Author: Alan Weissberger
Non-Terrestrial Networks (NTN) Tutorial: Architecture, Spectrum, and Technical Foundations
by Paresh Panchal, Principal Engineer – Charter Communications
Abstract:
Several Non-Terrestrial Network (NTN) related articles have appeared on the IEEE ComSoc Techblog over the past year. They include: Alan J Weissberger’s market overview (December 2025), the Keysight/Samsung frequency band n252 demonstration (January 2026), the Telecoms.com survey summary (July 2025), and the enterprise IoT hybrid-network article (January 2026). These contributions provide useful market context and early deployment perspective, but they do not fully address the engineering considerations that determine how an NTN system is actually designed, dimensioned, and deployed.
Importantly, they do not examine the 3GPP Release 18 NTN architecture options (A1–A4), which define key implementation choices for operator and satellite network integration. They also do not analyze NTN band planning and its regulatory variability across CEPT and FCC jurisdictions, or the propagation-delay effects that must be accounted for in HARQ timing, scheduling, and other RAN procedures. These issues are central to practical deployment planning and to the selection of an appropriate NTN architecture for a given use case.
This article fills that gap by providing a practitioner-oriented technical reference that complements the existing market-level coverage with engineering detail, e.g. NTN deployment options, spectrum applicability, and protocol-level implications. It is intended to serve as a practical guide for engineers and network planners assessing NTN architecture, spectrum strategy, and protocol behavior in real deployment scenarios.
Read my article at https://wireless-vector.com/ntn-article
……………………………………………………………………………………………………………….
About the Author:
Paresh is a wireless communications professional with deep expertise in RAN systems and architecture, network design, performance engineering, and network analytics. He’s been an active contributor to radio access network innovation with deep expertise in 5G/4G/CBRS RF design and optimization, specializing in cloud-native and O-RAN environments. Proven track record across multi-vendor, multi-country engagements covering greenfield and commercial networks. Core competencies span RF network modeling, performance analytics, and cross-functional program execution. Inventor with 25+ patent applications in radio network technologies.
………………………………………………………………………………………………………………………………………………………………………………………………………………………………..
References:
Non-Terrestrial Networks (NTNs): market, specifications & standards in 3GPP and ITU-R
Keysight Technologies Demonstrates 3GPP Rel-19 NR-NTN Connectivity in Band n252 (using Samsung modem chip set)
Telecoms.com’s survey: 5G NTNs to highlight service reliability and network redundancy
ITU-R recommendation IMT-2020-SAT.SPECS from ITU-R WP 5B to be based on 3GPP 5G NR-NTN and IoT-NTN (from Release 17 & 18)
China ITU filing to put ~200K satellites in low earth orbit while FCC authorizes 7.5K additional Starlink LEO satellites
Samsung announces 5G NTN modem technology for Exynos chip set; Omnispace and Ligado Networks MoU
Analysis and Implications of Deutsche Telekom’s potential full acquisition of T-Mobile
Deutsche Telekom is Europe’s largest telecommunications carrier with 273 million mobile customers in 50 countries. The German telco is reportedly considering a move to take full ownership of T-Mobile US by raising its stake from roughly 54% to 100%, thereby converting the publicly traded U.S. carrier into a wholly owned subsidiary. This potential merger/acquisition would represent a significant strategic and corporate restructuring within the global telecommunications landscape and would mark the consolidation of one of the largest wireless operators under a single parent entity.
During his 12-year tenure as Deutsche Telekom’s CEO Tim Höttges has helped turn T-Mobile from a money-losing underdog to the world’s most valuable telecom brand by market capitalization. Höttges has invested billions in T-Mobile’s U.S. infrastructure to support its 5G fixed-wireless home-internet product. T-Mobile contributes nearly two-thirds of Deutsche Telekom’s revenue.
Analysts say T-Mobile is currently constrained on large deals: Its high leverage makes borrowing expensive, and it can’t issue stock without diluting its German parent’s stake. By combining Deutsche Telekom and T-Mobile into a single $300 billion behemoth, the company would likely be able to raise debt at a lower cost, among other benefits. T-Mobile has been buying fiber-internet operators in the U.S. to compete with AT&T and Verizon and offer bundled wireless and home internet to more customers.
Deutsche Telekom CEO Höttges addressing shareholders at the firm’s annual meeting in April. Oliver Berg/
From a corporate strategy perspective, full ownership would provide Deutsche Telekom with complete control over T-Mobile US’s capital allocation, operational priorities, and long-term network investment strategy. Currently, while Deutsche Telekom exercises effective control through its majority stake, a 100% acquisition would eliminate minority shareholders, simplify governance structures, and allow the parent company to internalize the full economic value generated by T-Mobile US’s operations. This could be particularly relevant as the U.S. market continues to drive substantial cash flow and growth relative to Deutsche Telekom’s European operations.
The transaction would also have implications for corporate structure and financial reporting. Full ownership would enable Deutsche Telekom to restructure T-Mobile US within its corporate hierarchy, potentially integrating it more closely with other group entities or aligning its financial reporting more directly with parent-company objectives. Such consolidation could improve transparency for investors and reduce the complexity associated with managing a majority-owned public subsidiary.
However, executing a deal of this magnitude would present substantial challenges. The transaction would likely require extensive regulatory review in both the United States and Europe, including scrutiny from the Federal Communications Commission, the Department of Justice, and European competition authorities. Valuation would be a critical consideration, given T-Mobile US’s market position as the second-largest wireless carrier in the United States and its ongoing investments in 5G infrastructure, network modernization, and enterprise services. Financing the acquisition would also require careful consideration of debt levels, capital structure, and the impact on Deutsche Telekom’s balance sheet.
From a market perspective, the potential merger could be viewed as a consolidation move that reflects the increasing importance of the U.S. wireless market in global telecommunications strategy. T-Mobile US has emerged as a competitive leader in recent years, with strong performance in 5G deployment, spectrum efficiency, and customer acquisition. Full ownership would enable Deutsche Telekom to align these strengths more closely with its broader global strategy, potentially accelerating technology transfer, network architecture harmonization, and cross-border service integration.
Höttges has also put billions into expanding the German network, where fiber-internet subscribers have nearly tripled since 2023. He champions a cause popular with European regulators: tech sovereignty, or reducing reliance on American and Chinese technology. In February Deutsche Telekom opened Germany’s first AI gigafactory, a massive data center. The gigafactory uses AI GPU chips from Nvidia, which is of course an American company, based in Santa Clara, CA.
The CEO plans to retire at the end of 2028 and wants the right successor to be found first, said people familiar with the matter. He said on the German “OMR” podcast that his successor will need a different skill set. Artificial intelligence (AI) is overhauling the workforce and automating next-generation networks, transforming the industry at an astonishing pace. “Back then, a sober numbers guy was the right choice,” he said. “Today, I believe we need a visionary who understands the future architecture of modern infrastructure.”
In summary, Deutsche Telekom’s reported interest in acquiring full ownership of T-Mobile US represents a significant strategic consideration that would consolidate corporate control, simplify governance, and potentially enhance the parent company’s ability to capture the full financial benefits of its U.S. operations. While the strategic rationale is compelling, the transaction would entail substantial regulatory, financing, and valuation complexities that would need to be carefully addressed before any definitive agreement could be reached.
…………………………………………………………………………………………………………………………………………………………………………………
References:
https://www.telekom.com/en/company/companyprofile/company-profile-625808
https://www.wsj.com/business/telecom/t-mobile-deutsche-telekom-merger-4fdc8eba
https://finance.yahoo.com/markets/stocks/articles/deutsche-telekom-wants-whole-t-143743459.html
https://www.telekom.com/en/company/companyprofile/company-profile-625808
Analysis: OpenAI and Deutsche Telekom launch multi-year AI collaboration
Deutsche Telekom: successful completion of the 6G-TakeOff project with “3D networks”
Deutsche Telekom selects Iridium for NB-IoT direct-to-device (D2D) connectivity
Deutsche Telekom and Google Cloud partner on “RAN Guardian” AI agent
Deutsche Telekom offers 5G mmWave for industrial customers in Germany on 5G SA network
Deutsche Telekom migrates IP-based voice telephony platform to the cloud
Deutsche Telekom Network Day: Fiber, Mobile Network, Open RAN and 5G SA Launch in 2024
SK Telecom and Deutsche Telekom to Jointly Develop Telco-specific Large Language Models (LLMs)
Deutsche Telekom with AWS and VMware demonstrate a global enterprise network for seamless connectivity across geographically distributed data centers
T-Mobile expands FTTH footprint via 50-50 JVs with Oak Hill Capital and Wren House
T-Mobile US announces new broadband wireless and fiber targets, 5G-A with agentic AI and live voice call translation
Highlights of 2025 Broadband Nation Expo: Comcast, T-Mobile keynotes + selected quotes
AT&T deploys nationwide 5G SA while Verizon lags and T-Mobile leads
T-Mobile’s new CEO Srini Gopalan faces fierce competition from AT&T, Verizon and MVNOs
T-Mobile’s growth trajectory increases: 5G FWA, Metronet acquisition and MVNO deals with Charter & Comcast
Evercore: T-Mobile’s fiber business to boost revenue and achieve 40% penetration rate after 2 years
WSJ: T-Mobile hacked by cyber-espionage group linked to Chinese Intelligence agency
T-Mobile posts impressive wireless growth stats in 2Q-2024; fiber optic network acquisition binge to complement its FWA business
Dell’Oro: 2H2026 Data Center Capex to Accelerate due to massive AI Deployments
Dell’Oro Group has raised its worldwide data center capex outlook for 2026 as hyperscale AI deployments accelerated, complemented by continued investments in general-purpose infrastructure and rising component costs.
“Rising memory and storage pricing substantially increased overall server system costs in the quarter and will likely remain a major capex growth factor this year,” said Baron Fung, Senior Research Director at Dell’Oro Group. “At the same time, AI infrastructure deployments continue to accelerate rapidly, while hyperscalers also expanded general-purpose infrastructure to support public cloud growth, agentic AI workloads, and rising AI-related storage requirements.
“Despite exceptionally strong spending growth in 1H2026, capex growth is expected to accelerate further in 2H26, driven by the ramp of NVIDIA Rubin systems and refresh cycles for hyperscaler custom accelerator platforms. Beyond hyperscalers, select enterprise verticals and sovereign cloud providers are increasing AI infrastructure adoption, though growth remains constrained by uncertain returns and infrastructure readiness. While near-term demand remains healthy, some spending may have been pulled forward ahead of expected price increases later this year,” explained Fung.
Additional highlights from the 1Q 2026 Data Center IT Capex Quarterly Report:
- The global data center capex outlook was raised to more than $1 trillion for 2026.
- The Top 4 U.S. cloud providers—Amazon, Google, Meta, and Microsoft—increased data center capex by 78%.
- Dell led server OEM revenue in the quarter, followed by SuperMicro and Lenovo, while white-box vendors serving the hyperscale market accounted for the majority of server revenue. Nearly all server vendors benefited from higher memory-driven system pricing.
Image Credit: Futurum Group
………………………………………………………………………………………………………………………………………………………………
Dell’Oro Group’s Data Center IT Capex Quarterly Report details the data center infrastructure capital expenditures of each of the ten largest Cloud service providers, as well as the Rest-of-Cloud, Telco, and Enterprise customer segments. Allocation of the data center infrastructure capex for general-purpose and accelerated servers, storage systems, and other auxiliary data center equipment is provided. The report also discusses market trends, drivers of the leading Cloud service providers’ capex growth during the quarter, and the outlook for the next year. To purchase this report, please contact us at [email protected].
………………………………………………………………………………………………………………………………………………………………
Perplexity.ai generated- June 2026 forecast:
Approximately 75% of hyperscaler capex in 2026 is for AI infrastructure (~$450 billion).
Key Drivers of the Forecast Increase:
Additional Secondary Factors:
-
NVIDIA Rubin system ramp: Capex growth expected to accelerate further in 2H26 driven by Rubin system ramp
-
Hyperscaler custom accelerator refresh cycles: Refresh cycles for custom accelerator platforms will drive 2H26 growth
-
Enterprise verticals & sovereign cloud adoption: Select enterprise verticals and sovereign cloud providers increasing AI infrastructure adoption
-
Pulled-forward spending: Some spending pulled forward ahead of expected price increases later in 2026
………………………………………………………………………………………………………………………..
References:
https://futurumgroup.com/insights/ai-capex-2026-the-690b-infrastructure-sprint/
https://know.creditsights.com/insights/technology-hyperscaler-capex-2026-estimates/
Nvidia CEO Huang: AI is the largest infrastructure buildout in human history; AI Data Center CAPEX will generate new revenue streams for operators
Will Google Cloud’s AI and data analytics revenue +TPU IP licensing income offset huge AI CAPEX to produce a decent ROI?
Alphabet’s 2026 capex forecast soars; Gemini 3 AI model is a huge success
Hyperscaler capex > $600 bn in 2026 a 36% increase over 2025 while global spending on cloud infrastructure services skyrockets
Big tech spending on AI data centers and infrastructure vs the fiber optic buildout during the dot-com boom (& bust)
Gartner: AI spending >$2 trillion in 2026 driven by hyperscalers data center investments
Will billions of dollars big tech is spending on Gen AI data centers produce a decent ROI?
ABI Research: 6G Radio Installed Base by Region from 2029 to 2034
According to ABI Research, fewer than 0.35 million 6G-supported radios are expected to be deployed worldwide at the end of 2029. That’s understandable, because those “6G radios” will be pre-standard deployments. The IMT 2030 Radio Interface Technology (RIT) standards won’t be completed till the end of 2030!
By 2034, around 7.2 million 6G radios will have been deployed, as per the market research firm’s forecast. The Asia-Pacific region (4 million deployments by 2034) will likely see the first major global deployment of this new generation of radio equipment. North America (1.2 million deployments by 2034) and Europe (940K deployments by 2034) will also see significant deployments once telecom operators complete their own transition from 5G densification to 6G expansion.
6G deployments are forecast to start low, due to continuing ITU-R standards/3GPP specifications work and early network operator caution, with rapid growth after 2031 to 2032 as mobile operators complete their overall first phase of 6G deployments and more operators gain confidence for scaled rollouts.
ABI is treating 6G as a multi-layer infrastructure market: not just spectrum and radios, but also core evolution, advanced antenna systems, and sensing/AI convergence. In practice, that means the report should help answer questions like where to target product planning, how regional adoption may diverge, and which infrastructure subsegments could capture early 6G spend.
ABI’s broader 5G/6G research service also emphasizes spectrum and infrastructure planning, advanced antenna systems, and Open RAN-adjacent market intelligence, which fits the framing of this 6G report as part of a wider network infrastructure portfolio.
References:
https://www.abiresearch.com/news-resources/chart-data/6g-radios-installed-base-forecast
Analysis: Nvidia’s rumored new 6G AI-RAN – likely features/functions and industry impact
Comparing AI Native mode in 6G (IMT 2030) vs AI Overlay/Add-On status in 5G (IMT 2020)
Roles of 3GPP and ITU-R WP 5D in the IMT 2030/6G standards process
NVIDIA and global telecom leaders to build 6G on open and secure AI-native platforms + Linux Foundation launches OCUDU
AI wireless and fiber optic network technologies; IMT 2030 “native AI” concept
Analysis: Nvidia’s rumored new 6G AI-RAN – likely features/functions and industry impact
Executive Summary:
According to Light Reading, Nvidia is working on a GPU combo chip that would sit directly in the 6G radio unit [1.], extending its AI-RAN push from baseband/server into the radio itself. It’s reported to be a more hardware-integrated, sub-100W embedded design rather than just GPU acceleration in centralized RAN compute.
Note 1. 6G/IMT 2030 Radio Interface Technologies (RITs) have yet to be defined, let alone specified by 3GPP or ITU-R WP5D. They won’t be solidified until the end of 2030 so any specific silicon design won’t be completed until then or 2031!
……………………………………………………………………………………………………………………………………………………….
Light Reading’s headline frames it as a “radical new AI-RAN plan and they wrote that “the move was confirmed by knowledgeable sources, with Nvidia saying GPUs in more advanced radios will become “essential” in future. It marks a dramatic new development in the GPU giant’s “AI-RAN” strategy.”
If accurate, this would be a notable shift for Nvidia, because it would let them influence the whole RAN stack, not just centralized compute. That could matter for performance, power efficiency, and AI-native functions such as sensing, spectrum optimization, and real-time signal processing. Nvidia’s broader 6G messaging already emphasizes AI-native wireless, integrated sensing and communications, and spectrum agility as core themes.
The unconfirmed report fits Nvidia’s existing telecom roadmap rather than appearing out of nowhere. Nvidia has already announced an AI-native wireless stack for 6G with partners including Cisco, MITRE, Booz Allen, ODC, and T-Mobile, and it has promoted AI-RAN as a way to combine connectivity, computing, and sensing on one platform. It also aligns with the company’s recent partnership with Nokia, where Nvidia introduced the ARC-Pro 6G-ready accelerated computing platform and described it as a software-upgradable path from 5G-Advanced to 6G. That makes the rumored radio-chip move look like a vertical extension of the same strategy.
For wireless network operators, a radio-unit chip from Nvidia would be significant only if it improves cost, power, or flexibility versus incumbent RU silicon. The practical test will be whether it can deliver enough RF, baseband, and AI function integration to justify another architecture layer at the edge. It would also intensify competition in the radio-access supply chain and reinforce the trend toward AI-native, software-defined RANs. It also suggests Nvidia wants to shape not only the compute layer but the physical radio layer of 6G networks.
Possible AI Silicon Features and Functions:
Nvidia would most likely add AI-for-RAN features into radio silicon first, because those map directly to signal processing and link adaptation rather than to generic “AI at the edge.” Nvidia’s own AI-RAN materials emphasize embedding AI/ML into the radio signal-processing layer to improve spectral efficiency, coverage, capacity, and performance. Here are a few likely AI features/functions for the rumored 6G AI Nvidia super chip:
-
Neural channel estimation and equalization, to infer cleaner channel state from noisy RF observations and improve link reliability. Nvidia’s open-source Aerial release specifically calls out advanced neural models for channel estimation.
-
Real-time beam management, including beam selection, beam tracking, and beam refinement for massive MIMO and mmWave/upper-midband deployments. These are natural AI-RAN use cases because they depend on fast adaptation to changing propagation conditions.
-
Spectrum agility and interference mitigation, such as identifying jammed or congested resource blocks and dynamically avoiding them. NVIDIA and partners have already described spectrum agility applications that freeze only affected frequencies while keeping the rest of the system online.
-
Dynamic resource scheduling, using learned traffic and channel patterns to allocate PRBs, power, and compute more efficiently in real time. Nvidia describes AI-RAN as improving spectral efficiency and dynamic traffic handling through AI.
-
Integrated sensing and communications support, where the radio helps detect objects, motion, or environmental context in parallel with communication. Nvidia has already highlighted ISAC-style applications with camera/RF fusion and object tracking.
-
Edge inference hooks, letting the RU expose real-time PHY data to AI applications or a dApp-style framework. Nvidia’s open-source Aerial stack says third-party apps can access physical-layer data through secure APIs and modify RAN behavior in real time.
-
Self-optimization and closed-loop control, where the radio silicon learns local conditions and continuously retunes thresholds, coding, MCS selection, and precoding policies. That fits Nvidia’s broader framing of AI-native networks as software-defined and continuously adaptable.
The most plausible first wave is not a fully autonomous “AI radio,” but a hybrid RU chip that accelerates selected PHY functions and exposes telemetry/data paths to the rest of the AI-RAN stack. Nvidia’s current messaging emphasizes software-defined infrastructure, deterministic performance, and layered AI-RAN capabilities rather than replacing the entire RAN with a black-box model.
The real differentiator would be whether Nvidia can combine RF signal processing with its GPU/CUDA ecosystem, so the same platform handles channel learning, inference, and orchestration across RU/DU/CU tiers. That would let operators optimize for spectral efficiency and OPEX while still keeping a software-upgrade path to 6G. Radio electronics is constrained by power, latency, determinism, and certification, so Nvidia would need to prove these AI features help without destabilizing PHY timing. That is why the likely starting point is assistive AI inside the signal chain, not a fully learned end-to-end radio.
Image Credit: Nvidia
…………………………………………………………………………………………………………………………………………………………………………………………………………..
Competitive Analysis:
Nvidia’s reported move into a 6G radio-unit chip is most threatening to Marvell and Qualcomm at the silicon layer, while it is more of a strategic architecture challenge to Nokia and Ericsson at the system level. The immediate effect is less about a single chip and more about Nvidia trying to pull compute, connectivity, and AI deeper into the RAN value chain
Qualcomm is the closest direct competitor if Nvidia is trying to put silicon into the radio or near-radio layer. Qualcomm already has a Layer 1 strategy that combines silicon and software in SmartNIC/server-adjacent form factors, so Nvidia would be moving into a space where Qualcomm has both telecom credibility and established IP.
The risk for Qualcomm is that Nvidia can use its AI brand, CUDA ecosystem, and hyperscale relationships to redefine what “performance” means in RAN silicon, especially if AI-native functions become a buying criterion. The counterpoint is that Qualcomm still has a strong edge in wireless-specific silicon integration and standards heritage, which matters if the 6G radio path remains RF- and modem-centric.
Nokia looks less exposed in the short term because it is already partnering with Nvidia rather than treating it as a pure adversary. Nvidia and Nokia have publicly framed their relationship as an AI-native 5G-Advanced/6G platform effort, and Nokia says it will add NVIDIA-powered commercial AI-RAN products to its RAN portfolio.
Nonetheless, a Nvidia radio-chip push could still compress Nokia’s differentiation over time if more of the RAN stack becomes software-defined and GPU-centric. The strategic question is whether Nokia remains the integrator and operator-facing systems vendor, or whether Nvidia gradually becomes the architectural center of gravity.
Ericsson is the most structurally interesting case because it sits at the high end of global RAN share and has been more cautious about Nvidia as a Layer 1 option. Light Reading notes Ericsson is currently dismissive of Nvidia as a Layer 1 choice, even while the broader ecosystem explores AI-RAN collaboration.
For Ericsson, the threat is not immediate revenue loss from a single chip; it is erosion of the traditional assumption that RAN leadership comes from proprietary radio and baseband stacks. If Nvidia can make AI-native RAN a default design paradigm, Ericsson may be forced to defend its software and systems value rather than simply its box-selling model.
Samsung Electronics contacted Light Reading after their story was published to point out that it also works with AMD as a chip partner. “Samsung supports full Layer 1 (L1) processing using Intel’s telco CPUs (e.g., Xeon 6 Granite Rapids) and lookaside accelerator approach and in addition has successfully demonstrated full L1 processing on AMD’s CPUs without relying on dedicated L1 accelerators,” a Samsung spokesperson said via email.
Marvell is the most exposed chip supplier in this story because its telecom position is more concentrated in custom Layer 1 silicon. Light Reading specifically points out that Marvell is a critical supplier to Nokia in Layer 1, which makes a Nvidia radio-chip effort a direct substitution threat in portions of the stack.
If Nvidia succeeds, Marvell faces a two-sided squeeze: loss of design wins in telecom silicon and a narrative shift toward AI-native programmable platforms that favor Nvidia’s broader ecosystem. Marvell’s defense is that telecom operators still care about power, latency, and deterministic functionality, areas where custom silicon can remain more efficient than a generalized AI-compute approach.
…………………………………………………………………………………………………………………………………………………………………………
Summary Table:
| Company | Impact level | Why |
|---|---|---|
| Qualcomm | High | Direct silicon adjacency and overlapping Layer 1 ambitions. |
| Marvell | High | Telecom custom-silicon exposure, especially Layer 1. |
| Ericsson | Medium | Strategic and architectural threat more than immediate chip displacement. |
| Nokia | Medium to low near term | Partnered with Nvidia, so risk is more about future dependence and stack control. |
Source: Perplexity.ai
…………………………………………………………………………………………………………………………………………………………………………
Conclusions:
It’s unknown whether Nvidia’s rumored radio chip becomes a product, a reference design, or just an extension of its AI-RAN platform. If it ships, watch for operator trials, power-envelope disclosures, and whether it targets RU integration, DU acceleration, or a hybrid AI-RAN endpoint. If it stays at the partnership/reference-design level, the market impact will be more narrative than revenue-relevant.
Another unanswered question is whether Nokia and Ericsson keep treating Nvidia as a collaborator while preserving their own Physical layer control, or whether they start to see Nvidia as a platform owner in the making. That boundary will determine whether this is a tactical ecosystem play or the beginning of a deeper industry reset.
…………………………………………………………………………………………………………………………………………………………………………
References:
https://www.lightreading.com/6g/nvidia-has-a-radical-new-ai-ran-plan-a-6g-radio-unit-chip
https://www.lightreading.com/6g/analyst-insight-6g-coming-into-focus
https://www.nvidia.com/en-us/industries/telecommunications/ai-ran/
RAN Silicon Rethink- Part II; vRAN and General-Purpose Compute
Orange, Nokia, Nvidia, and Intel debate: ASICs vs. GPUs vs. General-Purpose CPUs for RAN Baseband Processing
RAN silicon rethink – from purpose built products & ASICs to general purpose processors or GPUs for vRAN & AI RAN
Dell’Oro: Analysis of the Nokia-NVIDIA-partnership on AI RAN
Nvidia pays $1 billion for a stake in Nokia to collaborate on AI networking solutions
Inside Nokia’s new AI Networking Innovation Lab
Analysis: Nvidia’s $2 billion investment in Marvell; NVLink Fusion ecosystem & RAN vendor silicon strategy
Marvell shrinking share of the RAN custom silicon market & acquisition of XConn Technologies for AI data center connectivity
Oriole Networks photonic networking platform to be integrated with AMD GPUs/CPUs for next-gen AI data center fabrics
London, England based Oriole Networks today announced continued progress in its collaboration with AMD in support of the UK’s Advanced Research & Invention Agency (ARIA) Scaling Inference Lab. The initiative integrates Oriole’s photonic interconnect architecture with AMD Instinct GPUs and AMD EPYC CPUs to evaluate next-generation data center fabrics capable of addressing the performance, latency, and energy constraints inherent in large-scale AI workloads.
The multi-year collaboration is advancing toward deployment of what is positioned as the first production-scale, all-photonic AI network fabric. The system is designed to deliver ultra-low latency and deterministic transport characteristics at the system level, leveraging optical circuit switching to optimize east-west traffic flows across accelerator clusters. The primary objective is to demonstrate how optical interconnect technologies can support large-scale inference and distributed AI processing under stringent performance and energy constraints.
Oriole’s PRISM photonic networking platform [2.] replaces conventional electronic switching in the network core with nanosecond-scale optical circuit switching. In contrast to packet-switched electronic fabrics, this approach is intended to reduce forwarding overhead, lower core power consumption, and improve end-to-end transport efficiency for accelerator-dense workloads. AMD is contributing compute hardware and technical collaboration to support modeling and execution of large-scale network workloads relevant to frontier AI systems. However, PRISM is not built for any single chip vendor. It works across any accelerator platform, giving the wider industry a path to frontier-scale system-wide performance without the need for proprietary stacks.
Note 1. Oriole Networks is a photonic networking company, developing disruptive technologies for AI/ML and HPC networking that will revolutionize data centers. These technologies address AI’s biggest challenges – speed, latency, and sustainability. Our holistic approach replaces energy-hungry electrical switching with photonic switching. By using only light to move data in the network, our solution will increase the efficiency of LLM training and inference to unprecedented levels while dramatically reducing the energy consumption of data centers, currently putting a huge strain on energy grids. We can offer faster, more efficient, and more sustainable AI without sacrificing the planet.
Note 2. Oriole’s PRISM is a fully photonic network system designed to provide port-level, all-to-all connectivity, eliminating the need for electrical switches and dramatically reducing the number of optical transceivers needed in the network. This evolution greatly reduces power consumption and latency, increases bandwidth, and strengthens network resilience by eliminating single points of failure.
Image Credit: Oriole Networks
…………………………………………………………………………………………………………………………………………………………………………………………………………………………………………..
The deployment also represents the first commercial implementation of Oriole’s technology following an R&D-to-production transition completed in approximately three years. The company states that its xPU-agnostic architecture is intended to support heterogeneous accelerator environments and broader industry rollout beginning in 2027.
Photonic networking architecture:
PRISM is designed to route data optically rather than electrically, using photonic circuit paths in place of conventional electronic switching elements. As AI training and inference workloads scale, data center interconnect requirements increasingly exceed the efficiency limits of traditional switch-based architectures, particularly in terms of power dissipation, thermal load, and communication latency.
By eliminating electronic switching in the fabric core, the PRISM architecture seeks to reduce core network power consumption and limit buffering- and queuing-related delay. The use of optical circuit switching is consistent with ongoing industry interest in photonic interconnects, co-packaged optics, and optical disaggregation as potential enablers of high-density AI clusters.
The company reports that the architecture can substantially reduce GPU idle time and improve system-level utilization by shortening data movement paths between compute nodes. It also indicates potential reductions in cooling demand and associated water usage due to lower network power dissipation.
Quotes:
James Regan, CEO of Oriole, said: “A year ago, we were proving the physics; today, we’re proving the business. Our collaboration with AMD has moved from concept to deployment to a system an order of magnitude larger, and the data proves this is already driving performance increases at pace. This is what it looks like when photonic networking stops being a research curiosity and starts being the foundation of how serious AI infrastructure gets built. There’s a big problem now with electrical switches, which are basically bottlenecking AI traffic, and it’s going to get worse. What we do is we replace all the electrical switches.”
“AMD is excited to collaborate with Oriole on the ARIA Scaling Inference Lab cluster,” said Madhu Rangarajan, corporate vice president, Compute and Enterprise AI business, AMD. “Oriole’s AI backend networking with nanosecond optical circuit switching represents a fundamentally different way to connect accelerators at scale. We are helping to validate how photonic fabrics can work alongside AMD compute to deliver the low-latency, high-bandwidth connectivity that AI Inference workloads demand.”
“Meeting the demands for modern AI requires rapidly identifying ways to improve the performance and cost-efficiency of large-scale AI clusters. ARIA is thrilled to collaborate with Oriole and AMD to demonstrate the benefits of this new technology and it’s exactly the type of collaboration, between innovative startups and industry leaders, that the Scaling Inference Lab was designed to foster,” said Suraj Bramhavar, Program Director at ARIA
Standards and interoperability context:
From a standards perspective, photonic AI fabrics remain an active area of industry development rather than a fully mature architectural class. Relevant technical domains include IEEE 802.3 optical Ethernet interfaces, ITU-T optical transport frameworks such as G.694 and G.709, and ecosystem work in optical interconnect and co-packaged optics initiatives.
A vendor-neutral, accelerator-agnostic photonic fabric may be of interest to standards and industry groups evaluating future data center interconnect models for AI and high-performance computing. The Oriole–AMD collaboration therefore provides an early reference point for assessing the operational characteristics, integration constraints, and interoperability implications of optical circuit-switched AI infrastructure.
……………………………………………………………………………………………………………………………………………………………………………………………………………………….
References:
Oriole to Deploy World’s First AI System with Pure Photonic Network to Supercharge Data Centers
https://www.fierce-network.com/cloud/oriole-networks-pushes-pure-photonic-networking-ai-data-centers
NTT’s IOWN is (finally) evolving to an All Photonics Network (APN); Physics based AI for enterprise OT
Goldman Sachs report: Optical Networking is the next mega trend in AI infrastructure
Hyperscaler design of networking equipment with ODM partners
Technavio: Silicon Photonics market estimated to grow at ~25% CAGR from 2024-2028
Amazon and Corning in Multi-Billion-Dollar Fiber Infrastructure Deal in North Carolina
Introduction:
The surge in optical fiber demand is intensifying as hyperscale cloud providers accelerate infrastructure buildouts to support AI-driven workloads and high-density data center interconnect (DCI). Corning [1.] today announced a multi‑billion‑dollar investment from Amazon to expand fiber manufacturing capacity in North Carolina—incremental to its previously announced $10 billion regional cloud infrastructure expansion—reflects a broader structural shift in how optical supply chains are being secured and scaled.
Note 1. Corning’s fiber-optic infrastructure uses highly pure strands of optical glass thinner than a human hair to transmit massive amounts of data as pulses of light. These networks serve as the backbone for modern communications, connecting everything from rural broadband rollouts to hyperscale data centers driving generative AI. In hyperscale cloud and AI data centers, Corning provides high-density optical hardware and cables, such as their GlassWorks AI™ solutions. These large setups feature massive fiber-optic trunk cables containing hundreds to thousands of individual fibers bundled together to link powerful processors and servers. For outdoor networks running underground or on utility poles, you will see ruggedized cables protected by thick jackets and aramid yarn. These cables are designed to withstand weather, crushing, and extreme temperatures.
Corning’s structured cable solutions for internal data center connectivity. Image Credit: Corning
…………………………………………………………………………………………………………………………………………………………………………………….
This trend is not isolated. Hyperscalers including Meta, Microsoft, and wireline network operator Lumen are proactively entering long-term supply and co-investment agreements with fiber and cable manufacturers, effectively reshaping the upstream optical ecosystem.
Recent Fiber Supply Agreements with Corning:
-
May 2026: NVIDIA committed $500 million to Corning to support construction of three new optical manufacturing facilities in North Carolina and Texas. This investment is expected to increase Corning’s U.S.-based optical connectivity manufacturing capacity by approximately 10× and expand domestic fiber production by over 50%, targeting AI cluster interconnect requirements characterized by high fiber count and low-latency links aligned with IEEE 802.3 Ethernet and emerging co-packaged optics ecosystems.
-
January 2026: Meta finalized a $6 billion agreement with Corning to secure fiber supply for large-scale data center fabrics. These fabrics increasingly rely on high-fiber-density architectures consistent with leaf-spine topologies and standards such as IEEE 802.3bs/ck (400G/800G Ethernet), as well as parallel single-mode fiber (PSM) and wavelength-division multiplexing (WDM) approaches defined in ITU-T G.694.x.
-
September 2025: Microsoft entered a manufacturing agreement with Corning and Heraeus focused on hollow-core fiber (HCF), a technology aligned with ITU-T G.650 characterization frameworks. HCF offers lower latency (reduced group index) and improved performance for latency-sensitive AI workloads and inter-data center transport.
-
August 2024: Corning and Lumen established a supply agreement for next-generation fiber optic cable to support AI-driven traffic growth. This aligns with ITU-T G.652.D and G.657 fiber standards for bend-insensitive and high-capacity terrestrial deployments, as well as evolving requirements for high-count ribbon fiber cables in dense metro and campus environments.
Structural Implications for the Optical Supply Chain:
Hyperscalers are transitioning from passive consumers of optical components to active participants in manufacturing scale-up, including:
-
Anchor tenancy models: As seen with Meta’s backing of Corning’s North Carolina facility, hyperscalers are underwriting capacity expansion, effectively securing preferential access to supply.
-
Vertical influence: Direct investments and long-term offtake agreements allow hyperscalers to influence fiber specifications, manufacturing roadmaps, and deployment architectures (e.g., optimized fiber types for short-reach vs. long-haul DCI).
-
Workforce development: Amazon and Corning’s collaboration with Catawba Valley Community College to expand fiber technician training reflects a strategic effort to address labor constraints in optical manufacturing and deployment, reinforcing domestic supply chain resilience.
Implications for Telecom Operators:
These developments introduce non-trivial risks and strategic considerations for telecom operators:
-
Supply prioritization: Hyperscaler-backed agreements may shift allocation dynamics, potentially constraining availability for traditional telecom buyers during periods of tight supply.
-
Pricing pressure: Long-term, high-volume contracts could influence pricing benchmarks, potentially disadvantaging operators without comparable scale or capital flexibility.
-
BEAD timing mismatch: U.S. operators anticipating fiber expansion funded by BEAD (Broadband Equity, Access, and Deployment) may face supply bottlenecks if hyperscaler demand absorbs near-term manufacturing output.
-
Architectural divergence: Hyperscaler-driven requirements—optimized for short-reach, ultra-high-capacity intra-data-center and DCI links—may skew innovation toward their use cases, potentially misaligning with traditional access network needs governed by ITU-T G.984 (GPON), G.9807 (XGS-PON), and emerging 25G/50G PON standards.
A useful analogy is the semiconductor industry, where hyperscaler influence has already reshaped foundry capacity allocation and advanced node prioritization. A similar dynamic is now emerging in optical fiber and connectivity, with hyperscalers effectively acting as quasi-industrial planners for next-generation optical infrastructure.
Quotes:
“Amazon’s investments in North Carolina have created more than 26,000 jobs across the state. This multibillion-dollar agreement with Corning continues that commitment, channeling investment into American manufacturing and creating 1,000 new jobs at their facilities near our data centers,” said Matt Garman, CEO of AWS. “We’re also partnering to train North Carolinians for highly skilled roles in fiber optics and fusion splicing. These long-term investments create long-term careers and real opportunity in the communities where we operate.”
“This agreement with Amazon represents a significant milestone for Corning and for American manufacturing,” said Wendell Weeks, chairman, CEO, and president of Corning. “For 175 years, Corning has pioneered the technologies that connect people and transform industries. Amazon’s investment will help us expand production, create 1,000 new advanced manufacturing jobs at our facilities, and lead the way toward building a resilient U.S. manufacturing base.”
Clearfield CEO Cheri Beranek told Fierce Network at Fiber Connect that supply chain issues are re-emerging, particularly around high-count fiber. “There’s absolutely a shortage of ribbon fiber,” she said, referring to a conversation with Hawaii Telecom, a Clearfield customer. “The high count for the ribbon fiber … everything over 432 is tough to get,” she said. “The fiber companies want to tell you that there’s enough American‑made fiber… but there can’t be.”
“In talking to fiber optic suppliers, they all say one thing, ‘It’s nice to finally be the cool kid on the block.’ Hyperscalers are finally realizing that they not only need compute, storage, chips, power, water and real estate, they also need fiber optic connectivity,” said Fierce Network’s Chief Analyst Linda Hardesty.
The net effect is a tightening coupling between AI infrastructure demand and optical supply chain strategy—one that telecom operators will need to actively manage through procurement strategy, vendor diversification, and potentially deeper participation in supply-side partnerships.
End Note:
Amazon’s long-term commitment to North Carolina goes beyond direct investments and jobs created in the state. Through workforce development, Career Choice, and upskilling programs, Amazon has already provided practical training for nearly 7,000 people in North Carolina, helping to open new pathways for higher-paying jobs and fulfilling careers.
In the last decade, Amazon has contributed more than $72 million to charities and organizations supporting local needs across North Carolina, with $10 million provided in 2025 alone to 26 local community partners. This includes contributions like $1.5 million to enhance public safety services for southeastern Hamlet and surrounding Richmond County communities by funding a new fire substation that is expected to lower emergency response times and homeowner insurance premiums.
References:
https://www.corning.com/data-center/au/en/home/applications/enterprise-private-data-center.html
https://www.aboutamazon.com/news/company-news/amazon-corning-fiber-optics-1000-jobs-north-carolina
Fiber Optic Boost: Corning and Meta in multiyear $6 billion deal to accelerate U.S data center buildout
Corning to Build New Fiber Optic Plant in Phoenix, AZ for AT&T Fiber Network Expansion
Calix and Corning Weigh In: When Will Broadband Wireline Spending Increase?
Verizon-Corning $1.05B fiber deal part of larger build-out or buy program
Hyperscalers Dominance of Subsea Cable Capacity to Increase in the AI Era
Hyperscalers (AWS, Google, Microsoft, Meta/FB) now dominate global subsea cable capacity. Their share of total international bandwidth has surged from negligible levels in 2010 to approximately 75% today. According to data from TeleGeography, hyperscalers are participating in over two-thirds of all planned submarine cable deployments, with Google alone anchoring eight new systems in the Asia-Pacific (APAC) region. Despite this shift, traditional telecommunications operators remain critical to the subsea ecosystem.
Tier-1 telecom carriers provide the deep terrestrial reach and last-mile connectivity that both regional service providers and large content providers require to access edge markets. However, those network operators must increasingly architect their Wide Area Network (WAN) and long-haul transport infrastructure to integrate seamlessly with these massive hyperscale topologies.
Brian Washburn, Chief Analyst at Omdia’s Telco B2B Solutions Intelligence Service, notes that carriers face intensifying pressure to align their infrastructure with hyperscaler technical requirements. To achieve complete architectural control and establish fully isolated private networks, hyperscalers frequently seek to deploy proprietary optical transport equipment directly within carrier landing stations and co-location facilities. This shift toward self-contained infrastructure creates visibility challenges for the industry. Washburn noted Google’s extensive transpacific cable network as a primary example. Because this hyperscaler traffic is routed over fully private, dark fiber subsea segments, it remains entirely invisible to carrier networks and traditional traffic-modeling metrics, rendering these massive data volumes completely opaque.
TeleGeography’s interactive submarine cable map shows the majority of active and planned international submarine cable systems and their landing stations. Selecting a cable route on the map provides access to data about the cable, including the cable’s name, ready-for-service (RFS) date, length, owners, website, and landing points. Selecting a landing point provides a list of all submarine cables landing at that station.
From a macro perspective, the deployment of next-generation physical infrastructure is increasingly tied to the rollout of raw, rack-scale data center capacity to support emerging AI workloads. Matt Walker, Chief Analyst at MTN Consulting, indicates that while Tier-1 US operators anticipate near-term traffic growth from centralized AI training models, they maintain a cautious, wait-and-see outlook regarding long-term network demand and the broader monetization of distributed inference at the edge. “With agentic, the potential for rapid growth in unexpected parts of the network is real, and it’s not clear how to plan for this,” he said. Operators are worried they will be stuck with the network costs to support “these pricey new AI-enabled services,” he also noted. Telco’s lack of visibility becomes a problem here. Walker stated in his research report: “The industry is flying partially blind. No comprehensive public study of AI traffic volumes, patterns, or growth exists. Nokia, Ericsson, and a handful of others have made partial contributions, but hyperscalers don’t share traffic data. For an industry spending over $600 billion in capex this year, this is a significant planning liability.”
MTN also revealed that telco capex remained subdued in 4Q2025, rising just 0.2% YoY to $86.6B as operators prioritized capital discipline, AI-enabled efficiency, and monetization of prior 5G investments. On an annualized basis, capex declined 0.9% to $295.7B, remaining below the $300B threshold for a second consecutive year. The strongest annualized capex growth rates were recorded by Swisscom (40.7%), Etisalat (40.5%), Airtel (24.4%), SoftBank (10.5%), and Deutsche Telekom (10.3%). The steepest capex declines came from China Telecom (-13.6%), Telefonica (-12.3%), China Unicom (-11.5%), Reliance Jio (-10.8%), and China Mobile (-8.1%).
Regionally, the Americas strengthened its lead in 4Q2025, accounting for 36.5% of global telecom revenues and 36.3% of capex, supported by resilient performance from T-Mobile US, AT&T, and Verizon. Asia’s revenue share moderated to 35.6% and capex share fell to 32.4%. This is notable given that Chinese telcos have been ramping AI and data center spending, while overall capex continues to decline as cuts to radio/hardware spending post-5G more than offset these gains.
References:
https://www.lightreading.com/ai-machine-learning/ai-is-going-to-transform-our-networks
https://www.submarinecablemap.com/
Cisco report: Agentic AI to reshape WAN traffic, AI inference will be ~25% of total traffic by 2035
Fiber Optic Networks & Subsea Cable Systems as the foundation for AI and Cloud services
Subsea cable systems: the new high-capacity, high-resilience backbone of the AI-driven global network
FCC updates subsea cable regulations; repeals 98 “outdated” broadcast rules and regulations
India’s Data Transmission Capacity to Quadruple in 2025 via New Submarine Cables
TechCrunch: Meta to build $10 billion Subsea Cable to manage its global data traffic
Google’s Bosun subsea cable to link Darwin, Australia to Christmas Island in the Indian Ocean
China seeks to control Asian subsea cable systems; SJC2 delayed, Apricot and Echo avoid South China Sea
“SMART” undersea cable to connect New Caledonia and Vanuatu in the southwest Pacific Ocean
Telstra International partners with: Trans Pacific Networks to build Echo cable; Google and APTelecom for central Pacific Connect cables
Orange Deploys Infinera’s GX Series to Power AMITIE Subsea Cable
Intentional or Accident: Russian fiber optic cable cut (1 of 3) by Chinese container ship under Baltic Sea
SK Telecom applies digital twins to SK Hynix semiconductor fabs using NVIDIA Omniverse libraries
SK Telecom (SKT) announced today that it has applied digital twins to SK Hynix semiconductor fabs [1.] using NVIDIA Omniverse libraries, optimizing the technology for complex, large-scale manufacturing environments. Digital twins recreate actual factories and equipment in virtual environments, enabling companies to simulate and verify the impact of process changes and equipment layout adjustments in advance. By enabling simulation of a wide range of scenarios in virtual environments, digital twins are gaining attention as a core physical AI technology that reduces trial and error while supporting data-driven decision-making. Last year, SKT completed a proof of concept (PoC) for applying digital twin technology to SK Hynix semiconductor fab. The company plans to proceed with commercialization in phases, aligning with SK Hynix’s roadmap to establish an “Autonomous Fab” by 2030.
Note 1. SK Hynix operates major semiconductor fabrication and packaging sites across South Korea and China, with new multibillion-dollar facilities under development in South Korea and the United States. While its core, multi-billion-dollar fabs are dedicated entirely to semiconductor memory production (DRAM, HBM, and NAND Flash), the company also operates a dedicated, separate pure-play foundry business that manufactures non-memory logic chips for external contract clients. : The main facilities in Icheon, Cheongju, and Yongin are specialized strictly for SK Hynix’s high-volume memory products like High-Bandwidth Memory (HBM), standard DRAM, and NAND flash. These massive facilities do not accept contract manufacturing orders for logic chips from external companies.
The Contract Foundry Business (External Clients): SK Hynix operates a wholly-owned subsidiary called SK Hynix System IC. This arm acts as a dedicated foundry for fabless semiconductor clients.
…………………………………………………………………………………………………………………………………………………………………………………………………………………………………..
Using the NVIDIA Agent Toolkit, SKT has also developed “Agentic Digital Twin Modeling” technology, which automates and intelligently processes diverse data—such as equipment and spatial structures at manufacturing sites—for digital twin environments. This technology enhances the efficiency of data conversion, scene optimization, and performance improvement tasks that arise during the development and operation of digital twins in manufacturing environments.
A virtual factory implementation using SK Telecom’s digital twin platform. /Courtesy of SK Telecom
………………………………………………………………………………………………………………………………………………………………………………………………………………………………………….
SKT is enhancing its platform by integrating NVIDIA Omniverse libraries to improve the loading speed of large-scale Open USD-based 3D scenes, execution performance, and GPU and memory usage efficiency. Through this, the company plans to implement a stable and scalable digital twin environment even in complex manufacturing environments with massive data volumes, such as semiconductor fabs.
“Semiconductor fabs are among the most challenging manufacturing environments, combining massive amounts of 3D data, complex equipment structures, and the need for high-level optimization,” said Mike Geyer, head of industrial digital twins at NVIDIA. “SKT has demonstrated a high level of technical capability in applying and validating NVIDIA Omniverse libraries, as well as the NVIDIA Agent Toolkit in real-world industrial settings within this environment.”
“Through our collaboration with NVIDIA, we have validated that manufacturing digital twins can evolve beyond simple 3D visualization into a physical AI platform capable of understanding and optimizing large-scale 3D manufacturing data,” said Cho Ik-hwan, Head of Physical AI at SKT. “Going forward, SKT will continue to expand its role as a physical AI technology partner with NVIDIA across various industrial sectors, including semiconductors.”
As a network provider equipped with end-to-end AI solutions — from AI infrastructure and models to services — SK Telecom plans to expand and strengthen its business targeting the enterprise and public sectors.
References:
https://www.thelec.net/news/articleView.html?idxno=10930
Inside Amazon’s new data center network architecture: quasi random network topology and passive optical devices
Amazon Web Services (AWS) claims it recently achieved a major breakthrough in Data Center Network (DCN) architecture and has been quietly deploying the new technology in its data centers since late last year. Amazon detailed its new networking design in a paper published May 21st titled “RNG: Flat Data Center Networks at Scale.” RNG, or “resilient network graphs,” is built around a quasi-random topology and new passive optical hardware. It’s a “quasi-random” design that combines elements of traditional, structured data networks with the performance advantages of more random architectures.
The goal is to move off conventional hierarchical “fat-tree” designs toward a flatter, more mesh-like fabric that uses far fewer routers and switches, offers more parallel paths, and therefore delivers higher effective throughput at lower power and capex.
“By essentially flattening the network, we eliminated the bottlenecks that come with traditional networking designs,” Matt Rehder, vice president of AWS Network Engineering, said in an exclusive interview with WIRED. “We think we’re the only ones who have done this at scale. RNG is a great fit for our core demands, but AI training data patterns are far more coordinated and centrally orchestrated, so they don’t approximate a random graph.”
The fact that Amazon is using this in the real world is “remarkable,” said Brighten Godfrey, a computer science professor at the University of Illinois Urbana-Champaign and an expert in networking, who was not involved in Amazon’s research. Godfrey coauthored a seminal 2012 paper on random network graphs, which he says are a “mind-bending problem to solve, in general.”
Classic cloud DCNs use structured topologies (Clos/fat-tree) where paths are highly regular and layered (Top of Rack (ToR)–aggregation–core). By contrast, random-graph theory says the most efficient routing networks are flat random graphs: each node connects to a small random subset of others, creating many short, diverse paths and graceful degradation under failures. The problem has always been practical: random cabling at scale is unmanageable, and routing across a huge random graph is nontrivial.
AWS’s “quasi-random” design essentially mixes determinism with randomness: key structural elements are fixed to keep the cabling and deployment manageable, while enough randomness is retained in the interconnect pattern to get the performance and resilience benefits of random graphs. The physical enabler is a new passive optical device called a ShuffleBox that standardizes how switches connect and internally permutes links so that, when many ShuffleBoxes are wired together, the resulting global topology is quasi-random without having to hand-design every link.
Image Credit: Amazon
………………………………………………………………………………………………………………………….
Key architectural pieces and claimed gains:
AWS reports that RNG-based fabrics now serve as the default network architecture for most new AWS data centers, after initial deployments beginning in 2024. The company claims the design:
-
Uses roughly 69% fewer routers/switches than traditional fat-tree DCNs, because the network is flatter and relies more on passive optical fanout.
-
Delivers up to about 33% higher throughput, due to more independent paths and better load spreading.
-
Cuts network equipment power consumption by on the order of 40%, with associated reductions in cooling and operational overhead.
On the control-plane side, AWS developed a routing scheme called Spraypoint. Instead of always following a strict shortest path from source to destination, Spraypoint first “sprays” traffic randomly to neighbors, then directs it via preselected “waypoints” using more conventional shortest-path routing. This hybrid behavior exploits the quasi-random topology to open many more independent paths than standard ECMP-style shortest-path routing would, which in turn improves utilization and resilience under congestion or failures.
Strategic implications:
For AWS’s cloud and AI build-out, this is positioned as a foundational infrastructure advantage: higher bisection bandwidth and lower network energy per bit directly benefit large-scale AI training clusters, storage backends, and multi-tenant cloud workloads. Fewer active devices and more passive optics also translate into lower capex and opex at hyperscale, so AWS is framing this as both a performance and cost/sustainability play that could save billions of dollars and reduce CO₂ emissions over time.
From a networking-theory standpoint, this is notable as one of the first reported at-scale, production deployments of a flat random-graph-inspired topology in a hyperscale DCN, rather than a purely academic or lab system.
In a quasi-random topology like AWS’s RNG fabric, the impact on latency and jitter comes from three main effects: path length distribution, load spreading, and failure behavior.
Baseline latency: path lengths and device count:
In a traditional Clos/fat-tree, average latency is dominated by a fixed number of stages (ToR → agg → core → agg → ToR), so hop count is tightly controlled but you pay for many active devices. A quasi-random, flat graph replaces that rigid hierarchy with many short, irregular paths; on average, shortest paths between any two switches are similar or slightly shorter in hop count than in a fat-tree, and there are fewer active routers in the path because the architecture offloads fanout to passive optics. That tends to keep or slightly reduce median/mean latency per flow, especially under moderate load, because packets traverse fewer serialized queueing points even if the physical graph looks “messier.”
Jitter: congestion and path diversity:
Jitter is driven much more by variable queueing delay than by fixed propagation or serialization. In a quasi-random fabric with many alternate paths and a load-balancing scheme like Spraypoint (random spray + waypoint-based shortest paths), flows can be spread more evenly across the network, reducing hot spots and thus reducing the variance of queueing delay across packets. That can lower jitter compared with a Clos under the same aggregate load, because the system is less likely to funnel many flows through the same few congested uplinks or spine devices.
However, because the routing intentionally uses many different paths, per-flow packet reordering becomes more likely unless constrained by per-flow hashing or waypointing, which can show up as effective jitter at higher layers. AWS’s description of Spraypoint suggests they mitigate this by using waypoints and policy to preserve some path structure, so you get the diversity benefits without unconstrained per-packet spraying.
Under failure and high load:
Where quasi-random really helps latency/jitter is under failure and partial congestion. In a Clos, link or spine failures can force large sets of flows to converge on a smaller subset of remaining equal-cost paths, driving up queueing delay and jitter nonlinearly. In a resilient random-graph-style fabric, node/edge failures simply remove a few edges from a highly connected graph; there are typically many alternative short paths, so the increase in hop count and queueing pressure is smaller and more diffuse. That tends to keep tail latency and jitter (P99, P99.9) better behaved, even if median latency looks similar to a Clos at low load.
So, qualitatively: median latency is roughly comparable to a well-designed Clos, sometimes better due to fewer active stages; jitter and tail latency should improve under realistic, bursty load and failure scenarios, provided the routing stack is designed to limit packet reordering.
Summary and Conclusions:
Quasi-random data center topologies like AWS’s RNG fabric replace rigid Clos/fat-tree hierarchies with a flatter, graph-like network that preserves short path lengths while dramatically increasing path diversity, which tends to hold median latency roughly steady or slightly better by reducing the number of active, queueing devices per path and offloading fanout to passive optics. They primarily improve jitter and tail latency by spreading flows across many alternative routes so congestion is less concentrated, making queueing delays less bursty and keeping P99/P99.9 behavior more stable under failures and hot spots, provided the routing layer (for example, AWS’s Spraypoint approach) constrains packet reordering through way pointing or per-flow consistency.
In conclusion, quasi-random fabrics are less about shaving a few microseconds off baseline latency and more about delivering more predictable end-to-end performance—especially for east–west, latency-sensitive cloud and AI workloads—by trading rigid structure for statistically robust, highly connected graphs that degrade more gracefully when links, nodes, or traffic patterns become pathological.
…………………………………………………………………………………………………………………………………………………………………….
References:
https://arxiv.org/pdf/2604.15261
https://www.wired.com/story/amazon-aws-ceo-matt-garman-ai-agents/
AWS to deploy AI inference chips from Cerebras in its data centers; Anapurna Labs/Amazon in-house AI silicon products
Amazon’s Jeff Bezos at Italian Tech Week: “AI is a kind of industrial bubble”
Data Center Networking Market to grow at a CAGR of 6.22% during 2022-2027 to reach $35.6 billion by 2027
TMR: Data Center Networking Market sees shift to user-centric & data-oriented business + CoreSite DC Tour
