AI in Networks Market
Ericsson integrates agentic AI into its NetCloud platform for self healing and autonomous 5G private networks
Ericsson is integrating agentic AI into its NetCloud platform to create self-healing and autonomous 5G private (enterprise) networks. This initiative upgrades the existing NetCloud Assistant (ANA), a generative AI tool, into a strategic partner capable of managing complex workflows and orchestrating multiple AI agents. The agentic AI agent aims to simplify private 5G adoption by reducing deployment complexity and the need for specialized administration. This new agentic architecture allows the new Ericsson system to interpret high-level instructions and autonomously assign tasks to a team of specialized AI agents.
Key AI features include:
- Agentic organizational hierarchy: ANA will be supported by multiple orchestrator and functional AI agents capable of planning and executing (with administrator direction). Orchestrator agents will be deployed in phases, starting with a troubleshooting agent planned in Q4 2025, followed by configuration, deployment, and policy agents planned in 2026. These orchestrators will connect with task, process, knowledge, and decision agents within an integrated agentic framework.
- Automated troubleshooting: ANA’s troubleshooting orchestrator will include automated workflows that address the top issues identified by Ericsson support teams, partners, and customers, such as offline devices and poor signal quality. Planned to launch in Q4 2025, this feature is expected to reduce downtime and customer support cases by over 20 percent.
- Multi-modal content generation: ANA can now generate dynamic graphs to visually represent trends and complex query results involving multiple data points.
- Explainable AI: ANA displays real-time process feedback, revealing steps taken by AI agents in order to enhance transparency and trust.
- Expanded AIOps insights: NetCloud AIOps will be expanded to provide isolation and correlation of fault, performance, configuration, and accounting anomalies for Wireless WAN and NetCloud SASE. For Ericsson Private 5G, NetCloud is expected to provide service health analytics including KPI monitoring and user equipment connectivity diagnostics. Planned availability Q4 2025.
Manish Tiwari, Head of Enterprise 5G, Ericsson Enterprise Wireless Solutions, adds: “With the integration of Ericsson Private 5G into the NetCloud platform, we’re taking a major step forward in making enterprise connectivity smarter, simpler, and adaptive. By building on powerful AI foundations, seamless lifecycle management, and the ability to scale securely across sites, we are providing flexibility to further accelerate digital transformation across industries. This is about more than connectivity: it is about giving enterprises the business-critical foundation they need to run IT and OT systems with confidence and unlock the next wave of innovation for their businesses.”
Pankaj Malhotra, Head of WWAN & Security, Ericsson Enterprise Wireless Solutions, says: “By introducing agentic AI into NetCloud, we’re enabling enterprises to simplify deployment and operations while also improving reliability, performance, and user experience. More importantly, it lays the foundation for our vision of fully autonomous, self-optimizing 5G enterprise networks, that can power the next generation of enterprise innovation.”
Agentic AI and the Future of Communications for Autonomous Vehicle (V2X)
Ericsson completes Aduna joint venture with 12 telcos to drive network API adoption
Ericsson reports ~flat 2Q-2025 results; sees potential for 5G SA and AI to drive growth
Ericsson revamps its OSS/BSS with AI using Amazon Bedrock as a foundation
Ericsson’s sales rose for the first time in 8 quarters; mobile networks need an AI boost
Highlights of Nokia’s Smart Factory in Oulu, Finland for 5G and 6G innovation
Nokia has opened a Smart Factory in Oulu, Finland, for 5G/6G design, manufacturing, and testing, integrating AI technologies and Industry 4.0 applications. It brings ~3,000 staff under one roof and is positioned as Europe’s flagship site for radio access (RAN) innovation.
The Oulu campus will initially focus on 5G, including: Standardization, System-on Chips as well as 5G radio hardware and software and patents. Oulu Factory, part of the new campus, will target New Production Introduction for Nokia’s 5G radio and baseband products. The new campus strengthens Oulu’s ecosystem as a global testbed for resilient and secure networks for both civilian and defense applications.
At Oulu “Home of Radio” campus, Nokia’s research and innovation underpins high quality, tested world class products readymade for customers across markets. Nokia’s experts will continue to foster innovation, from Massive MIMO radios like Osprey and Habrok to next-generation 6G solutions, creating secure, high-performance, future-proof connectivity.
Sustainability is integral to the facility. Renewable energy is used throughout the site, with additional energy used to heat 20,000 households in Oulu. The on-site energy station is one of the world’s largest CO2-based district heating and cooling plants.
Active 6G proof-of-concept trials will be tested using ~7 GHz and challenging propagation scenarios.
“Our teams in Oulu are shaping the future of 5G and 6G developing our most advanced radio networks. Oulu has a unique ecosystem that integrates Nokia’s R&D and smart manufacturing with an ecosystem of partners – including universities, start-ups and NATO’s DIANA test center. Oulu embodies our culture of innovation and the new campus will be essential to advancing connectivity necessary to power the AI supercycle,” said Justin Hotard, President and CEO of Nokia
Nokia Oulu Facts:
- Around 3,000 employees and 40 nationalities working on the campus.
- Oulu campus covers the entire product lifecycle of a product, from R&D to manufacturing and testing of the products.
- Footprint of the building is overall 55,000 square metres, including manufacturing, R&D and office space.
- Green campus with all energy purchased green and all surplus energy generated fed back into the district heating system and used to heat 20,000 local households.
- The campus boasts 100% waste utilization rate and 99% avoidance in CO2 emissions.
- Construction started in the second half of 2022, with the first employees moving into the facility in the first half of this year.
- YIT constructed the site and Arkkitehtitoimisto ALA were the architects.
References:
https://www.sdxcentral.com/analysis/behind-the-scenes-at-nokias-new-home-of-radio/
Will the wave of AI generated user-to/from-network traffic increase spectacularly as Cisco and Nokia predict?
Nokia’s Bell Labs to use adapted 4G and 5G access technologies for Indian space missions
Indosat Ooredoo Hutchison and Nokia use AI to reduce energy demand and emissions
Verizon partners with Nokia to deploy large private 5G network in the UK
Nokia selects Intel’s Justin Hotard as new CEO to increase growth in IP networking and data center connections
Nokia sees new types of 6G connected devices facilitated by a “3 layer technology stack”
Nokia and Eolo deploy 5G SA mmWave “Cloud RAN” network
OpenAI and Broadcom in $10B deal to make custom AI chips
Overview:
Late last October, IEEE Techblog reported that “OpenAI the maker of ChatGPT, was working with Broadcom to develop a new artificial intelligence (AI) chip focused on running AI models after they’ve been trained.” On Friday, the WSJ and FT (on-line subscriptions required) separately confirmed that OpenAI is working with Broadcom to develop custom AI chips, a move that could help alleviate the shortage of powerful processors needed to quickly train and release new versions of ChatGPT. OpenAI plans to use the new AI chip internally, according to one person close to the project, rather than make them available to external customers.
………………………………………………………………………………………………………………………………………………………………………………………….
Broadcom:
During its earnings call on Thursday, Broadcom’s CEO Hock Tan said that it had signed up an undisclosed fourth major AI developer as a custom AI chip customer, and that this new customer had committed to $10bn in orders. While Broadcom did not disclose the names of the new customer, people familiar with the matter confirmed OpenAI was the new client. Broadcom and OpenAI declined to comment, according to the FT. Tan said the deal had lifted the company’s growth prospects by bringing “immediate and fairly substantial demand,” shipping chips for that customer “pretty strongly” starting next year. “The addition of a fourth customer with immediate and fairly substantial demand really changes our thinking of what 2026 would be starting to look like,” Tan added.
Image credit: © Dado Ruvic/Reuters
HSBC analysts have recently noted that they expect to see a much higher growth rate from Broadcom’s custom chip business compared with Nvidia’s chip business in 2026. Nvidia continues to dominate the AI silicon market, with “hyperscalers” still representing the largest share of its customer base. While Nvidia doesn’t disclose specific customer names, recent filings show that a significant portion of their revenue comes from a small number of unidentified direct customers, which likely are large cloud providers like Microsoft, Amazon, Alphabet (Google), and Meta Platforms.
In August, Broadcom launched its Jericho networking chip, which is designed to help speed up AI computing by connecting data centers as far as 60 miles apart. By August, Broadcom’s market value had surpassed that of oil giant Saudi Aramco, making the chip firm the world’s seventh-largest publicly listed company.
……………………………………………………………………………………………………………………………………….
Open AI:
OpenAI CEO Sam Altman has been saying for months that a shortage of graphics processing units, or GPUs, has been slowing his company’s progress in releasing new versions of its flagship chatbot. In February, Altman wrote on X that ChatGPT-4.5, its then-newest large language model, was the closest the company had come to designing an AI model that behaved like a “thoughtful person,” but there were very high costs that came with developing it. “We will add tens of thousands of GPUs next week and roll it out to the plus tier then. (hundreds of thousands coming soon, and i’m pretty sure y’all will use every one we can rack up.)”
In recent years, OpenAI has relied heavily on so-called “off the shelf” GPUs produced by Nvidia, the biggest player in the chip-design space. But as demand from large AI firms looking to train increasingly sophisticated models has surged, chip makers and data-center operators have struggled to keep up. The company was one of the earliest customers for Nvidia’s AI chips and has since proven to be a voracious consumer of its AI silicon.
“If we’re talking about hyperscalers and gigantic AI factories, it’s very hard to get access to a high number of GPUs,” said Nikolay Filichkin, co-founder of Compute Labs, a startup that buys GPUs and offers investors a share in the rental income they produce. “It requires months of lead time and planning with the manufacturers.”
To solve this problem, OpenAI has been working with Broadcom for over a year to develop a custom chip for use in model training. Broadcom specializes in what it calls XPUs, a type of semiconductor that is designed with a particular application—such as training ChatGPT—in mind.
Last month, Altman said the company was prioritizing compute “in light of the increased demand from [OpenAI’s latest model] GPT-5” and planned to double its compute fleet “over the next 5 months.” OpenAI also recently struck a data-center deal with Oracle that calls for OpenAI to pay more than $30 billion a year to the cloud giant, and signed a smaller contract with Google earlier this year to alleviate computing shortages. It is also embarking on its own data-center construction project, Stargate, though that has gotten off to a slow start.
OpenAI’s move follows the strategy of tech giants such as Google, Amazon and Meta, which have designed their own specialized custom chips to run AI workloads. The industry has seen huge demand for the computing power to train and run AI models.
………………………………………………………………………………………………………………………………………………………………………………..
References:
https://www.ft.com/content/e8cc6d99-d06e-4e9b-a54f-29317fa68d6f
https://www.wsj.com/tech/ai/openai-broadcom-deal-ai-chips-5c7201d2
Reuters & Bloomberg: OpenAI to design “inference AI” chip with Broadcom and TSMC
Open AI raises $8.3B and is valued at $300B; AI speculative mania rivals Dot-com bubble
OpenAI announces new open weight, open source GPT models which Orange will deploy
OpenAI partners with G42 to build giant data center for Stargate UAE project
Generative AI Unicorns Rule the Startup Roost; OpenAI in the Spotlight
Will the wave of AI generated user-to/from-network traffic increase spectacularly as Cisco and Nokia predict?
Network operators are bracing themselves for a wave of AI traffic, partially based on a RtBrick survey, as well as forecasts by Cisco and Nokia, but that hasn’t happened yet. The heavy AI traffic today is East to West (or vice-versa) within cloud resident AI data centers and for AI data center interconnects.
1. Cisco believes that AI Inference agents will soon engage “continuously” with end-users, keeping traffic levels consistently high. has stated that AI will greatly increase network traffic, citing a shift toward new, more demanding traffic patterns driven by “agentic AI” and other applications. This perspective is a core part of Cisco’s business strategy, which is focused on selling the modernized infrastructure needed to handle the coming surge. Cisco identified three stages of AI-driven traffic growth, each with different network demands:
- Today’s generative AI models produce “spikey” traffic which spikes up when a user submits a query, but then returns to a low baseline. Current networks are largely handling this traffic without issues.
- Persistent “agentic” AI traffic: The next phase will involve AI agents that constantly interact with end-users and other agents. Cisco CEO Chuck Robbins has stated that this will drive traffic “beyond the peaks of current chatbot interaction” and keep network levels “consistently high”.
- Edge-based AI: A third wave of “physical AI” will require more computing and networking at the edge of the network to accommodate specialized use cases like industrial IoT.
“As we move towards agentic AI and the demand for inferencing expands to the enterprise and end user networking environments, traffic on the network will reach unprecedented levels,” Cisco CEO Chuck Robbins said on the company’s recent earnings call. “Network traffic will not only increase beyond the peaks of current chatbot interaction, but will remain consistently high with agents in constant interaction.”
2. Nokia recently predicted that both direct and indirect AI traffic on mobile networks will grow at a faster pace than regular, non-AI traffic.
- Direct AI traffic: This is generated by users or systems directly interacting with AI services and applications. Consumer examples: Using generative AI tools, interacting with AI-powered gaming, or experiencing extended reality (XR) environments. Enterprise examples: Employing predictive maintenance, autonomous operations, video and image analytics, or enhanced customer interactions.
- Indirect AI traffic: This occurs when AI algorithms are used to influence user engagement with existing services, thereby increasing overall traffic. Examples: AI-driven personalized recommendations for video content on social media, streaming platforms, and online marketplaces, which can lead to longer user sessions and higher bandwidth consumption.
The Finland based network equipment vendor warned that the AI wave could bring “a potential surge in uplink data traffic that could overwhelm our current network infrastructure if we’re not prepared,” noting that the rise of hybrid on-device and cloud tools will require much more than the 5-15 Mbps uplink available on today’s networks. Nokia’s Global Network Traffic 2030 report forecasts that overall traffic could grow by 5 to 9 times current levels by 2033. All told, Nokia said AI traffic is expected to hit 1088 exabytes (EB) per month by 2033. That means overall traffic will grow 5x in a best case scenario and 9x in a worse case.
To manage this anticipated traffic surge, Nokia advocates for radical changes to existing network infrastructure.
- Cognitive networks: The company states that networks must become “cognitive,” leveraging AI and machine learning (ML) to handle the growing data demand.
- Network-as-Code: As part of its Technology Strategy 2030, Nokia promotes a framework for more flexible and scalable networks that leverage AI and APIs.
- 6G preparation: Nokia Bell Labs is already conducting research and field tests to prepare for 6G networks around 2030, with a focus on delivering the capacity needed for AI and other emerging technologies.
- Optimizing the broadband edge: The company also highlights the need to empower the broadband network edge to handle the demands of AI applications, which require low latency and high reliability.
Nokia’s Global Network Traffic 2030 report didn’t mention agentic AI, which are artificial intelligence systems designed to autonomously perceive, reason, and act in their environment to achieve complex goals with less human oversight. Unlike generative AI, which focuses on creating content, agentic AI specializes in workflow automation and independent problem-solving by making decisions, adapting plans, and executing tasks over extended periods to meet long-term objectives.
3. Ericsson did point to traffic increases stemming from the use of AI-based assistants in its 2024 Mobility Report. In particular, it predicted the majority of traffic would be related to the use of consumer video AI assistants, rather than text-based applications and – outside the consumer realm – forecast increased traffic from “AI agents interacting with drones and droids. Accelerated consumer uptake of GenAI will cause a steady increase of traffic in addition to the baseline increase,” Ericsson said of its traffic growth scenario.
…………………………………………………………………………………………………………………………………………………………………………………..
Dissenting Views:
1. UK Disruptive Analysis Founder Dean Bubley isn’t a proponent of huge AI traffic growth. “Many in the telecom industry and vendor community are trying to talk up AI as driving future access network traffic and therefore demand for investment, spectrum etc., but there is no evidence of this at present,” he told Fierce Network.
Bubley argues that AI agents won’t really create much traffic on access networks to homes or businesses. Instead, he said, they will drive traffic “inside corporate networks, and inside and between data centers on backbone networks and inside the cloud. “There might be a bit more uplink traffic if video/images are sent to the cloud for AI purposes, but again that’s hypothetical,” he said.
2. In a LinkedIn post, Ookla analyst Mike Dano said he was a bit suspicious about “Cisco predicting a big jump in network traffic due to AI agents constantly wandering around the Internet and doing things.” While almost all of the comments agreed with Dano, it still is an open question whether the AI traffic Armageddon will actually materialize.
……………………………………………………………………………………………………………………………………………………………………………………….
References:
RtBrick survey: Telco leaders warn AI and streaming traffic to “crack networks” by 2030
https://www.fierce-network.com/cloud/will-ai-agents-really-raise-network-traffic-baseline
Analysis: Cisco, HPE/Juniper, and Nvidia network equipment for AI data centers
Both telecom and enterprise networks are being reshaped by AI bandwidth and latency demands of AI. Network operators that fail to modernize architectures risk falling behind. Why? AI workloads are network killers — they demand massive east-west traffic, ultra-low latency, and predictable throughput.
- Real-time observability is becoming non-negotiable, as enterprises need to detect and fix issues before they impact AI model training or inference.
- Self-driving networks are moving from concept to reality, with AI not just monitoring but actively remediating problems.
- The competitive race is now about who can integrate AI into networking most seamlessly — and HPE/Juniper’s Mist AI, Cisco’s assurance stack, and Nvidia’s AI fabrics are three different but converging approaches.
Cisco, HPE/Juniper, and Nvidia are designing AI-optimized networking equipment, with a focus on real-time observability, lower latency and increased data center performance for AI workloads. Here’s a capsule summary:
Cisco: AI-Ready Infrastructure:
- Cisco is embedding AI telemetry and analytics into its Silicon One chips, Nexus 9000 switches, and Catalyst campus gear.
- The focus is on real-time observability via its ThousandEyes platform and AI-driven assurance in DNA Center, aiming to optimize both enterprise and AI/ML workloads.
- Cisco is also pushing AI-native data center fabrics to handle GPU-heavy clusters for training and inference.
- Cisco claims “exceptional momentum” and leadership in AI: >$800M in AI infrastructure orders taken from web-scale customers in Q4, bringing the FY25 total to over $2B.
- Cisco Nexus switches now fully and seamlessly integrated with NVIDIA’s Spectrum-X architecture to deliver high speed networking for AI clusters
HPE + Juniper: AI-Native Networking Push:
- Following its $13.4B acquisition of Juniper Networks, HPE has merged Juniper’s Mist AI platform with its own Aruba portfolio to create AI-native, “self-driving” networks.
- Key upgrades include:
-Agentic AI troubleshooting that uses generative AI workflows to pinpoint and fix issues across wired, wireless, WAN, and data center domains.
-Marvis AI Assistant with enhanced conversational capabilities — IT teams can now ask open-ended questions like “Why is the Orlando site slow?” and get contextual, actionable answers.
-Large Experience Model (LEM) with Marvis Minis — digital twins that simulate user experiences to predict and prevent performance issues before they occur.
-Apstra integration for data center automation, enabling autonomous service provisioning and cross-domain observability
Nvidia: AI Networking at Compute Scale
- Nvidia’s Spectrum-X Ethernet platform and Quantum-2 InfiniBand (both from Mellanox acquisition) are designed for AI supercomputing fabrics, delivering ultra-low latency and congestion control for GPU clusters.
- In partnership with HPE, Nvidia is integrating NVIDIA AI Enterprise and Blackwell architecture GPUs into HPE Private Cloud AI, enabling enterprises to deploy AI workloads with optimized networking and compute together.
- Nvidia’s BlueField DPUs offload networking, storage, and security tasks from CPUs, freeing resources for AI processing.
………………………………………………………………………………………………………………………………………………………..
Here’s a side-by-side comparison of how Cisco, HPE/Juniper, and Nvidia are approaching AI‑optimized enterprise networking — so you can see where they align and where they differentiate:
| Feature / Focus Area | Cisco | HPE / Juniper | Nvidia |
|---|---|---|---|
| Core AI Networking Vision | AI‑ready infrastructure with embedded analytics and assurance for enterprise + AI workloads | AI‑native, “self‑driving” networks across campus, WAN, and data center | High‑performance fabrics purpose‑built for AI supercomputing |
| Key Platforms | Silicon One chips, Nexus 9000 switches, Catalyst campus gear, ThousandEyes, DNA Center | Mist AI platform, Marvis AI Assistant, Marvis Minis, Apstra automation | Spectrum‑X Ethernet, Quantum‑2 InfiniBand, BlueField DPUs |
| AI Integration | AI‑driven assurance, predictive analytics, real‑time telemetry | Generative AI for troubleshooting, conversational AI for IT ops, digital twin simulations | AI‑optimized networking stack tightly coupled with GPU compute |
| Observability | End‑to‑end visibility via ThousandEyes + DNA Center | Cross‑domain observability (wired, wireless, WAN, DC) with proactive issue detection | Telemetry and congestion control for GPU clusters |
| Automation | Policy‑driven automation in campus and data center fabrics | Autonomous provisioning, AI‑driven remediation, intent‑based networking | Offloading networking/storage/security tasks to DPUs for automation |
| Target Workloads | Enterprise IT, hybrid cloud, AI/ML inference & training | Enterprise IT, edge, hybrid cloud, AI/ML workloads | AI training & inference at hyperscale, HPC, large‑scale data centers |
| Differentiator | Strong enterprise install base + integrated assurance stack | Deep AI‑native operations with user experience simulation | Ultra‑low latency, high‑throughput fabrics for GPU‑dense environments |
Key Takeaways:
- Cisco is strongest in enterprise observability and broad infrastructure integration.
- HPE/Juniper is leaning into AI‑native operations with a heavy focus on automation and user experience simulation.
- Nvidia is laser‑focused on AI supercomputing performance, building the networking layer to match its GPU dominance.
- Cisco leverages its market leadership, customer base and strategic partnerships to integrate AI with existing enterprise networks.
- HPE/Juniper challenges rivals with an AI-native, experience-first network management platform.
- Nvidia aims to dominate the full-stack AI infrastructure, including networking.
SoftBank’s Transformer AI model boosts 5G AI-RAN uplink throughput by 30%, compared to a baseline model without AI
Softbank has developed its own Transformer-based AI model that can be used for wireless signal processing. SoftBank used its Transformer model to improve uplink channel interpolation which is a signal processing technique where the network essentially makes an educated guess as to the characteristics and current state of a signal’s channel. Enabling this type of intelligence in a network contributes to faster, more stable communication, according to SoftBank. The Japanese wireless network operator successfully increased uplink throughput by approximately 20% compared to a conventional signal processing method (the baseline method). In the latest demonstration, the new Transformer-based architecture was run on GPUs and tested in a live Over-the-Air (OTA) wireless environment. In addition to confirming real-time operation, the results showed further throughput gains and achieved ultra-low latency.
Editor’s note: A Transformer model is a type of neural network architecture that emerged in 2017. It excels at interpreting streams of sequential data associated with large language models (LLMs). Transformer models have also achieved elite performance in other fields of artificial intelligence (AI), including computer vision, speech recognition and time series forecasting. Transformer models are lightweight, efficient, and versatile – capable of natural language processing (NLP), image recognition and wireless signal processing as per this Softbank demo.
Significant throughput improvement:
- Uplink channel interpolation using the new architecture improved uplink throughput by approximately 8% compared to the conventional CNN model. Compared to the baseline method without AI, this represents an approximately 30% increase in throughput, proving that the continuous evolution of AI models leads to enhanced communication quality in real-world environments.
Higher AI performance with ultra-low latency:
- While real-time 5G communication requires processing in under 1 millisecond, this demonstration with the Transformer achieved an average processing time of approximately 338 microseconds, an ultra-low latency that is about 26% faster than the convolution neural network (CNN) [1.] based approach. Generally, AI model processing speeds decrease as performance increases. This achievement overcomes the technically difficult challenge of simultaneously achieving higher AI performance and lower latency. Editor’s note: Perhaps this can overcome the performance limitations in ITU-R M.2150 for URRLC in the RAN, which is based on an uncompleted 3GPP Release 16 specification.
Note 1. CNN-based approaches to achieving low latency focus on optimizing model architecture, computation, and hardware to accelerate inference, especially in real-time applications. Rather than relying on a single technique, the best results are often achieved through a combination of methods.
Using the new architecture, SoftBank conducted a simulation of “Sounding Reference Signal (SRS) prediction,” a process required for base stations to assign optimal radio waves (beams) to terminals. Previous research using a simpler Multilayer Perceptron (MLP) AI model for SRS prediction confirmed a maximum downlink throughput improvement of about 13% for a terminal moving at 80 km/h.*2
In the new simulation with the Transformer-based architecture, the downlink throughput for a terminal moving at 80 km/h improved by up to approximately 29%, and by up to approximately 31% for a terminal moving at 40 km/h. This confirms that enhancing the AI model more than doubled the throughput improvement rate (see Figure 1). This is a crucial achievement that will lead to a dramatic improvement in communication speeds, directly impacting the user experience.
The most significant technical challenge for the practical application of “AI for RAN” is to further improve communication quality using high-performance AI models while operating under the real-time processing constraint of less than one millisecond. SoftBank addressed this by developing a lightweight and highly efficient Transformer-based architecture that focuses only on essential processes, achieving both low latency and maximum AI performance. The important features are:
(1) Grasps overall wireless signal correlations
By leveraging the “Self-Attention” mechanism, a key feature of Transformers, the architecture can grasp wide-ranging correlations in wireless signals across frequency and time (e.g., complex signal patterns caused by radio wave reflection and interference). This allows it to maintain high AI performance while remaining lightweight. Convolution focuses on a part of the input, while Self-Attention captures the relationships of the entire input (see Figure 2).
(2) Preserves physical information of wireless signals
While it is common to normalize input data to stabilize learning in AI models, the architecture features a proprietary design that uses the raw amplitude of wireless signals without normalization. This ensures that crucial physical information indicating communication quality is not lost, significantly improving the performance of tasks like channel estimation.
(3) Versatility for various tasks
The architecture has a versatile, unified design. By making only minor changes to its output layer, it can be adapted to handle a variety of different tasks, including channel interpolation/estimation, SRS prediction, and signal demodulation. This reduces the time and cost associated with developing separate AI models for each task.
The demonstration results show that high-performance AI models like Transformer and the GPUs that run them are indispensable for achieving the high communication performance required in the 5G-Advanced and 6G eras. Furthermore, an AI-RAN that controls the RAN on GPUs allows for continuous performance upgrades through software updates as more advanced AI models emerge, even after the hardware has been deployed. This will enable telecommunication carriers to improve the efficiency of their capital expenditures and maximize value.
Moving forward, SoftBank will accelerate the commercialization of the technologies validated in this demonstration. By further improving communication quality and advancing networks with AI-RAN, SoftBank will contribute to innovation in future communication infrastructure. The Japan based conglomerate strongly endorsed AI RAN at MWC 2025.
References:
https://www.softbank.jp/en/corp/news/press/sbkk/2025/20250821_02/
https://www.telecoms.com/5g-6g/softbank-claims-its-ai-ran-tech-boosts-throughput-by-30-
https://www.telecoms.com/ai/softbank-makes-mwc-25-all-about-ai-ran
https://www.ibm.com/think/topics/transformer-model
https://www.itu.int/rec/R-REC-M.2150/en
Softbank developing autonomous AI agents; an AI model that can predict and capture human cognition
Dell’Oro Group: RAN Market Grows Outside of China in 2Q 2025
Dell’Oro: AI RAN to account for 1/3 of RAN market by 2029; AI RAN Alliance membership increases but few telcos have joined
Dell’Oro: RAN revenue growth in 1Q2025; AI RAN is a conundrum
Nvidia AI-RAN survey results; AI inferencing as a reinvention of edge computing?
OpenAI announces new open weight, open source GPT models which Orange will deploy
Deutsche Telekom and Google Cloud partner on “RAN Guardian” AI agent
RtBrick survey: Telco leaders warn AI and streaming traffic to “crack networks” by 2030
Respondents to a RtBrick survey of 200 senior telecom decision makers in the U.S., UK, and Australia finds that network operator leaders are failing to make key decisions and lack the motivation to change. The report exposes urgent warnings from telco engineers that their networks are on a five-year collision course with AI and streaming traffic. It finds that 93% of respondents report a lack of support from leadership to deploy disaggregated network equipment. Key findings:
- Risk-averse leadership and a lack of skills are the top factors that are choking progress.
- Majority are stuck in early planning, while AT&T, Deutsche Telekom, and Comcast lead large-scale disaggregation rollouts.
- Operators anticipate higher broadband prices but fear customer backlash if service quality can’t match the price.
- Organizations require more support from leadership to deploy disaggregation (93%).
- Complexity around operational transformation (42%), such as redesigning architectures and workflows.
- Critical shortage of specialist skills/staff (38%) to manage disaggregated systems.
The survey finds that almost nine in ten operators (87%) expect customers to demand higher broadband speeds by 2030, while roughly the same (79%) state their customers expect costs to increase, suggesting they will pay more for it. Yet half of all leaders (49%) admit they lack complete confidence in delivering services at a viable cost. Eighty-four percent say customer expectations for faster, cheaper broadband are already outpacing their networks, while 81% concede their current architectures are not well-suited to handling the future increases in bandwidth demand, suggesting they may struggle with the next wave of AI and streaming traffic.
“Senior leaders, engineers, and support staff inside operators have made their feelings clear: the bottleneck isn’t capacity, it’s decision-making,” said Pravin S Bhandarkar, CEO and Founder of RtBrick. “Disaggregated networks are no longer an experiment. They’re the foundation for the agility, scalability, and transparency operators need to thrive in an AI-driven, streaming-heavy future,” he added noting the intent to deploy disaggregation as per this figure:
However, execution continues to trail ambition. Only one in twenty leaders has confirmed they’re “in deployment” today, while 49% remain stuck in early-stage “exploration”, and 38% are still “in planning”. Meanwhile, big-name operators such as AT&T, Deutsche Telekom, and Comcast are charging ahead and already actively deploying disaggregation at scale, demonstrating faster rollouts, greater operational control, and true vendor flexibility. Here’s a snapshot of those activities:
- AT&T has deployed an open, disaggregated routing network in their core, powered by DriveNets Network Cloud software on white-box bare metal switches and routers from Taiwanese ODMs, according to Israel based DriveNets. DriveNets utilizes a Distributed Disaggregated Chassis (DDC) architecture, where a cluster of bare metal switches act as a single routing entity. That architecture has enabled AT&T to accelerate 5G and fiber rollouts and improve network scalability and performance. It has made 1.6Tb/s transport a reality on AT&T’s live network.
- Deutsche Telekom has deployed a disaggregated broadband network using routing software from RtBrick running on bare-metal switch hardware to provide high-speed internet connectivity. They’re also actively promoting Open BNG solutions as part of this initiative.
- Comcast uses network cloud software from DriveNets and white-box hardware to disaggregate their core network, aiming to increase efficiency and enable new services through a self-healing and consumable network. This also includes the use of disaggregated, pluggable optics from multiple vendors.
Nearly every leader surveyed also claims their organization is “using” or “planning to use” AI in network operations, including for planning, optimization, and fault resolution. However, nine in ten (93%) say they cannot unlock AI’s full value without richer, real-time network data. This requires more open, modular, software-driven architecture, enabled by network disaggregation.
“Telco leaders see AI as a powerful asset that can enhance network performance,” said Zara Squarey, Research Manager at Vanson Bourne. “However, the data shows that without support from leadership, specialized expertise, and modern architectures that open up real-time data, disaggregation deployments may risk further delays.”
When asked what benefits they expect disaggregation to deliver, operators focused on outcomes that could deliver the following benefits:
- 54% increased operational automation
- 54% enhanced supply chain resilience
- 51% improved energy efficiency
- 48% lower purchase and operational costs
- 33% reduced vendor lock-in
Transformation priorities align with those goals, with automation and agility (57%) ranked first, followed by vendor flexibility (55%), supply chain security (51%), cost efficiency (46%) and energy usage and sustainability (47%).
About the research:
The ‘State of Disaggregation’ research was independently conducted by Vanson Bourne in June 2025 and commissioned by RtBrick to identify the primary drivers and barriers to disaggregated network rollouts. The findings are based on responses from 200 telecom decision makers across the U.S., UK, and Australia, representing operations, engineering, and design/Research and Development at organizations with 100 to 5,000 or more employees.
References:
https://www.rtbrick.com/state-of-disaggregation-report-2
https://drivenets.com/blog/disaggregation-is-driving-the-future-of-atts-ip-transport-today/
Disaggregation of network equipment – advantages and issues to consider
OpenAI announces new open weight, open source GPT models which Orange will deploy
Overview:
OpenAI today introduced two new open-weight, open-source GPT models (gpt-oss-120b and gpt-oss-20b) designed to deliver top-tier performance at a lower cost. Available under the flexible Apache 2.0 license, these models outperform similarly sized open models on reasoning tasks, demonstrate strong tool use capabilities, and are optimized for efficient deployment on consumer hardware. They were trained using a mix of reinforcement learning and techniques informed by OpenAI’s most advanced internal models, including o3 and other frontier systems.
These two new AI models require much less compute power to run, with the gpt-oss20B version able to run on just 16 GB of memory. The smaller memory size and less compute power enables OpenAI’s models to run in a wider variety of environments, including at the network edge. The open weights mean those using the models can tweak the training parameters and customize them for specific tasks.
OpenAI has been working with early partner companies, including AI Sweden, Orange, and Snowflake to learn about real-world applications of our open models, from hosting these models on-premises for data security to fine-tuning them on specialized datasets. We’re excited to provide these best-in-class open models to empower everyone—from individual developers to large enterprises to governments—to run and customize AI on their own infrastructure. Coupled with the models available in our API, developers can choose the performance, cost, and latency they need to power AI workflows.
In lockstep with OpenAI, France’s Orange today announced plans to deploy the new OpenAI models in its regional cloud data centers as well as small on-premises servers and edge sites to meet demand for sovereign AI solutions. Orange’s deep AI engineering talent enables it to customize and distill the OpenAI models for specific tasks, effectively creating smaller sub-models for particular use-cases, while ensuring the protection of all sensitive data used in these customized models. This process facilitates innovative use-cases in network operations and will enable Orange to build on its existing suite of ‘Live Intelligence’ AI solutions for enterprises, as well as utilizing it for its own operational needs to improve efficiency, and drive cost savings.
Using AI to improve the quality and resilience of its networks, for example by enabling Orange to more easily explore and diagnose complex network issues with the help of AI. This can be achieved with trusted AI models that operate entirely within Orange sovereign data centers where Orange has complete control over the use of sensitive network data. This ability to create customized, secure, and sovereign AI models for network use cases is a key enabler in Orange’s mission to achieve higher levels of automation across all of its networks.
Steve Jarrett, Orange’s Chief AI Officer, noted the decision to use state-of-the-art open-weight models will allow it to drive “new use cases to address sensitive enterprise needs, help manage our networks, enable innovating customer care solutions including African regional languages, and much more.”
Performance of the new OpenAI models:
gpt-oss-120b outperforms OpenAI o3‑mini and matches or exceeds OpenAI o4-mini on competition coding (Codeforces), general problem solving (MMLU and HLE) and tool calling (TauBench). It furthermore does even better than o4-mini on health-related queries (HealthBench) and competition mathematics (AIME 2024 & 2025). gpt-oss-20b matches or exceeds OpenAI o3‑mini on these same evals, despite its small size, even outperforming it on competition mathematics and health.
Sovereign AI Market Forecasts:
Open-weight and open-source AI models play a significant role in enabling and shaping the development of Sovereign AI, which refers to a nation’s or organization’s ability to control its own AI technologies, data, and infrastructure to meet its specific needs and regulations.
Sovereign AI refers to a nation’s ability to control and manage its own AI development and deployment, including data, infrastructure, and talent. It’s about ensuring a country’s strategic autonomy in the realm of artificial intelligence, enabling them to leverage AI for their own economic, social, and security interests, while adhering to their own values and regulations.
Bank of America’s financial analysts recently forecast the sovereign AI market segment could become a “$50 billion a year opportunity, accounting for 10%–15% of the global $450–$500 billion AI infrastructure market.”
BofA analysts said, “Sovereign AI nicely complements commercial cloud investments with a focus on training and inference of LLMs in local culture, language and needs,” and could mitigate challenges such as “limited power availability for data centers in US” and trade restrictions with China.
References:
https://openai.com/index/introducing-gpt-oss/
https://newsroom.orange.com/orange-and-openai-collaborate-on-trusted-responsible-and-inclusive-ai/
https://finance.yahoo.com/news/nvidia-amd-targets-raised-bofa-162314196.html
Open AI raises $8.3B and is valued at $300B; AI speculative mania rivals Dot-com bubble
OpenAI partners with G42 to build giant data center for Stargate UAE project
Reuters & Bloomberg: OpenAI to design “inference AI” chip with Broadcom and TSMC
Nvidia’s networking solutions give it an edge over competitive AI chip makers
Nvidia’s networking equipment and module sales accounted for $12.9 billion of its $115.1 billion in data center revenue in its prior fiscal year. Composed of its NVLink, InfiniBand, and Ethernet solutions, Nvidia’s networking products (from its Mellanox acquisition) are what allow its GPU chips to communicate with each other, let servers talk to each other inside massive data centers, and ultimately ensure end users can connect to it all to run AI applications.
“The most important part in building a supercomputer is the infrastructure. The most important part is how you connect those computing engines together to form that larger unit of computing,” explained Gilad Shainer, senior vice president of networking at Nvidia.
In Q1-2025, networking made up $4.9 billion of Nvidia’s $39.1 billion in data center revenue. And it’ll continue to grow as customers continue to build out their AI capacity, whether that’s at research universities or massive data centers.
“It is the most underappreciated part of Nvidia’s business, by orders of magnitude,” Deepwater Asset Management managing partner Gene Munster told Yahoo Finance. “Basically, networking doesn’t get the attention because it’s 11% of revenue. But it’s growing like a rocket ship. “[Nvidia is a] very different business without networking,” Munster explained. “The output that the people who are buying all the Nvidia chips [are] desiring wouldn’t happen if it wasn’t for their networking.”
Nvidia senior vice president of networking Kevin Deierling says the company has to work across three different types of networks:
- NVLink technology connects GPUs to each other within a server or multiple servers inside of a tall, cabinet-like server rack, allowing them to communicate and boost overall performance.
- InfiniBand connects multiple server nodes across data centers to form what is essentially a massive AI computer.
- Ethernet connectivity for front-end network for storage and system management.
Note: Industry groups also have their own competing networking technologies including UALink, which is meant to go head-to-head with NVLink, explained Forrester analyst Alvin Nguyen.
“Those three networks are all required to build a giant AI-scale, or even a moderately sized enterprise-scale, AI computer,” Deierling explained. Low latency is key as longer transit times for data going to/from GPUs slows the entire operation, delaying other processes and impacting the overall efficiency of an entire data center.
Nvidia CEO Jensen Huang presents a Grace Blackwell NVLink72 as he delivers a keynote address at the Consumer Electronics Show (CES) in Las Vegas, Nevada on January 6, 2025. Photo by PATRICK T. FALLON/AFP via Getty Images
As companies continue to develop larger AI models and autonomous and semi-autonomous agentic AI capabilities that can perform tasks for users, making sure those GPUs work in lockstep with each other becomes increasingly important.
The AI industry is in the midst of a broad reordering around the idea of inferencing, which requires more powerful data center systems to run AI models. “I think there’s still a misperception that inferencing is trivial and easy,” Deierling said.
“It turns out that it’s starting to look more and more like training as we get to [an] agentic workflow. So all of these networks are important. Having them together, tightly coupled to the CPU, the GPU, and the DPU [data processing unit], all of that is vitally important to make inferencing a good experience.”
Competitor AI chip makers, like AMD are looking to grab more market share from Nvidia, and cloud giants like Amazon, Google, and Microsoft continue to design and develop their own AI chips. However, none of them have the low latency, high speed connectivity solutions provided by Nvidia (again, think Mellanox).
References:
https://www.nvidia.com/en-us/networking/
Networking chips and modules for AI data centers: Infiniband, Ultra Ethernet, Optical Connections
Nvidia enters Data Center Ethernet market with its Spectrum-X networking platform
Superclusters of Nvidia GPU/AI chips combined with end-to-end network platforms to create next generation data centers
Does AI change the business case for cloud networking?
The case for and against AI-RAN technology using Nvidia or AMD GPUs
Telecom sessions at Nvidia’s 2025 AI developers GTC: March 17–21 in San Jose, CA
Huawei launches CloudMatrix 384 AI System to rival Nvidia’s most advanced AI system
On Saturday, Huawei Technologies displayed an advanced AI computing system in China, as the Chinese technology giant seeks to capture market share in the country’s growing artificial intelligence sector. Huawei’s CloudMatrix 384 system made its first public debut at the World Artificial Intelligence Conference (WAIC), a three-day event in Shanghai where companies showcase their latest AI innovations, drawing a large crowd to the company’s booth.
The Huawei CloudMatrix 384 is a high-density AI computing system featuring 384 Huawei Ascend 910C chips, designed to rival Nvidia’s GB200 NVL72 (more below). The AI system employs a “supernode” architecture with high-speed internal chip interconnects. The system is built with optical links for low-latency, high-bandwidth communication. Huawei has also integrated the CloudMatrix 384 into its cloud platform. The system has drawn close attention from the global AI community since Huawei first announced it in April.
The CloudMatrix 384 resides on the super-node Ascend platform and uses high-speed bus interconnection capability, resulting in low latency linkage between 384 Ascend NPUs. Huawei says that “compared to traditional AI clusters that often stack servers, storage, network technology, and other resources, Huawei CloudMatrix has a super-organized setup. As a result, it also reduces the chance of facing failures at times of large-scale training.
Huawei staff at its WAIC booth declined to comment when asked to introduce the CloudMatrix 384 system. A spokesperson for Huawei did not respond to questions. However, Huawei says that “early reports revealed that the CloudMatrix 384 can offer 300 PFLOPs of dense BF16 computing. That’s double of Nvidia GB200 NVL72 AI tech system. It also excels in terms of memory capacity (3.6x) and bandwidth (2.1x).” Indeed, industry analysts view the CloudMatrix 384 as a direct competitor to Nvidia’s GB200 NVL72, the U.S. GPU chipmaker’s most advanced system-level product currently available in the market.
One industry expert has said the CloudMatrix 384 system rivals Nvidia’s most advanced offerings. Dylan Patel, founder of semiconductor research group SemiAnalysis, said in an April article that Huawei now had AI system capabilities that could beat Nvidia’s AI system. The CloudMatrix 384 incorporates 384 of Huawei’s latest 910C chips and outperforms Nvidia’s GB200 NVL72 on some metrics, which uses 72 B200 chips, according to SemiAnalysis. The performance stems from Huawei’s system design capabilities, which compensate for weaker individual chip performance through the use of more chips and system-level innovations, SemiAnalysis said.
Huawei has become widely regarded as China’s most promising domestic supplier of chips essential for AI development, even though the company faces U.S. export restrictions. Nvidia CEO Jensen Huang told Bloomberg in May that Huawei had been “moving quite fast” and named the CloudMatrix as an example.
Huawei says the system uses “supernode” architecture that allows the chips to interconnect at super-high speeds and in June, Huawei Cloud CEO Zhang Pingan said the CloudMatrix 384 system was operational on Huawei’s cloud platform.
According to Huawei, the Ascend AI chip-based CloudMatrix 384 with three important benefits:
- Ultra-large bandwidth
- Ultra-Low Latency
- Ultra-Strong Performance
These three perks can help enterprises achieve better AI training as well as stable reasoning performance for models. They could further retain long-term reliability.
References:
https://www.huaweicentral.com/huawei-launches-cloudmatrix-384-ai-chip-cluster-against-nvidia-nvl72/
https://semianalysis.com/2025/04/16/huawei-ai-cloudmatrix-384-chinas-answer-to-nvidia-gb200-nvl72/
U.S. export controls on Nvidia H20 AI chips enables Huawei’s 910C GPU to be favored by AI tech giants in China
Huawei’s “FOUR NEW strategy” for carriers to be successful in AI era
FT: Nvidia invested $1bn in AI start-ups in 2024
