.org/10.1016/j.apenergy.2024.125100

AI Power Grid Crisis Triggers $30 Billion “Fuel-Cell Boom” for Data Centers

Telecom data centers must be redesigned for the AI era with rack scale architectures, enhanced power & cooling requirements

Liquid Dreams: The Rise of Immersion Cooling and Underwater Data Centers

PwC: Global AI data center spending to hit $31.6tn by 2050; Role of full stack orchestration layer explained

AI Compute Has a Switchboard Problem: Orchestration & Data Center Fabric Explained

Dell’Oro: Data Center Physical Infrastructure revenue to grow at 22% CAGR from 2025-2025/forecast comparisons, analysis, risks

 

T-Mobile’s Strategy of Deploying Multi-Vendor AI-RAN on “Kinetic Edge” using AWS Pillars

Posted on by Alan Weissberger

Introduction:

During its Q3 2026 earnings call, T-Mobile US announced a definitive step away from traditional, single-purpose radio access networks (RANs) by detailing the real-world deployment of its “Kinetic Edge” [1.] architecture which operates on AWS infrastructure. By combining a multi-vendor AI-RAN footprint with a deeply distributed cloud orchestration layer in partnership with AWS, the U.S. network operator is attempting a fundamental structural shift: transforming cell towers into shared platforms capable of running concurrent cellular routing and third-party AI compute workloads.

Note 1.  The T-Mobile Kinetic Edge architecture (announced in late October 2026 in partnership with AWS) is a distributed intelligence infrastructure designed to support Physical AI at production scale. It eliminates the need for expensive onboard computing on machines (like humanoid robots or delivery drones) by moving sub-millisecond AI inference workloads directly to the network edge.  The underlying infrastructure layout, session parameters, and technical architecture of this deployment were discussed at the AWS re:Invent 2026 Event Catalog (Session IND384).That session outlined how T-Mobile operates its physical AI infrastructure at production scale for its 130M+ subscriber base using three integrated AWS pillars:

  • AWS Outposts (The Hardware Plane): Deployed directly within T-Mobile’s localized edge sites and central offices to provide a fully managed, low-latency hybrid hardware footprint. Outposts host the compute layer, keeping the execution of physical AI inference nodes inches away from the 5G user plane function (UPF) to maintain sub-millisecond execution loops. 
  • Amazon SageMaker (The Model Pipeline): Handles the training, optimization, and continuous refinement of the complex spatial telemetry, real-world vision, and physical reasoning models. It allows T-Mobile to continuously push optimized model variations out to edge containers. 
  • Amazon Bedrock AgentCore (The Orchestration Engine): Serves as the backbone platform to host, scale, and secure autonomous, long-running agentic workloads. It decouples the agent logic from the backend plumbing, utilizing serverless microVM environments for strict session isolation per workload.
    • AgentCore Identity & Policy: Enforces rigid, real-world permissions for unattended physical fleets via Cedar policies, ensuring autonomous edge actions remain safely sandboxed.
    • AgentCore Gateway & Memory:  Exposes internal APIs as unified Model Context Protocol (MCP) tool endpoints, while retaining stateful context across active device sessions.

Architectural Topology Breakdown:

Layer Technology Primary Function
Physical AI Edge Device Humanoid Robots, Vision Cameras, Fleet Sensors Gathers real-time sensory inputs, handles localized movement commands, offloads processing.
Deterministic Wireless Fabric T-Mobile 5G-Advanced Network & AI-RAN Provides steady, low-latency uplink/downlink to transfer raw sensor streams and return telemetry.
Local Compute Node AWS Outposts Processes sub-millisecond inferencing and local data breakouts away from the distant central cloud.
Distributed AI Infrastructure Amazon SageMaker & Bedrock AgentCore Handles localized vision, reasoning, and machine learning models dynamically.
………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………….

CEO Srini Gopalan framed this structural shift as a transition away from standard, legacy telecom metrics to running distributed token-based intelligence. He stated, “Our future value is no longer bound to basic SIM cards or standard data buckets.  We are engineering a platform to run distributed, token-based intelligence for millions of humanoid robots, drones, and autonomous systems by 2035.”

T-Mobile’s new network architecture serves as an ambitious architectural goal—and raises critical engineering questions—regarding how operators hope to eventually balance deterministic radio schedules with dynamic enterprise inference at the network edge.

The Baseband Convergence: Multi-Vendor AI-RAN at the Cell Site:

The foundational layer of the Kinetic Edge is T-Mobile’s verified execution of a commercial AI-RAN connection in Bellevue, Washington, which integrates Nvidia’s accelerated computing platform with Nokia’s 5G Standalone (SA) macro base stations.  In traditional architectures, the Distributed Unit (DU) relies on highly specialized, fixed-function ASICs to execute Layer 1 (Physical Layer or PHY) digital signal processing. Under the AI-RAN framework, these workloads are shifted onto software-defined, GPU-accelerated hardware profiles. This allows the physical cell site footprint to execute two distinct tasks simultaneously:

    • Inline Layer 1 Acceleration: Executing high-density 5G-Advanced signal processing—such as massive MIMO beamforming calculation and low-density parity-check (LDPC) decoding—directly within a programmable software stack.
    • Opportunistic AI Inference: Utilizing unallocated or transiently idle compute cycles on the exact same silicon to host localized machine learning models, edge vision analytics, and autonomous system telemetry.

Ankur Kapoor, EVP & Chief Network Officer, explained the software philosophy behind this infrastructure trial. “We cannot rely entirely on rigid, single-purpose ASICs if we want an agile network layer,” Kapoor noted. “Our engineering team is actively steering a hybrid model, developing software layers optimized for both specialized merchant silicon and accelerated GPU stacks, allowing us to pivot compute resources on demand.”  The immediate engineering challenge remains operational isolation. Unlike standard data center servers where a few milliseconds of compute jitter goes unnoticed, cellular networks operate under strict microsecond-level hybrid automatic repeat request (HARQ) deadlines. T-Mobile’s deployment must prove that heavy enterprise AI inference spikes cannot bleed into the compute budget reserved for timing-critical radio resource management.

Distributed Cloud Orchestration: The AWS “Kinetic Edge” Fabric:

To push deterministic connectivity out to the local cell edge, T-Mobile is wrapping its physical AI-RAN nodes in a distributed cloud framework managed via AWS. This architecture essentially stretches the AWS cloud control plane down to the operator’s regional hubs and central office cell aggregation sites.

       [ AWS Core Cloud Control Plane ]
                     │
                     ▼
       [ T-Mobile Regional Edge Hubs ]
   (Centralized Control & Heavy Orchestration)
                     │
                     ▼
       [ Localized Cell Site Node ]
 ┌───────────────────────────────────────────┐
 │  Co-located AI-RAN Compute Infrastructure │
 ├─────────────────────┬─────────────────────┤
 │   Deterministic     │    Opportunistic    │
 │  5G Layer 1/2 Stack │ Enterprise AI Stack │
 └─────────────────────┴─────────────────────┘

CTO Dr. John Saw detailed the core mechanics of this distributed offload strategy in a technical brief accompanying the announcement. “To unlock true physical AI, we must eliminate core network transport latency,” Dr. Saw stated. “The Kinetic Edge is designed to offload heavy vision, spatial telemetry, and real-world reasoning AI models onto distributed cell-site nodes, putting the execution engine inches away from the user plane function (UPF).”

Saw expects cellular traffic patterns in 2030 to be different from today as AI usage increases. Saw said his mobile traffic assumption for 2030 is “still downlink heavy but increasingly uplink rich,” driven by AI wearables, AI agents and “explosive growth by then of physical AI devices, like humanoids, robots, delivery drones, vehicles and factory machines.”  Check out Saw’s Physical AI blog here.

Chris Sambar, Chief Enterprise Officer, emphasized how this framework underpins T-Mobile’s enterprise monetization strategy. “Our B2B priority is monetizing the physical edge,” Sambar stated. “By exposing this low-latency AWS fabric to enterprise clients, we can guarantee deterministic, localized compute slots for industrial computer vision and robotics coordination directly on our infrastructure.”

The control plane must intelligently manage this distribution. Heavy macro-orchestration and model training remain centralized within regional edge hubs, while light, highly optimized model weights are pushed to individual edge nodes based on localized user density and enterprise demand.

Intelligent Radio Resource Management (RRM) Overlays:

Beyond hosting third-party workloads, the integration of AI directly into the RAN stack redefines internal network management. T-Mobile’s architecture leverages supervised machine learning models to build automated, predictive Radio Resource Management (RRM) overlays.

Traditional RRM relies on reactive, hard-coded algorithms to handle handovers, power allocation, and frequency scheduling based on past channel quality indicators (CQI). Ankur Kapoor highlighted how AI shifts this paradigm: “Our current live trials utilize AI-managed beamforming and predictive radio resource scheduling directly across subframes. The network no longer just reacts to interference; it models local spatial telemetry to preemptively mitigate link degradation before it impacts the terminal device”.

  1. Spatial Telemetry Ingestion: The AI-RAN layer continuously ingests massive streams of real-world spatial data and device behavior patterns across the sector.
  2. Predictive Beamforming Adjustment: Instead of waiting for a signal link to degrade, the predictive models anticipate localized network congestion, building clutter interference, or sudden user clustering.
  3. Dynamic Resource Re-allocation: The radio subframe allocation and multi-antenna beam vectors are adjusted dynamically ahead of time, maximizing spectral efficiency and keeping physical resource blocks (PRBs) optimally distributed.

The Interoperability and 6G Transition Horizon:

T-Mobile’s Q3 2026 milestones offer a compelling look at the future of network monetization, but the architecture’s long-term viability hinges on open standards. While this deployment successfully pairs Nokia and Nvidia architectures on a 5G Standalone core, the telecom industry must demand independently reproducible multi-vendor interoperability.

Dr. John Saw explicitly tied these architectural steps to the operator’s long-term 6G strategy, setting an aggressive timeline. “T-Mobile is anchoring a firm 2030 commercial deployment timeline for the nation’s first 6G network,” Dr. Saw declared. “6G will natively expand on this Kinetic Edge foundation, shifting from traditional cell communication to a ‘radar-behaving’ topology that pairs Integrated Sensing and Communication (ISAC) with uplink-dense architectures.”  The upcoming FCC spectrum auctions for upper C-band and 2.7GHz offer T-Mobile opportunity to add capacity for 6G as well as to boost uplink bandwidth, Saw added.

As 3GPP Release 20 and Release 21 move through their respective study and normative spec cycles towards 2027–2029, these real-world deployment lessons must be codified. Ensuring that shared-compute RAN platforms are open and interchangeable across competing hardware and software vendors will determine whether architectures like the Kinetic Edge become the universal foundation for 6G, or remain isolated, proprietary engineering triumphs.

………………………………………………………………………………………………………………………………………………………………………………………

References:

https://finance.yahoo.com/quote/TMUS/earnings/TMUS-Q3-2026-earnings_call-713930.html

https://www.t-mobile.com/news/network/physical-ai-is-leaving-the-lab

https://registration.awsevents.com/flow/awsevents/reinvent2026/eventcatalog/page/eventcatalog?search=IND384

https://aws.amazon.com/blogs/industries/your-guide-to-aws-for-telecom-at-reinvent-2026/

https://aws.amazon.com/blogs/architecture/lets-architect-architecting-for-the-edge/

https://www.lightreading.com/6g/t-mobile-we-ll-be-first-with-6g-and-no-to-spacex-mvno

T-Mobile US announces new broadband wireless and fiber targets, 5G-A with agentic AI and live voice call translation

AWS to deploy AI inference chips from Cerebras in its data centers; Anapurna Labs/Amazon in-house AI silicon products

AT&T and AWS to deliver last mile connectivity for AI workloads; AT&T Geo Modeler™ AI simulation tool

PsiQuantum Targets AI Data Center Networks with Silicon-Photonic Optical Switching

Posted on by Alan Weissberger

Disclaimer:  ChatGPT was used extensively for research contained in this article.

……………………………………………………………………………………………………………………………………..

Introduction:

As AI workloads expand and AI clusters grow, the networks connecting GPUs and other accelerators are becoming an increasingly important constraint on performance, power consumption and cost. Optical circuit switching (OCS) offers one way to improve connectivity by establishing direct optical paths between network ports without converting optical signals into electrical signals for packet-by-packet switching.

On October 28th, PsiQuantum [1.] unveiled an 8×8 optical circuit-switching demonstration built on its low-loss silicon-photonics platform, originally developed for quantum computing. The company reports promising laboratory results and is sharing demonstration samples with selected prospective customers. The key question is whether this technology can scale into a commercially competitive product for AI data centers.

Note 1.  PsiQuantum was founded on the premise that photonic qubits, together with leverage of mature semiconductor manufacturing processes, would provide the fastest path to commercially useful, fault-tolerant quantum computers. Today, PsiQuantum builds tens of thousands of wafers of quantum chips in a tier-1 foundry, operates cryogenic cabinets supporting hundreds of chips, advances fault-tolerant algorithm development to solve critical problems across industries, and is developing a modular quantum computing platform designed to increase capability and scale. Each milestone brings us closer to delivering commercially useful quantum computing. The company’s headquarters is in Palo Alto, CA.  Source: PsiQuantum

Optical Circuit Switching (OCS) for AI Data Centers:

Large AI clusters rely on high-speed interconnects to move model parameters, gradients, activations and other data among processors. Network congestion, communication overhead and power consumption can limit how effectively expensive accelerators are utilized.  Ethernet and InfiniBand networks rely primarily on packet switches to route traffic dynamically.

OCS takes a different approach: it establishes direct optical connections between selected ports, allowing light to pass through the switching fabric without optical-electrical-optical conversion at the switch. This approach can complement conventional packet switching in AI clusters, particularly when communication patterns are predictable or recurring. A network controller can establish optical paths between selected racks or groups of accelerators, while packet switches handle traffic requiring flexible, fine-grained routing.

OCS is therefore a potential complement to conventional switching fabrics, not a wholesale replacement for Ethernet or InfiniBand. Its benefits depend on network architecture, traffic patterns, connection utilization and the overhead of reconfiguring optical paths.

Here’s an illustration of optical circuit switching in an AI data-center network. Optical circuit switches establish direct optical paths between selected network ports, complementing conventional packet switching. Conceptual illustration—not a depiction of PsiQuantum’s specific product or a verified implementation.

Image Credit: ChatGPT

………………………………………………………………………………………………………………………………………………………………………………………………………….

PsiQuantum’s 8×8 Demonstration:

PsiQuantum describes its first publicly disclosed device as a strictly non-blocking 8×8 switch, meaning any input can connect to any output. The silicon-photonic device operates without moving parts.


 

 

 

 

 

 

 

 

 

 

 

 

 

The company reports the following laboratory results:

Parameter Company-reported result
Switch configuration 8×8
Typical insertion loss 1.2 dB
Crosstalk 40 dB
Reconfiguration time Less than 3 ms
Optical amplification Not required

Source: PsiQuantum, October 28, 2026. These are company-reported laboratory results, not independently verified benchmarks.

“This opportunity exists because we chose to solve the hardest scaling problems at the beginning of our roadmap to utility-scale quantum computing,” said Victor Peng, Chief Executive Officer of PsiQuantum. “As AI computing pushes infrastructure to its limits, we now have an opportunity to put our state-of-the-art technology to work today while continuing toward our ultimate goal: building and deploying the world’s first useful, fault-tolerant quantum computers.”

“AI is driving extraordinary growth in the scale and complexity of computing infrastructure, and optics will play an increasingly important role in enabling that growth,” said Rob Soderbery, Executive Vice President at PsiQuantum. “PsiQuantum has spent nearly a decade developing silicon photonics against some of the most demanding requirements imaginable. What we are demonstrating today is that this platform can deliver exceptional performance for optical circuit switching for AI computing.”

PsiQuantum has also developed barium titanate (BTO) integration at the 300mm wafer scale and uses a 64×64 BTO switch in its quantum systems. BTO is not required for the demonstration vehicle or initial products announced today, highlighting the performance of the underlying silicon photonics platform while preserving an additional path to faster, more advanced future products.

………………………………………………………………………………………………………………………………………………………………………………………………

EE Analysis:

Insertion loss measures the optical power lost as a signal passes through the switch. The reported 1.2 dB figure suggests that relatively little power is lost in the device itself.

Crosstalk measures unwanted signal leakage between channels, while reconfiguration time indicates how quickly the switch can establish a new connection arrangement.

A switching time below 3 milliseconds could support some scheduled or recurring AI communication patterns. However, suitability depends on workload characteristics, connection-management overhead and how effectively established paths are utilized.

PsiQuantum says its silicon-photonics platform leverages semiconductor manufacturing at Global Foundries in New York. It has also developed barium titanate integration at 300-mm wafer scale, but the company says this technology is not required for the demonstrated switch or its initial products. Its 8×8 device should therefore be judged on its own reported performance, rather than the potential capabilities of future products.

……………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………..

Google Provides a Production-Scale Benchmark:

Optical circuit switching is already deployed in hyperscale data-center networks. Google’s Jupiter architecture provides an important real-world comparison.

In a 2022 ACM SIGCOMM paper, Google engineers described how Jupiter evolved from a conventional Clos network toward a direct-connect topology between aggregation blocks. Its Apollo optical circuit switches use micro-electromechanical systems (MEMS) mirrors to steer light between fiber ports. A software-defined networking (SDN) control plane dynamically adjusts the network’s logical topology and traffic paths, while packet switches continue to handle traffic routing.

Google’s paper reported that the broader evolution of Jupiter delivered a fivefold increase in network speed and capacity, a 30% reduction in capital expenditure and a 41% reduction in power consumption while serving live production traffic. These improvements resulted from the overall architecture—including optical switching, traffic engineering and network-management changes—not from the optical switches alone.

The distinction is important. Google has demonstrated production-scale OCS integrated with network control and traffic engineering. PsiQuantum has reported promising component-level results from an 8×8 silicon-photonic laboratory demonstration. The two systems also use different switching technologies.

Google’s experience nevertheless demonstrates that OCS can deliver substantial benefits when integrated into a carefully engineered network. PsiQuantum’s challenge is to show that its silicon-photonics approach can provide comparable practical value at scale.

The Commercial Test:

PsiQuantum must now demonstrate that its reported performance can be maintained in larger configurations and integrated into real AI data-center networks. The main questions are:

  • Scalability: Can port counts increase without unacceptable optical losses, crosstalk, complexity or cost?
  • Network integration: Can the switches work effectively alongside existing Ethernet and InfiniBand fabrics, with appropriate traffic scheduling and network management?
  • Power and economics: Will the complete implementation reduce system-level power consumption or operating costs enough to justify adoption?
  • Manufacturing and reliability: Can PsiQuantum deliver consistent production quality, operational reliability and competitive costs at the volumes data-center customers require?

The company says it plans to announce commercial products in the coming months. Larger configurations, independent performance measurements, customer trials and system-level demonstrations will be important milestones.

Lessons from the All-Optical Networking Hype Dictate Caution:

During the 1998–2001 telecom boom, all-optical networking was heavily promoted as the future of telecommunications, promising to eliminate electronic switching bottlenecks by keeping traffic optical throughout the network. Despite substantial investment and years of development, that broad vision never became the dominant commercial network architecture. NTT’s current IOWN and All-Photonics Network initiatives reflect renewed ambitions for end-to-end optical connectivity, but they do not erase the industry’s long history of overpromising.

This history warrants caution in evaluating PsiQuantum’s 8×8 demonstration. Optical circuit switching has proven useful in specific architectures, including Google’s production data-center network, but that does not establish a broad emerging market for every new OCS technology. PsiQuantum must demonstrate that its silicon-photonic approach can scale economically and deliver measurable system-level benefits in real AI data centers—not simply promising laboratory specifications.

Earlier all-optical-networking initiatives illustrate why attractive component specifications do not guarantee commercial adoption. However, OCS already has production deployments. The question here is whether PsiQuantum’s implementation offers a competitive advantage in a clearly defined deployment—not whether optical switching itself is viable.

Conclusions:

PsiQuantum’s 8×8 demonstration extends its silicon-photonics technology beyond quantum computing into a potentially valuable AI infrastructure market. Its reported insertion loss, crosstalk and sub-3-millisecond reconfiguration time warrant attention, but do not yet establish commercial scalability or an economic advantage over existing technologies.

Google’s Jupiter network shows that optical circuit switching can deliver benefits at hyperscale when combined with software-defined networking and traffic engineering. PsiQuantum must demonstrate that its distinct silicon-photonic approach can deliver similar practical value through larger switches, reliable integration and measurable system-level savings.

For AI data-center operators, the ultimate test is whether optical switching can deliver reliable, scalable and economically attractive connectivity that keeps expensive AI accelerators productively utilized.

The most informative next milestone would be a customer trial reporting port count, fiber-to-fiber loss, reconfiguration interruption, reliability, and workload-level performance and power against a specified baseline. Until then, PsiQuantum has presented a promising switching demonstration, not yet a demonstrated AI-networking advantage.

……………………………………………………………………………………………………………………………………………………………………………………………………..

References:

PsiQuantum, “Optical Circuit Switch.” https://www.psiquantum.com/optical-circuit-switch

PsiQuantum, “PsiQuantum Unveils Optical Circuit Switch Technology Built on Advanced, Low-Loss Silicon Photonics Platform,” October 28, 2026. https://www.businesswire.com/news/home/20261028890054/en/

Leon Poutievski et al., “Jupiter Evolving: Transforming Google’s Datacenter Network via Optical Circuit Switches and Software-Defined Networking,” ACM SIGCOMM 2022. https://doi.org/10.1145/3544216.3544265

Google Cloud, “The Evolution of Google’s Jupiter Data Center Network,” August 23, 2022. https://cloud.google.com/blog/topics/systems/the-evolution-of-googles-jupiter-data-center-network/

NTT’s IOWN is (finally) evolving to an All Photonics Network (APN); Physics based AI for enterprise OT

NTT’s IOWN provides ultra low latency and energy efficiency in Japan and Hong Kong

Oriole Networks photonic networking platform to be integrated with AMD GPUs/CPUs for next-gen AI data center fabrics

Technavio: Silicon Photonics market estimated to grow at ~25% CAGR from 2024-2028

LightCounting: Silicon Photonics chip market to hit $3 billion in 2029

Light Counting on Silicon Photonics and Optical Switching at SC22

Synopsys and Juniper Networks form new company to pursue “open” silicon photonics platform

 

Oracle Integrates NVIDIA and Google AI Accelerators with its High-Performance Cloud Networking Infrastructure

Posted on by Alan Weissberger

Disclaimer:  ChatGPT was used extensively for research included in this article.

…………………………………………………………………………………………………………………………………………………………………

Introduction:

Oracle’s multiple announcements on October 27th illustrate how AI cloud infrastructure is evolving beyond simply providing access to GPUs. Oracle Cloud Infrastructure (OCI) is combining NVIDIA’s latest AI computing systems, Google’s Tensor Processing Units (TPUs), and its Acceleron networking architecture to support increasingly demanding AI training, inference and agentic workloads.

The strategy reflects a fundamental change in the requirements for large-scale AI: accelerator performance alone is insufficient. The network connecting those accelerators, the movement of data between compute and storage, and the ability to sustain performance across very large clusters are equally important.

NVIDIA Vera Rubin systems target large-scale AI workloads:

Oracle and NVIDIA announced that NVIDIA Vera Rubin NVL72 systems and Vera CPUs are generally available on OCI. They also outlined plans to expand OCI Superclusters using Vera Rubin systems to multi-gigawatt scale. Oracle says its Superclusters can scale across hundreds of thousands of GPUs.

The NVL72 architecture connects 72 GPUs within a rack using NVIDIA NVLink scale-up networking. Beyond the rack, Oracle uses an Ethernet-based network fabric to connect systems into much larger clusters.

This distinction between scale-up and scale-out networking is critical. Scale-up interconnects provide high-bandwidth communication among closely coupled accelerators. Scale-out fabrics must move data between racks while managing congestion, latency, failures and the traffic generated by thousands of parallel computing devices.

Large AI training jobs are particularly sensitive to communication delays. Because many operations synchronize across accelerators, a congested link or a slow communication path can leave otherwise productive GPUs waiting for data or results from other GPUs.

Oracle’s challenge is therefore not simply to provide more computing capacity, but to keep that capacity productively utilized as clusters grow.

Google TPUs introduce an alternative AI computing architecture:

In a separate announcement, Oracle and Google said Google’s TPUs will become available through OCI, making Oracle the first hyperscale cloud provider outside Google to offer the accelerators.

Google’s eighth-generation TPU portfolio includes two architectures with different priorities:

  • TPU 8t: Designed primarily for large-scale AI training.
  • TPU 8i: Designed for inference, including workloads involving iterative reasoning and AI agents.

Google’s TPUs are purpose-built accelerators supported by Google’s software ecosystem. Oracle plans to provide the data-center infrastructure, power, cooling and networking needed to deploy them at scale.

However, the service will not be immediately available: Oracle says TPU availability will begin in calendar 2027 and ramp over multiple years.

The addition of TPUs gives OCI customers another option alongside NVIDIA-based infrastructure. Different accelerators can have different performance characteristics, memory architectures, software requirements and economics, depending on the model and workload.

This does not mean the systems are interchangeable. Moving an application between accelerator platforms may require changes to software, kernels, frameworks, deployment tools and performance optimization. The practical value of choice will depend on how easily customers can adapt their workloads and achieve competitive performance.

Oracle Acceleron addresses the networking challenge:

Oracle’s third major development is its continued expansion of Acceleron, OCI’s networking and infrastructure technology portfolio.

Oracle describes an architecture combining high-performance Ethernet networking, remote direct memory access (RDMA), host-level acceleration and multipath network designs. Its NVIDIA-based infrastructure incorporates Spectrum-X Ethernet switches and other NVIDIA networking components alongside Oracle’s own network architecture and services.  RDMA enables systems to transfer data between memories with less intervention from host CPUs than conventional networking approaches. This can reduce overhead and improve communication efficiency in tightly synchronized distributed workloads.

Oracle also describes a multiplanar network architecture that distributes traffic across multiple paths. Its Multipath Reliable Connection (MRC) transport can spread packets across many paths, helping the network respond to congestion or unstable paths without relying on a single route.  The objective is to maintain predictable communication performance as clusters grow. A network failure or congestion event can affect far more than one connection when thousands of accelerators participate in a coordinated training job.

Image Credit: Oracle

Oracle’s October 27 networking announcement also describes improvements in storage connectivity, congestion management and network resilience. These capabilities are intended to keep distributed AI workloads operating efficiently while improving security and supporting multitenant cloud environments.

The company presents these capabilities as ways to reduce network friction and improve utilization. Their actual benefits will depend on workload characteristics, network configuration, software implementation and operational conditions; independent, comparable benchmarks would be needed to quantify the advantages over competing architectures.

What does this mean for AI cloud architecture?

The combined announcements point to a broader industry shift: AI infrastructure is increasingly designed as an integrated system rather than a collection of independent hardware products.

Three technical issues will be especially important:

1. Communication becomes a limiting factor. As accelerator counts increase, the ability to exchange data efficiently can constrain overall performance. Network topology, congestion control, latency and fault recovery can be as consequential as raw link bandwidth.

2. Heterogeneous computing creates opportunities and complexity. Offering NVIDIA GPUs and Google TPUs allows OCI customers to evaluate different architectures for training and inference. However, software portability, developer tools, model compatibility and operational complexity will influence whether that choice delivers practical benefits.

3. Utilization determines system efficiency. Expensive accelerators generate value only when they are doing useful work. Faster data movement, efficient collective communication, resilient networking and appropriate workload placement can help reduce idle time. The same principle applies to power consumption: infrastructure efficiency must be evaluated across the complete system, not just individual chips.

Oracle is also extending its infrastructure strategy to CPUs and specialized accelerators for inference and agentic applications. That reflects the growing importance of the work surrounding model execution, including tool calls, code execution, data retrieval and orchestration—not just the computation performed inside an AI model.

Oracle’s AI cloud infrastructure integrates accelerator hardware, high-density computing systems and high-performance networking to support distributed AI workloads:

Image Credit: ChatGPT

……………………………………………………………………………………………………………………………………………………………………………………………………………………….

Conclusions:

Oracle’s October 27th announcements bring together three elements of its AI cloud strategy: NVIDIA’s latest accelerator systems, Google’s alternative TPU architecture, and Oracle’s high-performance cloud networking: 

The NVIDIA systems combine NVLink scale-up interconnects with Ethernet-based scale-out fabrics, while Google’s TPU architecture integrates custom silicon, high-bandwidth memory and specialized interconnects. Oracle Acceleron adds RDMA connectivity, congestion control, multipath traffic distribution and automatic failover to help sustain performance across large, distributed AI clusters.

The important technical question is whether these components can deliver reliable, scalable and cost-effective performance for real AI workloads. Hardware specifications and announced deployment capacity are useful starting points, but customers ultimately need end-to-end benchmarks measuring training throughput, inference latency and cost per token, network utilization, storage performance, software compatibility, fault recovery, and power efficiency. For distributed training in particular, goodput—the proportion of time a cluster spends performing productive computation rather than waiting on communication or recovering from failures—is a more meaningful measure than peak compute performance alone.

The competitive battleground is no longer just the AI chip. It is the complete cloud computing and networking system: accelerators, memory, interconnects, Ethernet fabrics, storage and software working together to maximize useful computation per dollar and per watt.

Google says TPU chip-to-chip traffic uses ICI and OCS, while Oracle Acceleron connects TPU capacity to the rest of OCI. That differs from the NVIDIA NVLink scale-up and RoCE scale-out arrangement. Offering both accelerator families through OCI therefore should not be read as establishing a common GPU–TPU interconnect or transparent mixed-accelerator training.

Sidebar – Larry Ellison’s remarkable change of direction on cloud computing:

Oracle’s current AI infrastructure ambitions are particularly striking given Larry Ellison’s earlier skepticism about cloud computing. In 2008, Ellison publicly criticized cloud computing as a repackaging of established technologies under a fashionable label. Oracle’s business at the time centered on enterprise databases, business applications, software licensing and support. Its traditional model depended heavily on customers running Oracle software in their own data centers.

At an Oracle event I attended in spring 2011, that strategy remained intact: Oracle proposed licensing its software to cloud developers rather than becoming a cloud infrastructure provider. “We do not plan to become a cloud service provider,” the Oracle representative told the audience.

That began to change in a big way.  In October 2011, Oracle introduced its public cloud, offering Oracle software and development services over the internet. In 2012, Ellison articulated a broader cloud strategy, although he continued to argue that cloud computing was an evolution of existing technology rather than a fundamentally new computing paradigm.

Oracle’s transformation subsequently accelerated. In 2016, the company launched its modern-generation cloud infrastructure platform, competing more directly with established infrastructure providers such as Amazon Web Services (AWS) and Microsoft Azure.  The strategic change went well beyond delivering existing Oracle software as an online service. Oracle invested in compute, storage, networking, data centers and infrastructure designed for demanding enterprise workloads. Its later expansion into large-scale AI infrastructure has taken that evolution further.

Today, Oracle is deploying NVIDIA’s newest AI systems, preparing to offer Google’s TPUs, and developing networking capabilities intended to support distributed AI clusters at enormous scale.  The historical irony is clear: a company long associated with proprietary enterprise software and database licensing has become a major contender in the infrastructure underpinning modern AI.

Whatever the merits of Ellison’s initial criticism of cloud terminology, Oracle’s subsequent decision to build and operate cloud infrastructure and provide cloud services reversed the position its representative had communicated to attendees in spring 2011. Ellison’s evolution also illustrates an important lesson in technology strategy. A company need not accept every new technology label or prevailing industry narrative to recognize that customer requirements and competitive economics are changing. But adapting successfully may ultimately require much more than adding new products to an existing business—it can require rebuilding the business around a different infrastructure model.

………………………………………………………………………………………………………………………………………………………………………………………………

References:

Oracle and NVIDIA — Vera Rubin systems and Vera CPUs on OCI
https://blogs.oracle.com/cloud-infrastructure/nvidia-vera-rubin-vera-cpus-oci

2. Oracle and Google — Bringing Google TPUs to Oracle Cloud Infrastructure
https://www.oracle.com/news/announcement/ai-world-oracle-brings-google-tpus-to-oracle-cloud-infrastructure-2026-10-27/

3. Oracle — Acceleron networking innovations for AI

https://www.oracle.com/europe/news/announcement/ai-world-oracle-unveils-new-cloud-networking-innovations-for-ai-2026-10-27/
https://blogs.oracle.com/cloud-infrastructure/first-principles-networking-for-ai-scale

https://blogs.oracle.com/cloud-infrastructure/wp-content/uploads/sites/83/2026/10/HP-Fabric-01-1024×464.png

4. WIRED — “Ellison Answers Benioff With Oracle Cloud” (October 2011)
https://www.wired.com/2011/10/ellison-answers-benioff-with-oracle-cloud/

Editorial note: The first three links cover the technical announcements. The WIRED article provides historical context for the sidebar on Larry Ellison’s changing position on cloud computing. Oracle’s official October 27th networking announcement is also available here: https://www.oracle.com/europe/news/announcement/ai-world-oracle-unveils-new-cloud-networking-innovations-for-ai-2026-10-27/

Custom AI Silicon: OpenAI, Anthropic, Hyperscalers and the Battle with Nvidia, AMD and Broadcom

Google’s TPU Business Outpaces Rivals as Hyperscalers Accelerate Custom AI Silicon Strategies

AWS to deploy AI inference chips from Cerebras in its data centers; Anapurna Labs/Amazon in-house AI silicon products

2026 OCP Summit: Full-Stack AI Networking Silicon from Nvidia/Cisco and Broadcom; Ethernet vs InfiniBand Interconnects

Huawei’s Ascend Silicon Roadmap Extends From AI Chips to Cluster-Scale Infrastructure

Cisco’s Silicon One G300 as the dominant AI networking fabric, competing with Broadcom’s Tomahawk 6 series

AWS, Google, Microsoft, Oracle win $9B DoD cloud computing contract (JWCC)

Oracle and Microsoft Enhance Interoperability of their Cloud Platforms (facilitating multi-cloud)

 

Analysis: Why Carriers Are Fighting Mid-Band Satellite D2D Spectrum at the FCC

Posted on by Alan Weissberger

Disclaimer:  Google Gemini was used for research in this article.

………………………………………………………………………………………………………………………………………………..

Executive Summary:

Ahead of the FCC’s October 29th open meeting on Supplemental Coverage from Space (SCS), CTIA filed a joint petition on behalf of AT&T, Verizon, and T-Mobile on October 22nd. The filing urges the regulators to exclude 290 MHz of mid-band spectrum—specifically the 2.5 GHz and 3.45 GHz bands—from the proposed satellite-to-cellular spectrum rules.

The U.S. carriers want to exclude the 2.5 GHz (Band n41) and 3.45 GHz (Band n77/n78) channels. Their argument centers on a hard physical constraint: orbital downlinks disrupt terrestrial Time Division Duplexing (TDD) networks.  From the Backgrounder section of the CTIA Petition:

“This Notice of Proposed Rulemaking (Notice) would recognize American innovation and investment in Direct-to-Device (D2D) technology by proposing to modernize the Commission’s rules for D2D services in licensed spectrum. The Notice would propose to make an additional 482 megahertz of spectrum available for Supplemental Coverage from Space (SCS), eliminate burdensome SCS requirements that constrain innovation, and harmonize the SCS and mobile-satellite service (MSS) rules.

Further, the Notice would consider measures to modernize service rules in MSS-primary bands, including whether to confirm the exclusive nature of MSS bands. The item would seek comment on appropriate D2D performance requirements. These reforms would promote more flexible terrestrial, satellite, and hybrid business arrangements and incentivize high-quality D2D service for the benefit of consumers.’

TDD Timing Failures and Slot Overlap:

Terrestrial 5G in mid-band spectrum relies on strict TDD synchronization. Macro sites use GPS timing to align Downlink (DL) and Uplink (UL) frames. Short Guard Periods separate these slots to account for local signal propagation delays across 1 to 5 kilometers.

Low Earth Orbit (LEO) satellites operating between 300 and 550 kilometers destroy this timing model:

  • Propagation Delays: Orbital distances introduce propagation delays of 2 to 10 milliseconds.

  • Cross-Link Interference: A satellite downlinking in a nominal DL slot arrives at the ground out of phase. The signal bleeds past the Guard Period and hits terrestrial base stations during their scheduled Uplink receive slot.

  • Elevated Noise Floor: Because ground sites cannot adjust their receive windows for orbiting transmitters, these uncoordinated downlinks blind base station uplink receivers.

Handset Desensitization (The Near-Far Problem):

The carriers also modeled the uplink impact of standard mobile handsets attempting to reach space.

  • Maximum Handset Power: To establish a link with a LEO satellite overhead, a smartphone must transmit at its absolute limit—typically 23 dBm to 26 dBm.

  • Base Station Interference: If a user near a terrestrial cell sector edge fires up a satellite link, that high-power uplink transmission floods nearby ground base stations.

  • Sector Outages: T-Mobile’s RF modeling showed that a single high-power Direct-to-Device (D2D) handset can raise the uplink noise floor of a terrestrial cell site 1 to 3 miles away, effectively knocking out terrestrial uplink capacity for that sector.

Chart Generated by Google Gemini:

…………………………………………………………………………………………………………………………………………………………………………………………………..

3GPP Rel-17/18 NTN Specs [1.] vs. Proprietary Mid-Band D2D:

Note 1. ATIS, acting on behalf of 3GPP, submitted 3GPP Release 17/18 candidate specifications for 3GPP 5G-NTN: RIT (Radio Interface Technology) and 3GPP 5G-NTN: SRIT (Set of Radio Interface Technologies) to ITU-R WP 4B under Resolution ITU-R 65. The ITU-R formally adopted these 3GPP NTN specifications into Recommendation ITU-R M.2177-0 “Detailed specifications of the satellite radio interfaces of International Mobile Telecommunications-2020,” published in February 2026.

………………………………………………………………………………………………………………………………………………………………

The core conflict arises because proprietary mid-band SCS bypasses the physical layer synchronization built into official 3GPP Non-Terrestrial Network (NTN) standards.

                                ┌─────────────────────────────────────────────────────────┐
                                │Supplemental Coverage from Space ARCHITECTURES            
                                └────────────────────────────┬────────────────────────────┘
                                                             │
                              ┌──────────────────────────────┴──────────────────────────────┐
                              ▼                                                             ▼
┌───────────────────────────────────────────────────────────┐ ┌───────────────────────────────────────────────────────────┐
│              3GPP Rel-17/18 NTN Standardized             │ │          Proprietary Mid-Band SCS (Rel-14/15 UEs)        │
├───────────────────────────────────────────────────────────┤ ├───────────────────────────────────────────────────────────┤
│ • Spectrum: Dedicated MSS Bands (e.g., S-Band n255/n256)  │ │ • Spectrum: Shared Terrestrial TDD (2.5 & 3.45 GHz)       │
│ • Timing: GNSS UE Self-Pre-compensation (TA Adjustment)   │ │ • Timing: Uncoordinated Orbital Delays (2ms–10ms)         │
│ • Frequency: Autonomous GNSS Doppler Offset Tracking      │ │ • Frequency: Uncorrected Doppler Shift vs. Terrestrial UE │
│ • Network Impact: Zero Cross-Link Interference            │ │ • Network Impact: Severe TDD Slot & Sector Blindness      │
└───────────────────────────────────────────────────────────┘ └───────────────────────────────────────────────────────────┘

  • Timing Advance (TA) Pre-Compensation: In 3GPP Release 17/18 NTN, the device uses its GNSS location and satellite orbital data (ephemeris) to compute physical range. The UE pre-adjusts its transmit timing before sending a frame, ensuring it arrives within the network’s expected window. Proprietary mid-band D2D attempts to serve legacy, non-NTN UEs that lack these satellite timing advance algorithms.

  • Doppler Compensation: LEO satellites move at ~7.5 km/s, causing carrier frequency offsets up to $\pm 40\text{ kHz}$. 3GPP Rel-17/18 requires explicit pre-compensation at the UE or network level to preserve subcarrier spacing. Uncoordinated terrestrial mid-band access lacks this Doppler tracking loop, driving adjacent-channel interference.

  • Spectrum Isolation: 3GPP NTN operates in dedicated Mobile Satellite Service (MSS) bands (like S-band n255/n256). In contrast, using non-standardized satellite overlays in primary terrestrial TDD spectrum forces space and ground networks to share the exact same radio channels.

Aggregate Power Flux Density (PFD) Limits:

Finally, the carriers warned against relying on single-beam interference models. A fleet of thousands of VLEO satellites creates moving, overlapping beam patterns over high-density population centers.

CTIA requested that if the FCC keeps mid-band spectrum in the SCS framework, it must enforce strict aggregate Power Flux Density (PFD) limits. These limits would force satellite operators to throttle downlink power over populated landmasses, protecting ground-based 5G networks.

……………………………………………………………………………………………………………………………………………….

Addendum:

For additional technical context on physical layer synchronization in satellite networks, see this video tutorial on 3GPP 5G NTN Doppler Shift & Timing Advance Explained. This visual guide breaks down how Release 17/18 user equipment calculates propagation delays and pre-compensates for high LEO velocity before transmitting.

……………………………………………………………………………………………………………………………………………….

References:

https://www.fcc.gov/ecfs/document/26110083526/1

https://www.fcc.gov/ecfs/filing/26110083510

https://docs.fcc.gov/public/attachments/DOC-425711A1.pdf

https://communicationsdaily.com/article/2026/10/26/ctia-calls-for-cautious-approach-on-scs-notice-2610230014?BC=bc_6adc8818469bb

https://www.androidheadlines.com/2026/10/att-tmobile-verizon-fcc-satellite-spectrum-spacex.html

https://shattered.io/carriers-oppose-fcc-satellite-spectrum-290-mhz-starlink-2026/

https://www.3gpp.org/dynreport/38821.htm

https://www.3gpp.org/specifications-technologies/releases/release-18

SpaceX Moves to Acquire Nationwide 800 MHz Spectrum for Starlink Mobile D2D Service

Analysis & Opinion: SpaceX to build, lease or buy a hybrid/terrestrial satellite network?

Analysis: Amazon <- Globalstar – a strategic move for D2D and spectrum parity

European network operators in talks to form consortium to bid for EU satellite spectrum and provide D2M service

European Consortium 5G NTN transmission paves the way for standards based direct to device (D2D) connectivity

Tutorial: LEO Satellite Internet connectivity, D2D, and major providers

Ookla on the Global D2D Market

Analyst firms wide forecasts for the LEO satellite direct-to-device (D2D) market

Page 1 of 374
1 2 3 … 374

Archives

Archives

Recent Posts