AI
AI wireless and fiber optic network technologies; IMT 2030 “native AI” concept
To date, the main benefit of AI for telecom has been to reduce headcount/layoff employees. Light Reading’s Iain Morris wrote, “Telecom operators and vendors, nevertheless, are already using AI as the excuse for thousands of job cuts made and promised. So far, those cuts have not brought any improvement in the sector’s fortunes. Meanwhile, ceding basic but essential skills to systems that hardly anyone understands seems incredibly risky.” Some say that will change with 6G/ IMT 2030, but that’s a long way off. Others point to AI RAN, but that has not gotten any real market traction with wireless telcos.
As Gen AI development accelerates, robust wireless and fiber optic network infrastructure will be essential to accommodate the substantial data and communication volume generated by AI systems. Initially, the existing network ecosystem—encompassing wireless, wireline, broadband, and satellite services—will absorb this traffic load. However, the expanding requirements of AI are anticipated to drive the future emergence of entirely new network architectures and communication paradigms.
For sure, AI needs massive, fast, reliable connectivity to function, driving demand for low latency optical networks and 6G/ IMT 2030, which AI itself will optimize, leading to better efficiency, security, resource management, and new services like real-time AR/VR, ultimately boosting telecom revenue and innovation across the entire digital ecosystem.
Source: Pitinan Piyavatin/Alamy Stock Photo
……………………………………………………………………………………………………………………………………………………………………..
- AI Backend Scale-Out and Scale-Up Networks: These are specialized, private networks within and across data centers designed to connect numerous GPUs and enable them to function as one massive compute resource. They utilize technologies like:
- InfiniBand: A long-standing high-bandwidth, low-latency technology that has become a top choice for connecting GPU clusters in AI training environments.
- Optimized Ethernet: Ethernet is gaining ground for AI workloads through the development of enhanced, open standards via the Ultra Ethernet Consortium (UEC). These enhancements aim to provide lossless, low-latency fabrics that can match or exceed InfiniBand’s performance at scale.
- High-Speed Optics: The use of 400 Gbps and 800 Gbps (and soon 1.6 Tbps) optical interconnects is critical for meeting the massive bandwidth and power requirements within and between AI data centers.
- Edge AI Networking: As AI inferencing (generating responses from AI models) moves closer to the end-user or device (e.g., in autonomous vehicles, smart hospitals, or factories), specialized edge networks are needed. These networks must ensure low latency and localized processing to enable real-time responses.
- AI-Native 6G Networks: The upcoming sixth-generation (6G) wireless networks are being designed with AI integration as a core principle, rather than an add-on.
- These networks are expected to be fully automated and self-evolving, using AI to optimize resource allocation, predict issues, and enhance security autonomously.
- They will support extremely high data rates (up to 1 Tbps), ultra-low latency (around 1 ms), and new technologies like AI-RAN (Radio Access Network) that integrate AI capabilities directly into the network infrastructure.
- More in next section below.
- Self-Evolving Networks: The ultimate goal is the development of “self-evolving networks” where AI agents manage and optimize the network infrastructure autonomously, adapting to new demands and challenges without human intervention.
……………………………………………………………………………………………………………………………………………………………………..
In IMT 2030/6G networks, AI will shift from being an “add-on” optimization tool (as in 5G) to a native, foundational component of the entire network architecture. This deep integration will enable the network to be self-organizing, highly efficient, and capable of supporting advanced AI applications as a service. Native AI for IMT-2030 (6G) means building AI directly into the network’s core architecture, making it AI-first and pervasive, rather than adding AI as an overlay; this enables self-optimizing, intelligent networks that can autonomously manage resources, provide ubiquitous AI services, and offer seamless, context-aware experiences with minimal human intervention, fundamentally transforming both network operations and user applications by 2030.
- Ubiquitous Intelligence: Embedding AI everywhere, enabling distributed intelligence for AI model training, inference, and deployment directly within the network infrastructure, extending to the network edge.
- Autonomous Operations: AI handles complex tasks like network optimization, resource allocation, and automated maintenance (O&M) in real-time, reducing reliance on manual intervention.
- AI-as-a-Service (AIaaS): The network transforms into a unified platform providing both communication and AI capabilities, making AI accessible for various applications.
- Intelligent Processing: AI drives functions across the air interface, resource management, and control planes for highly efficient operations.
- Data-Driven Automation: Leverages big data and real-time analytics to predict issues, optimize performance, and automate complex decision-making.
- Seamless User Experience: Moves beyond touchscreens to AI-driven interactions, offering more natural and contextual computing.
- Autonomous Operations: AI will enable self-monitoring, self-optimization, and self-healing networks, drastically reducing the need for human intervention in operation and maintenance (O&M).
- Dynamic Resource Management: ML algorithms will analyze massive amounts of network data in real-time to predict traffic patterns and user demands, dynamically allocating bandwidth, power, and computing resources to ensure optimal performance and energy efficiency.
- AI-Native Air Interface: AI/ML models will replace traditional, manually engineered signal processing blocks in the physical layer (e.g., channel estimation, beam management) to adapt dynamically to complex and time-varying wireless environments, improving spectral efficiency.
- Enhanced Security: AI will be critical for real-time threat detection and automated incident response across the hyper-connected 6G ecosystem, identifying anomalies and mitigating security risks that are not well understood by current systems.
- Digital Twins: AI will power the creation and management of real-time digital twins (virtual replicas) of the physical network, allowing for sophisticated simulations and testing of network changes before real-world deployment.
- Pervasive Edge AI: AI model training and inference will be distributed throughout the network, from the cloud to the edge (devices, base stations), reducing latency and enabling real-time, localized decision-making for applications like autonomous driving and industrial automation.
- Support for Advanced Use Cases: The massive data rates (up to 1 Tbps), ultra-low latency, and high reliability enabled by AI in 6G will facilitate new applications such as holographic communication, remote robotic surgery with haptic feedback, and collaborative robotics that were not feasible with 5G.
- Federated Learning: The network will support distributed machine learning techniques, such as federated learning, which allow AI models to be trained on local data across various devices without the need to centralize sensitive user data, thus ensuring data privacy and security.
- Integrated Sensing and Communication (ISAC): AI will process the rich environmental data gathered through 6G’s new sensing capabilities (e.g., precise positioning, motion detection, environmental monitoring), allowing the network to interact with and understand the physical world in a holistic manner for applications like smart city management or augmented reality.
……………………………………………………………………………………………………………………………………………………………………..
AI‑native air interface and RAN:
IMT‑2030 explicitly expects a new AI‑native air interface that uses AI/ML models for core PHY/MAC functions such as channel estimation, symbol detection/decoding, beam management, interference handling, and CSI feedback. This enables adaptive waveforms and link control that react in real time to channel and traffic conditions, going beyond deterministic algorithms in 5G‑Advanced.
At the RAN level, IMT‑2030 envisions “native‑AI enabled” architectures that are simpler but more intelligent, with data‑driven operation and distributed learning across gNBs, edge nodes, and devices. AI/ML will be applied end‑to‑end for resource allocation, mobility, energy optimization, and fault management, effectively turning the RAN into a self‑optimizing, self‑healing system.
Integrated AI and communication services:
The framework defines “Artificial Intelligence and Communication” (often phrased as Integrated AI and Communication) as a specific usage scenario where the network provides AI compute, model hosting, and inference as a service. Example use cases include IMT‑2030‑assisted automated driving, cooperative medical robotics, digital twins, and offloading heavy computation from devices to edge/cloud via the 6G network.
To support this, IMT‑2030 includes “applicable AI‑related capabilities” such as distributed data processing, distributed learning, AI model execution and inference, and AI‑aware scheduling as native capabilities of the system. Computing and data services (not just connectivity) are treated as integral IMT‑2030 components, especially at the edge for low‑latency, energy‑efficient AI workloads.
System intelligence and new use cases:
AI is central to several new IMT‑2030 usage scenarios beyond classic eMBB/mMTC/URLLC, including Immersive Communication, Integrated Sensing and Communication, and Integrated AI and Communication. In integrated sensing, AI fuses multi‑dimensional radio sensing data (position, motion, environment, even human behavior) to provide contextual awareness for applications like smart cities, industrial control, and XR.
Embedding intelligence across air interface, edge, and cloud is seen as necessary to manage 6G complexity and enable “Intelligence of Everything,” including real‑time digital twins and AIGC‑driven services. The vision is for the 6G/IMT‑2030 network to act as a distributed neural system that tightly couples communication, sensing, and computing.
IMT 2030 Goals:
- To create self-healing, self-optimizing networks that can adapt to diverse demands.
- To enable new AI-driven applications, from intelligent digital twins to advanced immersive experiences.
- To build a truly intelligent communication fabric that supports a hyper-connected, AI-enhanced world.
Summary table: AI’s roles in IMT‑2030:
| Dimension | AI role in IMT‑2030 |
|---|---|
| Air interface | AI‑native PHY/MAC for channel estimation, decoding, beamforming, interference control. |
| RAN/core architecture | Native‑AI enabled, data‑driven, self‑optimizing/self‑healing network functions. |
| Compute and data services | Built‑in edge/cloud compute for AI training, inference, and data processing. |
| Usage scenarios | Dedicated “Integrated AI and Communication” plus AI‑rich sensing and immersive use cases. |
| Applications and ecosystems | Support for digital twins, automated driving, robotics, AIGC, and industrial automation. |
In summary, AI in IMT‑2030 is both an internal engine for network intelligence and an exported capability the network offers to verticals, making 6G effectively AI‑native end‑to‑end.
………………………………………………………………………………………………………………………………………………
References:
https://www.lightreading.com/ai-machine-learning/the-lessons-of-pluribus-for-telecom-s-genai-fans
https://www.ericsson.com/en/reports-and-papers/white-papers/ai-native
https://www.5gamericas.org/wp-content/uploads/2024/08/ITUs-IMT-2030-Vision_Id.pdf
ITU-R WP 5D Timeline for submission, evaluation process & consensus building for IMT-2030 (6G) RITs/SRITs
ITU-R WP 5D reports on: IMT-2030 (“6G”) Minimum Technology Performance Requirements; Evaluation Criteria & Methodology
Ericsson and e& (UAE) sign MoU for 6G collaboration vs ITU-R IMT-2030 framework
Nokia and Rohde & Schwarz collaborate on AI-powered 6G receiver years before IMT 2030 RIT submissions to ITU-R WP5D
NTT DOCOMO successful outdoor trial of AI-driven wireless interface with 3 partners
Verizon’s 6G Innovation Forum joins a crowded list of 6G efforts that may conflict with 3GPP and ITU-R IMT-2030 work
ITU-R WP5D IMT 2030 Submission & Evaluation Guidelines vs 6G specs in 3GPP Release 20 & 21
Dell’Oro: Analysis of the Nokia-NVIDIA-partnership on AI RAN
Highlights of 3GPP Stage 1 Workshop on IMT 2030 (6G) Use Cases
Draft new ITU-R recommendation (not yet approved): M.[IMT.FRAMEWORK FOR 2030 AND BEYOND]
Analysis: OpenAI and Deutsche Telekom launch multi-year AI collaboration
Deutsche Telekom (DT) has formalized a strategic, multi-year collaboration with OpenAI to integrate advanced artificial intelligence (AI) solutions across its internal operations and customer engagement platforms. The partnership aims to co-develop “simple, personal, and multi-lingual AI experiences” focused on enhancing communication and productivity. Initial pilot programs are slated for deployment in Q1 2026. AI will also play a larger role in customer care, internal copilots, and network operations as the Group advances toward more autonomous, self-healing networks.DT plans a company-wide rollout of ChatGPT Enterprise, leveraging AI to streamline core functions including:
- Customer Care: Deploying sophisticated virtual assistants to manage billing inquiries, service outages, plan modifications, roaming support, and device troubleshooting [1].
- Internal Operations: Utilizing AI copilots to increase internal efficiency.
- Network Management: Optimizing core network provisioning and operations.
- Sovereign Cloud (2021): DT’s T-Systems division partnered with Google Cloud to offer sovereign cloud services.
- T Cloud Suite (Early 2025): The launch of a comprehensive suite providing sovereign public, private, and AI cloud options leveraging hybrid infrastructure.
- Industrial AI Cloud (Early 2025): A collaboration with Nvidia to build a dedicated industrial AI data center in Munich, scheduled for Q1 2026 operations.
- Edge AI compute services for enterprises.
- Vertical AI solutions tailored for healthcare, retail, and manufacturing sectors.
- Integrated private 5G and AI bundles for industrial logistical hubs.
“Telcos – if they execute – will have a big play in the edge inferencing space as well as providing hosting and colo services that can host domain specific SLMs that need to be run closer to the user data,” he said. “Furthermore, telcos will play a role in connectivity services across Neocloud providers such as CoreWeave, Lambda Labs, Digital Ocean, Vast.AI etc. OpenAI does not want to lose the opportunity to partner with telcos so they are striking early,” Nag added.
Other Voices:
- Roger Entner notes the model is highly applicable to European incumbents (e.g., Orange, Telefonica) due to the relative scarcity of existing AI data centers in the region, allowing operators to fill a critical infrastructure gap. Conversely, the model is less viable for U.S. operators, where hyperscalers already dominate the extensive data center market.
- AvidThink Founder and colleague Roy Chua cautions that while DT presents a robust “reference blueprint,” replicating this strategy requires significant scale, substantial financial investment, and regulatory alignment—factors not easily accessible to all network operators.
- Futurum Group VP and Practice Lead Nick Patience told Fierce Network, “This deal elevates DT from being a user of AI to being a co-developer, which is pretty significant. DT is one of the few operators building a full-stack AI story. This is an example of OpenAI treating telcos as high-scale distribution and data channels – customer care, billing, network telemetry, national reach and government relationships. This suggests OpenAI is deliberately building an operator channel in key regions (U.S., Korea, EU) but still in partnership with existing cloud and infra providers rather than displacing them.”
OpenAI has established significant partnerships with several telecom network providers and related technology companies to integrate AI into network operations, enhance customer experience, and develop new AI-native platforms. Those deals and collaborations include:
- T-Mobile: T-Mobile has a multi-year agreement with OpenAI and is actively testing the integration of AI (specifically IntentCX) into its business operations for customer service improvements. T-Mobile is also collaborating with Nokia and Nvidia on AI-RAN (Radio Access Network) technologies for 6G innovation.
- SK Telecom (SKT): SK Telecom has an in-house AI company and collaborates with OpenAI and other AI leaders like Anthropic to enhance its AI capabilities, build sovereign AI infrastructure, and explore new services for its customers in South Korea and globally. They are also reportedly integrating Perplexity into their offerings.
- Deutsche Telekom (DT): DT is partnering with OpenAI to offer ChatGPT Enterprise across its business to help teams work more effectively, improve customer service, and automate network operations.
- Circles: This global telco technology company and OpenAI announced a strategic global collaboration to build a fully AI-native telco SaaS platform, which will first launch in Singapore. The platform aims to revolutionize the consumer experience and drive operational efficiencies for telcos worldwide.
- Rakuten: Rakuten and OpenAI launched a strategic partnership to develop AI tools and a platform aimed at leveraging Rakuten’s Open RAN expertise to revolutionize the use of AI in telecommunications.
- Orange: Orange is working with OpenAI to drive new use cases for enterprise needs, manage networks, and enable innovative customer care solutions, including those that support African regional languages.
- Indian Telecoms (Reliance Jio, Airtel): Telecom providers in India are integrating AI tools from companies like Google and Perplexity into their mobile subscriptions, providing millions of users access to advanced intelligence resources.
- Nokia & Nvidia: In a broader industry collaboration, Nvidia invested $1 billion in Nokia to add Nvidia-powered AI-RAN products to Nokia’s portfolio, enabling telecom service providers to launch AI-native 5G-Advanced and 6G networks. This partnership also includes T-Mobile US for testing.
Conclusions:
With more than 261 million mobile customers globally, Deutsche Telekom provides a strong foundation to bring AI into everyday use at scale. The new collaboration marks the next step in Deutsche Telekom’s AI journey – moving from early pilots to large-scale products that make AI useful for everyone
References:
https://www.telekom.com/en/media/media-information/archive/openai-and-telekom-collaborate-1100164
https://www.telekom.com/en/company/companyprofile/company-profile-625808
Deutsche Telekom: successful completion of the 6G-TakeOff project with “3D networks”
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
Open AI raises $8.3B and is valued at $300B; AI speculative mania rivals Dot-com bubble
OpenAI and Broadcom in $10B deal to make custom AI chips
Custom AI Chips: Powering the next wave of Intelligent Computing
OpenAI orders HBM chips from SK Hynix & Samsung for Stargate UAE project
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
Reuters & Bloomberg: OpenAI to design “inference AI” chip with Broadcom and TSMC
Custom AI Chips: Powering the next wave of Intelligent Computing
by the Indxx team of market researchers with Alan J Weissberger
The Market for AI Related Semiconductors:
Several market research firms and banks forecast that revenue from AI-related semiconductors will grow at about 18% annually over the next few years—five times faster than non-AI semiconductor market segments.
- IDC forecasts that global AI hardware spending, including chip demand, will grow at an annual rate of 18%.
- Morgan Stanley analysts predict that AI-related semiconductors will grow at an 18% annual rate for a specific company, Taiwan Semiconductor (TSMC).
- Infosys notes that data center semiconductor sales are projected to grow at an 18% CAGR.
- MarketResearch.biz and the IEEE IRDS predict an 18% annual growth rate for AI accelerator chips.
- Citi also forecasts aggregate chip sales for potential AI workloads to grow at a CAGR of 18% through 2030.
AI-focused chips are expected to represent nearly 20% of global semiconductor demand in 2025, contributing approximately $67 billion in revenue [1]. The global AI chip market is projected to reach $40.79 billion in 2025 [2.] and continue expanding rapidly toward $165 billion by 2030.
…………………………………………………………………………………………………………………………………………………
Types of AI Custom Chips:
Artificial intelligence is advancing at a speed that traditional computing hardware can no longer keep pace with. To meet the demands of massive AI models, lower latency, and higher computing efficiency, companies are increasingly turning to custom AI chips which are purpose-built processors optimized for neural networks, training, and inference workloads.
Those AI chips include Application Specific Integrated Circuits (ASICs) and Field- Programmable Gate Arrays (FPGAs) to Neural Processing Units (NPUs) and Google’s Tensor Processing Units (TPUs). They are optimized for core AI tasks like matrix multiplications and convolutions, delivering far higher performance-per-watt than CPUs or GPUs. This efficiency is key as AI workloads grow exponentially with the rise of Large Language Models (LLMs) and generative AI.
OpenAI – Broadcom Deal:
Perhaps the biggest custom AI chip design is being done by an OpenAI partnership with Broadcom in a multi-year, multi-billion dollar deal announced in October 2025. In this arrangement, OpenAI will design the hardware and Broadcom will develop custom chips to integrate AI model knowledge directly into the silicon for efficiency.
Here’s a summary of the partnership:
- OpenAI designs its own AI processors (GPUs) and systems, embedding its AI insights directly into the hardware. Broadcom develops and deploys these custom chips and the surrounding infrastructure, using its Ethernet networking solutions to scale the systems.
- Massive Scale: The agreement covers 10 gigawatts (GW) of AI compute, with deployments expected over four years, potentially extending to 2029.
- Cost Savings: This custom silicon strategy aims to significantly reduce costs compared to off-the-shelf Nvidia or AMD chips, potentially saving 30-40% on large-scale deployments.
- Strategic Goal: The collaboration allows OpenAI to build tailored hardware to meet the intense demands of developing frontier AI models and products, reducing reliance on other chip vendors.
AI Silicon Market Share of Key Players:
- Nvidia, with its extremely popular AI GPUs and CUDA software ecosystem., is expected to maintain its market leadership. It currently holds an estimated 86% share of the AI GPU market segment according to one source [2.]. Others put NVIDIA’s market AI chip market share between 80% and 92%.
- AMD holds a smaller, but growing, AI chip market share, with estimates placing its discrete GPU market share around 4% to 7% in early to mid-2025. AMD is projected to grow its AI chip division significantly, aiming for a double-digit share with products like the MI300X. In response to the extraordinary demand for advanced AI processors, AMD’s Chief Executive Officer, Dr. Lisa Su, presented a strategic initiative to the Board of Directors: to pivot the company’s core operational focus towards artificial intelligence. Ms. Su articulated the view that the “insatiable demand for compute” represented a sustained market trend. AMD’s strategic reorientation has yielded significant financial returns; AMD’s market capitalization has nearly quadrupled, surpassing $350 billion [1]. Furthermore, the company has successfully executed high-profile agreements, securing major contracts to provide cutting-edge silicon solutions to key industry players, including OpenAI and Oracle.
- Intel accounts for approximately 1% of the discrete GPU market share, but is focused on expanding its presence in the AI training accelerator market with its Gaudi 3 platform, where it aims for an 8.7% share by the end of 2025. The former microprocessor king has recently invested heavily in both its design and manufacturing businesses and is courting customers for its advanced data-center processors.
- Qualcomm, which is best known for designing chips for mobile devices and cars, announced in October that it would launch two new AI accelerator chips. The company said the new AI200 and AI250 are distinguished by their very high memory capabilities and energy efficiency.
Big Tech Custom AI chips vs Nvidia AI GPUs:
Big tech companies, including Google, Meta, Amazon, and Apple—are designing their own custom AI silicon to reduce costs, accelerate performance, and scale AI across industries. Yet nearly all rely on TSMC for manufacturing, thanks to its leadership in advanced chip fabrication technology [3.]
- Google recently announced Ironwood, its 7th-generation Tensor Processing Unit (TPU), a major AI chip for LLM training and inference, offering 4x the performance of its predecessor (Trillium) and massive scalability for demanding AI workloads like Gemini, challenging Nvidia’s dominance by efficiently powering complex AI at scale for Google Cloud and major partners like Meta. Ironwood is significantly faster, with claims of over 4x improvement in training and inference compared to the previous Trillium (6th gen) TPU. It allows for super-pods of up to 9,216 interconnected chips, enabling huge computational power for cutting-edge models. It’s optimized for high-volume, low-latency AI inference, handling complex thinking models and real-time chatbots efficiently.
- Meta is in advanced talks to purchase and rent large quantities of Google’s custom AI chips (TPUs), starting with cloud rentals in 2026 and moving to direct purchases for data centers in 2027, a significant move to diversify beyond Nvidia and challenge the AI hardware market. This multi-billion dollar deal could reshape AI infrastructure by giving Meta access to Google’s specialized silicon for workloads like AI model inference, signaling a major shift in big tech’s chip strategy, notes this TechRadar article.
- According to a Wall Street Journal report published on December 2, 2025, Amazon’s new Trainium3 custom AI chip presents a challenge to Nvidia’s market position by providing a more affordable option for AI development. Four times as fast as its previous generation of AI chips, Amazon said Trainium3 (produced by AWS’s Annapurna Labs custom-chip design business) can reduce the cost of training and operating AI models by up to 50% compared with systems that use equivalent graphics processing units, or GPUs. AWS acquired Israeli startup Annapurna Labs in 2015 and began designing chips to power AWS’s data-center servers, including network security chips, central processing units, and later its AI processor series, known as Inferentia and Trainium. “The main advantage at the end of the day is price performance,” said Ron Diamant, an AWS vice president and the chief architect of the Trainium chips. He added that his main goal is giving customers more options for different computing workloads. “I don’t see us trying to replace Nvidia,” Diamant said.
- Interestingly, many of the biggest buyers of Amazon’s chips are also Nvidia customers. Chief among them is Anthropic, which AWS said in late October is using more than one million Trainium2 chips to build and deploy its Claude AI model. Nvidia announced a month later that it was investing $10 billion in Anthropic as part of a massive deal to sell the AI firm computing power generated by its chips.
Image Credit: Emil Lendof/WSJ, iStock
Other AI Silicon Facts and Figures:
- Edge AI chips are forecast to reach $13.5 billion in 2025, driven by IoT and smartphone integration.
- AI accelerators based on ASIC designs are expected to grow by 34% year-over-year in 2025.
- Automotive AI chips are set to surpass $6.3 billion in 2025, thanks to advancements in autonomous driving.
- Google’s TPU v5p reached 30% faster matrix math throughput in benchmark tests.
- U.S.-based AI chip startups raised over $5.1 billion in venture capital in the first half of 2025 alone.
Conclusions:
Custom silicon is now essential for deploying AI in real-world applications such as automation, robotics, healthcare, finance, and mobility. As AI expands across every sector, these purpose-built chips are becoming the true backbone of modern computing—driving a hardware race that is just as important as advances in software. More and more AI firms are seeking to diversify their suppliers by buying chips and other hardware from companies other than Nvidia. Advantages like cost-effectiveness, specialization, lower power consumption and strategic independence that cloud providers gain from developing their own in-house AI silicon. By developing their own chips, hyperscalers can create a vertically integrated AI stack (hardware, software, and cloud services) optimized for their specific internal workloads and cloud platforms. This allows them to tailor performance precisely to their needs, potentially achieving better total cost of ownership (TCO) than general-purpose Nvidia GPUs
However, Nvidia is convinced it will retain a huge lead in selling AI silicon. In a post on X, Nvida wrote that it was “delighted by Google’s success with its TPUs,” before adding that Nvidia “is a generation ahead of the industry—it’s the only platform that runs every AI model and does it everywhere computing is done.” The company said its chips offer “greater performance, versatility, and fungibility” than more narrowly tailored custom chips made by Google and AWS.
The race is far from over, but we can expect to surely see more competition in the AI silicon arena.
………………………………………………………………………………………………………………………………………………………………………………….
Links for Notes:
2. https://sqmagazine.co.uk/ai-chip-statistics/
3. https://www.ibm.com/think/news/custom-chips-ai-future
References:
https://www.wsj.com/tech/ai/amazons-custom-chips-pose-another-threat-to-nvidia-8aa19f5b
https://www.wsj.com/tech/ai/nvidia-ai-chips-competitors-amd-broadcom-google-amazon-6729c65a
AI infrastructure spending boom: a path towards AGI or speculative bubble?
OpenAI and Broadcom in $10B deal to make custom AI chips
Reuters & Bloomberg: OpenAI to design “inference AI” chip with Broadcom and TSMC
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
Cisco CEO sees great potential in AI data center connectivity, silicon, optics, and optical systems
Expose: AI is more than a bubble; it’s a data center debt bomb
China gaining on U.S. in AI technology arms race- silicon, models and research
AI infrastructure spending boom: a path towards AGI or speculative bubble?
by Rahul Sharma, Indxx with Alan J Weissberger, IEEE Techblog
Introduction:
The ongoing wave of artificial intelligence (AI) infrastructure investment by U.S. mega-cap tech firms marks one of the largest corporate spending cycles in history. Aggregate annual AI investments, mostly for cloud resident mega-data centers, are expected to exceed $400 billion in 2025, potentially surpassing $500 billion by 2026 — the scale of this buildout rivals that of past industrial revolutions — from railroads to the internet era.[1]
At its core, this spending surge represents a strategic arms race for computational dominance. Meta, Alphabet, Amazon and Microsoft are racing to secure leadership in artificial intelligence capabilities — a contest where access to data, energy, and compute capacity are the new determinants of market power.
AI Spending & Debt Financing:
Leading technology firms are racing to secure dominance in compute capacity — the new cornerstone of digital power:
- Meta plans to spend $72 billion on AI infrastructure in 2025.
- Alphabet (Google) has expanded its capex guidance to $91–93 billion.[3]
- Microsoft and Amazon are doubling data center capacity, while AWS will drive most of Amazon’s $125 billion 2026 investment.[4]
- Even Apple, typically conservative in R&D, has accelerated AI infrastructure spending.
Their capex is shown in the chart below:
Analysts estimate that AI could add up to 0.5% to U.S. GDP annually over the next several years. Reflecting this optimism, Morgan Stanley forecasts $2.9 trillion in AI-related investments between 2025 and 2028. The scale of commitment from Big Tech is reshaping expectations across financial markets, enterprise strategies, and public policy, marking one of the most intense capital spending cycles in corporate history.[2]
Meanwhile, OpenAI’s trillion-dollar partnerships with Nvidia, Oracle, and Broadcom have redefined the scale of ambition, turning compute infrastructure into a strategic asset comparable to energy independence or semiconductor sovereignty.[5]
Growth Engine or Speculative Bubble?
As Big Tech pours hundreds of billions of dollars into AI infrastructure, analysts and investors remain divided — some view it as a rational, long-term investment cycle, while others warn of a potential speculative bubble. Yet uncertainty remains — especially around Meta’s long-term monetization of AGI-related efforts.[8]
Some analysts view this huge AI spending as a necessary step towards achieving Artificial General Intelligence (AGI) – an unrealized type of AI that possesses human-level cognitive abilities, allowing it to understand, learn, and adapt to any intellectual task a human can. Unlike narrow AI, which is designed for specific functions like playing chess or image recognition, AGI could apply its knowledge to a wide range of different situations and problems without needing to be explicitly programmed for each one.
Other analysts believe this is a speculative bubble, fueled by debt that can never be repaid. Tech sector valuations have soared to dot-com era levels – and, based on price-to-sales ratios, are well beyond them. Some of AI’s biggest proponents acknowledge the fact that valuations are overinflated, including OpenAI chairman Bret Taylor: “AI will transform the economy… and create huge amounts of economic value in the future,” Taylor told The Verge. “I think we’re also in a bubble, and a lot of people will lose a lot of money,” he added.
Here are a few AI bubble points and charts:
- AI-related capex is projected to consume up to 94% of operating cash flows by 2026, according to Bank of America.[6]
- Over $75 billion in AI-linked corporate bonds have been issued in just two months — a signal of mounting leverage. Still, strong revenue growth from AI services (particularly cloud and enterprise AI) keeps optimism alive.[7]
- Meta, Google, Microsoft, Amazon and xAI (Elon Musk’s company) are all using off-balance-sheet debt vehicles, including special-purpose vehicles (SPVs) to fund part of their AI investments. A slowdown in AI demand could render the debt tied to these SPVs worthless, potentially triggering another financial crisis.
- Alphabet’s (Google’s parent company) CEO Sundar Pichai sees “elements of irrationality” in the current scale of AI investing which is much more than excessive investments during the dot-com/fiber optic built-out boom of the late 1990s. If the AI bubble bursts, Pichai said that no company will be immune, including Alphabet, despite its breakthrough technology, Gemini, fueling gains in the company’s stock price.
…………………………………………………………………………………………………………………..
From Infrastructure to Intelligence:
Executives justify the massive spend by citing acute compute shortages and exponential demand growth:
- Microsoft’s CFO Amy Hood admitted, “We’ve been short on capacity for many quarters” and confirmed that the company will increase its spending on GPUs and CPUs in 2026 to meet surging demand.
- Amazon’s Andy Jassy noted that “every new tranche of capacity is immediately monetized”, underscoring strong and sustained demand for AI and cloud services.
- Google reported billions in quarterly AI revenue, offering early evidence of commercial payoff.
Macro Ripple Effects – Industrializing Intelligence:
AI data centers have become the factories of the digital age, fueling demand for:
- Semiconductors, especially GPUs (Nvidia, AMD, Broadcom)
- Cloud and networking infrastructure (Oracle, Cisco)
- Energy and advanced cooling systems for AI data centers (Vertiv, Schneider Electric, Johnson Controls, and other specialists such as Liquid Stack and Green Revolution Cooling).
| Company Name | Core Expertise | Key Solutions for AI Data Centers |
|---|---|---|
| Vertiv | Critical infrastructure (power & cooling) | Offers full-stack solutions with air and liquid cooling, power distribution units (PDUs), and monitoring systems, including the AI-ready Vertiv 360AI portfolio. |
| Schneider Electric | Energy management & automation | Provides integrated power and thermal management via its EcoStruxure platform, specializing in modular and liquid cooling solutions for HPC and AI applications. |
| Johnson Controls | HVAC & building solutions | Offers integrated, energy-efficient solutions from design to maintenance, including Silent-Aire cooling and YORK chillers, with a focus on large-scale operations. |
| Eaton | Power management | Specializes in electrical distribution systems, uninterruptible power supplies (UPS), and switchgear, which are crucial for reliable energy delivery to high-density AI racks. |
- LiquidStack: A leader in two-phase and modular immersion cooling and direct-to-chip systems, trusted by large cloud and hardware providers.
- Green Revolution Cooling (GRC): Pioneers in single-phase immersion cooling solutions that help simplify thermal management and improve energy efficiency.
- Iceotope: Focuses on chassis-level precision liquid cooling, delivering dielectric fluid directly to components for maximum efficiency and reduced operational costs.
- Asetek: Specializes in direct-to-chip (D2C) liquid cooling solutions and rack-level Coolant Distribution Units (CDUs) for high-performance computing.
- CoolIT Systems: Known for its custom direct liquid cooling technologies, working closely with server OEMs (Original Equipment Manufacturers) to integrate cold plates and CDUs for AI and HPC workloads.
–>This new AI ecosystem is reshaping global supply chains — but also straining local energy and water resources. For example, Meta’s massive data center in Georgia has already triggered environmental concerns over energy and water usage.
Global Spending Outlook:
- According to UBS, global AI capex will reach $423 billion in 2025, $571 billion by 2026 and $1.3 trillion by 2030, growing at a 25% CAGR during the period 2025-2030.
Compute demand is outpacing expectations, with Google’s Gemini saw 130 times rise in AI token usage over the past 18 months, highlighting soaring compute and Meta’s infrastructure needs expanding sharply.[9]
Conclusions:
The AI infrastructure boom reflects a bold, forward-looking strategy by Big Tech, built on the belief that compute capacity will define the next decade’s leaders. If Artificial General Intelligence (AGI) or large-scale AI monetization unfolds as expected, today’s investments will be seen as visionary and transformative. Either way, the AI era is well underway — and the race for computational excellence is reshaping the future of global markets and innovation.
…………………………………………………………………………………………………………………………………………………………………………………………………………………………….
Footnotes:
[1] https://www.investing.com/news/stock-market-news/ai-capex-to-exceed-half-a-trillion-in-2026-ubs-4343520?utm_medium=feed&utm_source=yahoo&utm_campaign=yahoo-www
[2] https://www.venturepulsemag.com/2025/08/01/big-techs-400-billion-ai-bet-the-race-thats-reshaping-global-technology/#:~:text=Big%20Tech’s%20$400%20Billion%20AI%20Bet:%20The%20Race%20That’s%20Reshaping%20Global%20Technology,-3%20months%20ago&text=The%20world’s%20largest%20technology%20companies,enterprise%20strategy%2C%20and%20public%20policy.
[3] https://www.businessinsider.com/big-tech-capex-spending-ai-earnings-2025-10?
[4] https://www.investing.com/analysis/meta-plunged-12-amazon-jumped-11–same-ai-race-different-economics-200669410
[5] https://www.cnbc.com/2025/10/15/a-guide-to-1-trillion-worth-of-ai-deals-between-openai-nvidia.html
[6] https://finance.yahoo.com/news/bank-america-just-issued-stark-152422714.html
[7] https://news.futunn.com/en/post/64706046/from-cash-rich-to-collective-debt-how-does-wall-street?level=1&data_ticket=1763038546393561
[8] https://www.businessinsider.com/big-tech-capex-spending-ai-earnings-2025-10?
[9] https://finance.yahoo.com/news/ai-capex-exceed-half-trillion-093015889.html
……………………………………………………………………………………………………………………………………………………………………………………………………………………………
About the Author:
Rahul Sharma is President & Co-Chief Executive Officer at Indxx – a provider of end-to-end indexing services, data and technology products. He has been instrumental in leading the firm’s growth since 2011. Raul manages Indxx’s Sales, Client Engagement, Marketing and Branding teams while also helping to set the firm’s overall strategic objectives and vision.
Rahul holds a BS from Boston College and an MBA with Beta Gamma Sigma honors from Georgetown University’s McDonough School of Business.
……………………………………………………………………………………………………………………………………………………………………………………………………………………………
References:
Curmudgeon/Sperandeo: New AI Era Thinking and Circular Financing Deals
Expose: AI is more than a bubble; it’s a data center debt bomb
Can the debt fueling the new wave of AI infrastructure buildouts ever be repaid?
AI spending boom accelerates: Big tech to invest an aggregate of $400 billion in 2025; much more in 2026!
Big tech spending on AI data centers and infrastructure vs the fiber optic buildout during the dot-com boom (& bust)
FT: Scale of AI private company valuations dwarfs dot-com boom
Amazon’s Jeff Bezos at Italian Tech Week: “AI is a kind of industrial bubble”
AI Data Center Boom Carries Huge Default and Demand Risks
Will billions of dollars big tech is spending on Gen AI data centers produce a decent ROI?
Dell’Oro: Analysis of the Nokia-NVIDIA-partnership on AI RAN
RAN silicon rethink – from purpose built products & ASICs to general purpose processors or GPUs for vRAN & AI RAN
Nokia in major pivot from traditional telecom to AI, cloud infrastructure, data center networking and 6G
Reuters: US Department of Energy forms $1 billion AI supercomputer partnership with AMD
………………………………………………………………………………………………………………………………………………………………………….
GSMA, ETSI, IEEE, ITU & TM Forum: AI Telco Troubleshooting Challenge + TelecomGPT: a dedicated LLM for telecom applications
The GSMA — along with ETSI, IEEE GenAINet, the ITU, and TM Forum — today opened an innovation challenge calling on telco operators, AI researchers, and startups to build large-language models (LLMs) capable of root-cause analysis (RCA) for telecom network faults. The AI Telco Troubleshooting Challenge is supported by Huawei, InterDigital, NextGCloud, RelationalAI, xFlowResearch and technical advisors from AT&T.
The new competition invites teams to submit AI models in three categories: Generalization to New Faults will assess the best performing LLMs for RCA; Small Models at the Edge will evaluate lightweight edge-deployable models; and Explainability/Reasoning will focus on the AI systems that clearly explain their reasoning. Additional categories will include securing edge-cloud deployments and enabling AI services for application developers.
The goal is to deliver AI tools that help operators automatically identify, diagnose, and (eventually) remediate network problems — potentially reducing both downtime and operational costs. This marks a concrete step toward turning “telco-AI” from pilot projects into operational infrastructure.
As telecom networks scale (5G, 5G-Advanced, edge, IoT), faults and failures become costlier. Automating fault detection and troubleshooting with AI could significantly boost network resilience, reduce manual labor, and enable faster recovery from outages.
“Large Language Models have become instrumental in the pursuit of autonomous, resilient and adaptive networks,” said Prof. Merouane Debbah, General Chair of IEEE GenAINet ETI. “Through this challenge, we are tackling core research and engineering challenges, such as generalisation to unseen network faults, interpretability and edge-efficient AI, that are vital for making AI-native telecom infrastructures a reality. IEEE GenAINet ETI is proud to support this initiative, which serves as a testbed for future-ready innovations across the global telco ecosystem.”
“ITU’s global AI challenges connect innovators with computing resources, datasets, and expert mentors to nurture AI innovation ecosystems worldwide,” said Seizo Onoe, Director of the ITU Telecommunication Standardization Bureau. “Crowdsourcing new solutions and creating conditions for them to scale, our challenges boost business by helping innovations achieve meaningful impact.”
“The future of telecoms depends on the autonomation of network resiliency – shifting from static infrastructure to AI-driven, context-aware, self-optimising networks. TM Forum’s AI-Native Blueprint provides the architectural foundation to make this reality, and the AI Telco Troubleshooting Challenge aligns perfectly to support the industry in moving beyond isolated pilots to production-grade resilient autonomation,” said Guy Lupo, AI and Data Mission lead at TM Forum.
The initiative builds on recent breakthroughs in applying AI to network operations, leveraging curated datasets such as TeleLogs and benchmarking frameworks developed by GSMA and its partners under the GSMA Open-Telco LLM Benchmarks community, which includes a leaderboard that highlights how various LLMs perform on telco-specific use cases.
“Network faults cost operators millions annually and root cause analysis is a critical pain point for operators,” said Louis Powell, Director of AI Technologies at GSMA. “By harnessing AI models capable of reasoning and diagnosing unseen faults, the industry can dramatically improve reliability and reduce operational costs. Through this challenge, we aim to accelerate the development of LLMs that combine reasoning, efficiency and scalability.”
“We are encouraged by the upside of this challenge after our team at AT&T fine-tuned a 4-billion-parameter small language model that topped all other evaluated models on the GSMA Open-Telco LLM Benchmarks (TeleLogs RCA task), including frontier models such as GPT-5, Claude Sonnet 4.5 and Grok-4,” said Andy Markus, Chief Data Officer at AT&T. “This challenge has the right mix of an important business problem and a technical opportunity, and we welcome the industry’s collaboration to take it to the next level.”
The AI Telco Troubleshooting Challenge is open for submissions on the 28th November and it closes on 1st February 2026, with the winners announced at a dedicated prize-giving session at MWC26 Barcelona.
…………………………………………………………………………………………………………………………………………………………………………
Separately, the GSMA Foundry and Khalifa University announced a strategic collaboration to develop “TelecomGPT,” a dedicated LLM for telecom applications, plus an Open-Telco Knowledge Graph based on 3GPP specifications.
-
These assets are intended to help the industry overcome limitations of general-purpose LLMs, which often struggle with telecom-specific technical contexts. PR Newswire+2Mobile World Live+2
-
The plan: make TelecomGPT and related knowledge tools available for operators, vendors and researchers to accelerate AI-driven telco innovations. PR Newswire+1
Why it matters: A specialized “telco-native” LLM could improve automation, operations, R&D and standardization efforts — for example, helping operators configure networks, analyze logs, or build AI-powered services. It represents a shift toward embedding AI more deeply into core telecom infrastructure and operations.
…………………………………………………………………………………………………………………………………………………………………………………..
About GSMA
The GSMA is a global organization unifying the mobile ecosystem to discover, develop and deliver innovation foundational to positive business environments and societal change. Our vision is to unlock the full power of connectivity so that people, industry, and society thrive. Representing mobile operators and organizations across the mobile ecosystem and adjacent industries, the GSMA delivers for its members across three broad pillars: Connectivity for Good, Industry Services and Solutions, and Outreach. This activity includes advancing policy, tackling today’s biggest societal challenges, underpinning the technology and interoperability that make mobile work, and providing the world’s largest platform to convene the mobile ecosystem at the MWC and M360 series of events.
We invite you to find out more at gsma.com
About ETSI
ETSI is one of only three bodies officially recognized by the European Union as a European Standards Organization (ESO). It is an independent, not-for-profit body dedicated to ICT standardisation. With over 900 member organizations from more than 60 countries across five continents, ETSI offers an open and inclusive environment for members representing large and small private companies, research institutions, academia, governments, and public organizations. ETSI supports the timely development, ratification, and testing of globally applicable standards for ICT‑enabled systems, applications, and services across all sectors of industry and society. More on: etsi.org
About IEEE GenAINet
The aim of the IEEE Large Generative AI Models in Telecom Emerging Technology Initiative (GenAINet ETI) is to create a dynamic platform of research and innovation for academics, researchers, and industry leaders to advance the research on large generative AI in Telecom, through collaborative efforts across various disciplines, including mathematics, information theory, wireless communications, signal processing, networking, artificial intelligence, and more. More on: https://genainet.committees.comsoc.org
About ITU
The International Telecommunication Union (ITU) is the United Nations agency for digital technologies, driving innovation for people and the planet with 194 Member States and a membership of over 1,000 companies, universities, civil society, and international and regional organizations. Established in 1865, ITU coordinates the global use of the radio spectrum and satellite orbits, establishes international technology standards, drives universal connectivity and digital services, and is helping to make sure everyone benefits from sustainable digital transformation, including the most remote communities. From artificial intelligence (AI) to quantum, from satellites and submarine cables to advanced mobile and wireless broadband networks, ITU is committed to connecting the world and beyond. Learn more: www.itu.int
About TM Forum
TM Forum is an alliance of over 800 organizations spanning the global connectivity ecosystem, including the world’s top ten Communication Service Providers (CSPs), top three hyperscalers and Network Equipment Providers (NEPs), vendors, consultancies and system integrators, large and small. We provide a place for our Members to collaborate, innovate, and deliver lasting change. Together, we are building a sustainable future for the industry in connectivity and beyond. To find out more, visit: www.tmforum.org
References:
The AI Telco Troubleshooting Challenge Launches to Transform Network Reliability
AI Telco Troubleshooting Challenge global launch webinar
GSMA Vision 2040 study identifies spectrum needs during the peak 6G era of 2035–2040
Gartner: Gen AI nearing trough of disillusionment; GSMA survey of network operator use of AI
Expose: AI is more than a bubble; it’s a data center debt bomb
We’ve previously described the tremendous debt that AI companies have assumed, expressing serious doubts that it will ever be repaid. This article expands on that by pointing out the huge losses incurred by the AI startup darlings and that AI poster child Open AI won’t have the cash to cover its costs 9which are greater than most analysts assume). Also, we quote from the Wall Street Journal, Financial Times, Barron’s, along with a dire forecast from the Center for Public Enterprise.
In Saturday’s print edition, The Wall Street Journal notes:
OpenAI and Anthropic are the two largest suppliers of generative AI with their chatbots ChatGPT and Claude, respectively, and founders Sam Altman and Dario Amodei have become tech celebrities.
What’s only starting to become clear is that those companies are also sinkholes for AI losses that are the flip side of chunks of the public-company profits.
OpenAI hopes to turn profitable only in 2030, while Anthropic is targeting 2028. Meanwhile, the amounts of money being lost are extraordinary.
It’s impossible to quantify how much cash flowed from OpenAI to big tech companies. But OpenAI’s loss in the quarter equates to 65% of the rise in underlying earnings of Microsoft, Nvidia, Alphabet, Amazon and Meta together. That ignores Anthropic, from which Amazon recorded a profit of $9.5B from its holding in the loss-making company in the quarter.
OpenAI committed to spend $250 billion more on Microsoft’s cloud and has signed a $300 billion deal with Oracle, $22 billion with CoreWeave and $38 billion with Amazon, which is a big investor in rival Anthropic.
OpenAI doesn’t have the income to cover its costs. It expects revenue of $13 billion this year to more than double to $30 billion next year, then to double again in 2027, according to figures provided to shareholders. Costs are expected to rise even faster, and losses are predicted to roughly triple to more than $40 billion by 2027. Things don’t come back into balance even in OpenAI’s own forecasts until total computing costs finally level off in 2029, allowing it to scrape into profit in 2030.
The losses at OpenAI that has helped boost the profits of Big Tech may, in fact, understate the true nature of the problem. According to the Financial Times:
OpenAI’s running costs may be a lot more than previously thought, and that its main backer Microsoft is doing very nicely out of their revenue share agreement.
OpenAI appears to have spent more than $12.4bn at Azure on inference compute alone in the last seven calendar quarters. Its implied revenue for the period was a minimum of $6.8bn. Even allowing for some fudging between annualised run rates and period-end totals, the apparent gap between revenues and running costs is a lot more than has been reported previously.
The apparent gap between revenues and running costs is a lot more than has been reported previously. If the data is accurate, then it would call into question the business model of OpenAI and nearly every other general-purpose LLM vendor.
Also, the financing needed to build out the data centers at the heart of the AI boom is increasingly becoming an exercise in creative accounting. The Wall Street Journal reports:
The Hyperion deal is a Frankenstein financing that combines elements of private-equity, project finance and investment-grade bonds. Meta needed such financial wizardry because it already issued a $30B bond in October that roughly doubled its debt load overnight.
Enter Morgan Stanley, with a plan to have someone else borrow the money for Hyperion. Blue Owl invested about $3 billion for an 80% private-equity stake in the data center, while Meta retained 20% for the $1.3 billion it had already spent. The joint venture, named Beignet Investor after the New Orleans pastry, got another $27 billion by issuing bonds that pay off in 2049, $18 billion of which Pimco purchased. That debt is on Beignet’s balance sheet, not Meta’s.
Dan Fuss, vice chairman of Loomis Sayles told Barrons: “We are good at taking credit risk,” Dan said, cheerfully admitting to having the scars to show for it. That is, he added, if they know the credit. But that’s become less clear with the recent spate of mind-bendingly complex megadeals, with myriad entities funding multibillion-dollar data centers. Fuss thinks current data-center deals are too speculative. The risk is too great and future revenue too uncertain. And yields aren’t enough to compensate, he concluded.
Increased wariness about monster hyper-scaler borrowings has sent the cost of insuring their debt against default soaring. Credit default swaps (CDS) more than doubled for Oracle since September, after it issued $18 billion in public bonds and took out a $38 billion private loan. CoreWeave’s CDS gapped higher this past week, mirroring the slide of the data-center company’s stock.
According to the Bank Credit Analyst (BCA), capex busts weigh on the economy, which further hits asset prices, the firm says. Following the dot-com bust, a housing bubble grew, which burst in the 2008-09 financial crisis. “It is far from certain that a new bubble will emerge (after the AI bubble bursts) this time around, in which case the resulting recession could be more severe than the one in 2001,” BCA notes.
………………………………………………………………………………………………………………………………………………
The widening gap between the expenditures needed to build out AI data centers and the cash flows generated by the products they enable creates a colossal risk which could crash asset values of AI companies. The Center for Public Enterprise reports that it’s “Bubble or Nothing.”
Should economic conditions in the tech sector sour, the burgeoning artificial intelligence (AI) boom may evaporate—and, with it, the economic activity associated with the boom in data center development.
Circular financing, or “roundabouting,” among so-called hyperscaler tenants—the leading tech companies and AI service providers—create an interlocking liability structure across the sector. These tenants comprise an incredibly large share of the market and are financing each others’ expansion, creating concentration risks for lenders and shareholders.
Debt is playing an increasingly large role in the financing of data centers. While debt is a quotidian aspect of project finance, and while it seems like hyperscaler tech companies can self-finance their growth through equity and cash, the lack of transparency in some recent debt-financed transactions and the interlocked liability structure of the sector are cause for concern.
If there is a sudden stop in new lending to data centers, Ponzi finance units ‘with cash flow shortfalls will be forced to try to make position by selling out position’—in other words to force a fire sale—which is ‘likely to lead to a collapse of asset values.’
The fact that the data center boom is threatened by, at its core, a lack of consumer demand and the resulting unstable investment pathways, is itself an ironic miniature of the U.S. economy as a whole. Just as stable investment demand is the linchpin of sectoral planning, stable aggregate demand is the keystone in national economic planning. Without it, capital investment crumbles.
……………………………………………………………………………………………………………..
Postscript (November 23, 2025):
In addition to cloud/hyperscaler AI spending, AI start-ups (especially OpenAI) and newer IT infrastructure companies (like Oracle) play a prominent role. It’s often a “scratch my back and I’ll scratch yours” type of deal. Let’s look at the “circular financing” arrangement between Nvidia and OpenAI where capital flows from Nvidia to OpenAI and then back to Nvidia. That ensures Nvidia a massive, long-term customer and providing OpenAI with the necessary capital and guaranteed access to critical, high-demand hardware. Here’s the scoop:
- Nvidia has agreed to invest up to $100 billion in OpenAI over time. This investment will be in cash, likely for non-voting equity shares, and will be made in stages as specific data center deployment milestones are met.
- OpenAIhas committed to building and deploying at least 10 gigawatts of AI data center capacity using Nvidia’s silicon and equipment, which will involve purchasing millions of Nvidia expensive GPU chips.
Here’s the Circular Flow of this deal:
- Nvidia provides a cash investment to OpenAI.
- OpenAI uses that capital (and potentially raises additional debt using the commitment as collateral) to build new data centers.
- OpenAI then uses the funds to purchase Nvidia GPUs and other data center infrastructure.
- The revenue from these massive sales flows back to Nvidia, helping to justify its soaring stock price and funding further investments.
What’s wrong with such an arrangement you ask? Anyone remember the dot-com/fiber optic boom and bust? Critics have drawn parallels to the “vendor financing” practices of the dot-com era, arguing these interconnected deals could create a “mirage of growth” and potentially an AI bubble, as the actual organic demand for the products is difficult to assess when companies are essentially funding their own sales.
However, supporters note that, unlike the dot-com bubble, these deals involve the creation of tangible physical assets (data centers and chips) and reflect genuine, booming demand for AI compute capacity although it’s not at all certain how they’ll be paid for.
There’s a similar cozy relationship with the $1B Nvidia invested in Nokia with the Finnish company now planning to ditch Marvell’s silicon and replace it by buying the more expensive, power hungry Nvidia GPUs for its wireless network equipment. Nokia, has only now become a strong supporter of Nvidia’s AI RAN (Radio Access Network), which has many telco skeptics.
………………………………………………………………………………………………………………………………………………….
References:
https://www.wsj.com/tech/ai/big-techs-soaring-profits-have-an-ugly-underside-openais-losses-fe7e3184
https://www.ft.com/content/fce77ba4-6231-4920-9e99-693a6c38e7d5
https://www.wsj.com/tech/ai/three-ai-megadeals-are-breaking-new-ground-on-wall-street-896e0023
Can the debt fueling the new wave of AI infrastructure buildouts ever be repaid?
AI Data Center Boom Carries Huge Default and Demand Risks
Big tech spending on AI data centers and infrastructure vs the fiber optic buildout during the dot-com boom (& bust)
AI spending boom accelerates: Big tech to invest an aggregate of $400 billion in 2025; much more in 2026!
Gartner: AI spending >$2 trillion in 2026 driven by hyperscalers data center investments
Amazon’s Jeff Bezos at Italian Tech Week: “AI is a kind of industrial bubble”
FT: Scale of AI private company valuations dwarfs dot-com boom
Indosat Ooredoo Hutchison, Nokia and Nvidia AI-RAN research center in Indonesia amongst telco skepticism
Indosat Ooredoo Hutchison (Indosat) Nokia, and Nvidia have officially launched the AI-RAN Research Centre in Surabaya, a strategic collaboration designed to advance AI-native wireless networks and edge AI applications across Indonesia. This collaboration, aims to support Indonesia’s digital transformation goals and its “Golden Indonesia Vision 2045.” The facility will allow researchers and engineers to experiment with combining Nokia’s RAN technologies with Nvidia’s accelerated computing platforms and Indosat’s 5G network.
According to the partners, the research facility will serve as a collaborative environment for engineers, researchers, and future digital leaders to experiment, learn, and co-create AI-powered solutions. Its work will centre on integrating Nokia’s advanced RAN technologies with Nvidia’s accelerated computing platforms and Indosat’s commercial 5G network. The three companies view the project as a foundation for AI-driven growth, with applications spanning education, agriculture, and healthcare.
The AI-RAN infrastructure enables high-performance software-defined RAN and AI workloads on a single platform, leveraging Nvidia’s Aerial RAN Computer 1 (ARC-1). The facility will also act as a distributed computing extension of Indosat’s sovereign AI Factory, a national AI platform powered by Nvidia, creating an “AI Grid” that connects datacentres and distributed 5G nodes to deliver intelligence closer to users.
Nezar Patria, vice minister of communication and digital affairs of the Republic of Indonesia said: “The inauguration of the AI-RAN Research Centre marks a concrete step in strengthening Indonesia’s digital sovereignty. The collaboration between the government, industry, and global partners such as Indosat, Nokia, and Nvidia demonstrates that Indonesia is not merely a user but also a creator of AI technology. This initiative supports the acceleration of the Indonesia Emas 2045 vision by building an inclusive, secure, and globally competitive AI ecosystem.”
Vikram Sinha, president director and CEO of Indosat Ooredoo Hutchison said: “As Indonesia accelerates its digital transformation, the AI-RAN Research Centre reflects Indosat’s larger purpose of empowering Indonesia. When connectivity meets compute, it creates intelligence, delivered at the edge, in a sovereign manner. This is how AI unlocks real impact, from personalised tutors for children in rural areas to precision farming powered by drones. Together with Nokia and Nvidia, we’re building the foundation for AI-driven growth that strengthens Indonesia’s digital future.”
From a network perspective, the project demonstrates how AI-RAN architectures can optimize wireless network performance, energy efficiency, and scalability through machine learning–based radio signal processing.
Ronnie Vasishta, senior vice president of telecom at Nvidia added: “The AI Grid is the biggest opportunity for telecom providers to make AI as ubiquitous as connectivity and distribute intelligence at scale by tapping into their nationwide wireless networks.”
Pallavi Mahajan, chief technology and AI officer at Nokia said: “This initiative represents a major milestone in our journey toward the future of AI-native networks by bringing AI-powered intelligence into the hands of every Indonesian.”
………………………………………………………………………………………………………………………………………………………..
Wireless Telcos are Skeptical about AI-RAN:
According to Light Reading, the AI RAN uptake is facing resistance from telcos. The problem is Nvidia’s AI GPUs are very costly and not GPUs power-efficient enough to reside in wireless base stations, central offices or even small telco data centers.
Nvidia references 300 watts for the power consumption of ARC-Pro, which is much higher than the peak of 40 watts that Qualcomm claimed more than two years ago for its own RAN silicon when supporting throughput of 24 Gbit/s. How ARC-Pro would measure up on a like-for-like basis in a commercial network is obviously unclear.
Nvidia also claims a Gbit/s-per-watt performance “on par with” today’s traditional custom silicon. Yet the huge energy consumption of GPU-filled telco data centers does not bear that out.
“Is there a case for a wide-area indiscriminate rollout? I am not sure,” said Verizon CTO Yago Tenorio, during the Brooklyn 6G Summit, another telecom event, last week. “It depends on the unit cost of the GPU, on the power efficiency of the GPU, and the main factor will always be just doing what’s best for the basestation. Don’t try to just overcomplicate the whole thing monetizing that platform, as there are easier ways to do it.”
“We have no way to justify a business case like that,” said Bernard Bureau, the vice president of wireless strategy for Canada’s Telus, at FYUZ. “Our COs [central offices] are not necessarily the best places to run a data center. It would mean huge investments in space and power upgrades for those locations, and we’ve got sunk investment that can be leveraged in our cell sites.”
Light Reading’s Iain Morris wrote, “Besides turning COs into data centers, operators would need to invest in fiber connections between those facilities and their masts.”
How much spectral efficiency can be gained by using Nvidia GPUs as RAN silicon?
“It’s debatable if it’s going to improve the spectral efficiency by 50% or even 100%. It depends on the case,” said Tenorio. Whatever the exact improvement, it would be “really good” and is something the industry needs, he told the audience.
In April, Nokia’s rival Ericsson said it had tested “AI-native” link adaptation, a RAN algorithm, in the network of Bell Canada without needing any GPU. “That’s an algorithm we have optimized for decades,” said Per Narvinger, the head of Ericsson’s mobile networks business group. “Despite that, through a large language model, but a really small one, we gained 10% of spectral efficiency.”
Before Nvidia invested in Nokia, the latter claimed to have sufficient AI and machine-learning capabilities in the custom silicon provided by Marvell Technology, its historical supplier of 5G base station chips.
Executives at Cohere Technology praises Nvidia’s investment in Nokia, seeing it as an important AI spur for telecom. Yet their own software does not run on Nvidia GPUs. It promises to boost spectral efficiency on today’s 5G networks, massively reducing what telcos would have to spend on new radios. It has won plaudits from Vodafone’s Pignatelli as well as Bell Canada and Telstra, both of which have invested in Cohere. The challenge is getting the kit vendors to accommodate a technology that could hurt their own sales. Regardless, Bell Canada’s recent field trials of Cohere have used a standard Dell server without GPUs.
Finally, if GPUs are so critical in AI for RAN, why has neither Ericsson or Samsung using Nvidia GPU’s in their RAN equipment?
Morris sums up:
“Currently, the AI-RAN strategy adopted by Nokia looks like a massive gamble on the future. “The world is developing on Nvidia,” Vasishta told Light Reading in the summer, before the company’s share price had gained another 35%. That vast and expanding ecosystem holds attractions for RAN developers bothered by the diminishing returns on investment in custom silicon.”
“Intel’s general-purpose chips and virtual RAN approach drew interest several years ago for all the same reasons. But Intel’s recent decline has made Nvidia shine even more brightly. Telcos might not have to worry. Nvidia is already paying a big 5G vendor (Nokia) to use its technology. For a company that is so outrageously wealthy, paying a big operator to deploy it would be the next logical step.
…………………………………………………………………………………………………………………………………………………
References:
https://capacityglobal.com/news/indosat-nokia-and-nvidia-launch-ai-ran-research-centre-in-indonesia/
https://www.telecoms.com/ai/indosat-nokia-and-nvidia-open-ai-ran-research-centre-in-indonesia
https://www.lightreading.com/5g/nokia-and-nvidia-s-ai-ran-plan-hits-telco-resistance
https://resources.nvidia.com/en-us-aerial-ran-computer-pro
Nvidia pays $1 billion for a stake in Nokia to collaborate on AI networking solutions
Dell’Oro: AI RAN to account for 1/3 of RAN market by 2029; AI RAN Alliance membership increases but few telcos have joined
Nvidia AI-RAN survey results; AI inferencing as a reinvention of edge computing?
Dell’Oro: RAN revenue growth in 1Q2025; AI RAN is a conundrum
The case for and against AI-RAN technology using Nvidia or AMD GPUs
AI RAN Alliance selects Alex Choi as Chairman
AI spending boom accelerates: Big tech to invest an aggregate of $400 billion in 2025; much more in 2026!
- Meta Platforms says it continues to experience capacity constraints as it simultaneously trains new AI models and supports existing product infrastructure. Meta CEO Mark Zuckerberg described an unsatiated appetite for more computing resources that Meta must work to fulfill to ensure it’s a leader in a fast-moving AI race. “We want to make sure we’re not underinvesting,” he said on an earnings call with analysts Wednesday after posting third-quarter results. Meta signaled in the earnings report that capital expenditures would be “notably larger” next year than in 2025, during which Meta expects to spend as much as $72 billion. He indicated that the company’s existing advertising business and platforms are operating in a “compute-starved state.” This condition persists because Meta is allocating more resources toward AI research and development efforts rather than bolstering existing operations.
- Microsoft reported substantial customer demand for its data-center-driven services, prompting plans to double its data center footprint over the next two years. Concurrently, Amazon is working aggressively to deploy additional cloud capacity to meet demand. Amy Hood, Microsoft’s chief financial officer, said: “We’ve been short [on computing power] now for many quarters. I thought we were going to catch up. We are not. Demand is increasing.” She further elaborated, “When you see these kinds of demand signals and we know we’re behind, we do need to spend.”
- Alphabet (Google’s parent company) reported that capital expenditures will jump from $85 billion to between $91 billion and $93 billion. Google CFO Anat Ashkenazi said the investments are already yielding returns: “We already are generating billions of dollars from AI in the quarter. But then across the board, we have a rigorous framework and approach by which we evaluate these long-term investments.”
- Amazon has not provided a specific total dollar figure for its planned AI investment in 2026. However, the company has announced it expects its total capital expenditures (capex) in 2026 to be even higher than its 2025 projection of $125 billion, with the vast majority of this spending dedicated to AI and related infrastructure for Amazon Web Services (AWS).
- Apple: Announced it is also increasing its AI investments, though its overall spending remains smaller in comparison to the other tech giants.
As big as the spending projections were this week, they look pedestrian compared with OpenAI, which has announced roughly $1 trillion worth of AI infrastructure deals of late with partners including Nvidia , Oracle and Broadcom.
Despite the big capex tax write-offs (due to the 2025 GOP tax act) there is a large degree of uncertainty regarding the eventual outcomes of this substantial AI infrastructure spending. The companies themselves, along with numerous AI proponents, assert that these investments are essential for machine-learning systems to achieve artificial general intelligence (AGI), a state where they surpass human intelligence.
Yet skeptics question whether investing billions in large-language models (LLMs), the most prevalent AI system, will ultimately achieve that objective. They also highlight the limited number of paying users for existing technology and the prolonged training period required before a global workforce can effectively utilize it.
During investor calls following the earnings announcements, analysts directed incisive questions at company executives. On Microsoft’s call, one analyst voiced a central market concern, asking: “Are we in a bubble?” Similarly, on the call for Google’s parent company, Alphabet, another analyst questioned: “What early signs are you seeing that gives you confidence that the spending is really driving better returns longer term?”
Bank of America (BofA) credit strategists Yuri Seliger and Sohyun Marie Lee write in a client note that capital spending by five of the Magnificent Seven megacap tech companies (Amazon.com, Alphabet, and Microsoft, along with Meta and Oracle) has been growing even faster than their prodigious cash flows. “These companies collectively may be reaching a limit to how much AI capex they are willing to fund purely from cash flows,” they write. Consensus estimates of AI capex suggest will climb to 94% of operating cash flows, minus dividends and share repurchases, in 2025 and 2026, up from 76% in 2024. That’s still less than 100% of cash flows, so they don’t need to borrow to fund spending, “but it’s getting close,” they add.
………………………………………………………………………………………………………………………………………………………………….
Google, which projected a rise in its full-year capital expenditures from $85 billion to a range of $91 billion to $93 billion, indicated that these investments were already proving profitable. Google’s Ashkenazi stated: “We already are generating billions of dollars from AI in the quarter. But then across the board, we have a rigorous framework and approach by which we evaluate these long-term investments.”
Microsoft reported that it expects to face capacity shortages that will affect its ability to power both its current businesses and AI research needs until at least the first half of the next year. The company noted that its cloud computing division, Azure, is absorbing “most of the revenue impact.”
Amazon informed investors of its expedited efforts to bring new capacity online, citing its ability to immediately monetize these investments.
“You’re going to see us continue to be very aggressive in investing capacity because we see the demand,” said Amazon Chief Executive Andy Jassy. “As fast as we’re adding capacity right now, we’re monetizing it.”
Meta’s chief financial officer, Susan Li, stated that the company’s capital expenditures—which have already nearly doubled from last year to $72 billion this year—will grow “notably larger” in 2026, though specific figures were not provided. Meta brought this year’s biggest investment-grade corporate bond deal to market, totaling some $30 billion, the latest in a parade of recent data-center borrowing.
Apple confirmed during its earnings call it is also increasing investments in AI . However, its total spending levels remain significantly lower compared to the outlays planned by the other major technology firms.
Skepticism and Risk:
While proponents argue the investments are necessary for AGI and offer a competitive advantage, skeptics question if huge spending (capex) on AI infrastructure and large-language models will achieve this goal and point to limited paying users for current AI technology. Meta CEO Zuckerberg addressed this by telling investors the company would “simply pivot” if its AGI spending strategy proves incorrect.
The mad scramble by mega tech companies and Open AI to build AI data centers is largely relying on debt markets, with a slew of public and private mega deals since September. Hyperscalers would have to spend 94% of operating cash flow to pay for their AI buildouts so are turning to debt financing to help defray some of that cost, according to Bank of America. Unlike earnings per share, cash flow can’t be manipulated by companies. If they spend more on AI than they generate internally, they have to finance the difference.
Hyperscaler debt taken on so far this year have raised almost as much money as all debt financings done between 2020 and 2024, the BofA research said. BofA calculates $75 billion of AI-related public debt offerings just in the past two months!
In bubbles, everyone gets caught up in the idea that spending on the hot theme will deliver vast profits — eventually. When the bubble is big enough, it shifts the focus of the market as a whole from disliking capital expenditure, and hating speculative capital spending in particular, to loving it. That certainly seems the case today with surging AI spending. For much more, please check-out the References below.
Postscript: November 23, 2025:
In this new AI era, consumers and workers are not what drives the economy anymore. Instead, it’s spending on all things AI, mostly with borrowed money or circular financing deals.
BofA Research noted that Meta and Oracle issued $75 billion in bonds and loans in September and October 2025 alone to fund AI data center build outs, an amount more than double the annual average over the past decade. They warned that “The AI boom is hitting a money wall” as capital expenditures consume a large portion of free cash flow. Separately, a recent Bank of America Global Fund Manager Survey found that 53% of participating fund managers felt that AI stocks had reached bubble proportions. This marked a slight decrease from a record 54% in the prior month’s survey, but the concern has grown over time, with the “AI bubble” cited as the top “tail risk” by 45% of respondents in the November 2025 poll.
JP Morgan Chase estimates up to $7 trillion of AI spending will be with borrowed money. “The question is not ‘which market will finance the AI-boom?’ Rather, the question is ‘how will financings be structured to access every capital market?’ according to strategists at the bank led by Tarek Hamid.
As an example of AI debt financing, Meta did a $27 billion bond offering. It wasn’t on their balance sheet. They paid 100 basis points over what it would cost to put it on their balance sheet. Special purpose vehicles happen at the tail end of the cycle, not the early part of the cycle, notes Rajiv Jain of GQG Partners.
References:
Market research firms Omdia and Dell’Oro: impact of 6G and AI investments on telcos
Market research firm Omdia (owned by Informa) this week forecast that 6G and AI investments are set to drive industry growth in the global communications market. As a result, global communications providers’ revenue is expected to reach $5.6 trillion by 2030, growing at a 6.2% CAGR from 2025. Investment momentum is also expected to shift toward mobile networks from 2028 onward, as tier 1 markets prepare for 6G deployments. Telecoms capex is forecast to reach $395 billion by 2030, with a 3.6% CAGR, while technology capex will surge to $545 billion, reflecting a 9.3% CAGR.
Fixed telecom capex will gradually decline due to market saturation. Meanwhile, AI infrastructure, cloud services, and digital sovereignty policies are driving telecom operators to expand data centers and invest in specialized hardware.
Key market trends:
- CP capex per person will increase from $74 in 2024 to $116 in 2030, with CP capex reaching 2.5% of global GDP investment.
-
Capital intensity in telecom will decline until 2027, then rise due to mobile network upgrades.
-
Regional leaders in revenue and capex include North America, Oceania & Eastern Asia, and Western Europe, with Central & Southern Asia showing the highest growth potential.
Dario Talmesio, research director at Omdia said, “telecom operators are entering a new phase of strategic investment. With 6G on the horizon and AI infrastructure demands accelerating, the connectivity business is shifting from volume-based pricing to value-driven connectivity.”
Omdia’s forecast is based on a comprehensive model incorporating historical data from 67 countries, local market dynamics, regulatory trends, and technology migration patterns.
…………………………………………………………………………………………………………………………………………………
Separately, Dell’Oro Group sees 6G capex ramping around 2030, although it warns that the RAN market remains flat, “raising key questions for the industry’s future.” Cumulative 6G RAN investments over the 2029-2034 period are projected to account for 55% to 60% of the total RAN capex over the same forecast period.
“Our long-term position and characterization of this market have not changed,” said Stefan Pongratz, Vice President of RAN and Telecom Capex research at Dell’Oro Group. “The RAN network plays a pivotal role in the broader telecom market. There are opportunities to expand the RAN beyond the traditional MBB (mobile broadband) use cases. At the same time, there are serious near-term risks tilted to the downside, particularly when considering the slowdown in data traffic,” continued Pongratz.
Additional highlights from Dell’Oro’s October 2025 6G Advanced Research Report:
- The baseline scenario is for the broader RAN market to stay flat over the next 10 years. This is built on the assumption that the mobile network will run into utilization challenges by the end of the decade, spurring a 6G capex ramp dominated by Massive MIMO systems in the Sub-7GHz/cm Wave spectrum, utilizing the existing macro grid as much as possible.
- The report also outlines more optimistic and pessimistic growth scenarios, depending largely on the mobile data traffic growth trajectory and the impact beyond MBB, including private wireless and FWA (fixed wireless access).
- Cumulative 6G RAN investments over the 2029-2034 period are projected to account for 55 to 60 percent of the total RAN capex over the same forecast period.
Dell’Oro Group’s 6G Advanced Research Report offers an overview of the RAN market by technology, with tables covering manufacturers’ revenue for total RAN over the next 10 years. 6G RAN is analyzed by spectrum (Sub-7 GHz, cmWave, mmWave), by Massive MIMO, and by region (North America, Europe, Middle East and Africa, China, Asia Pacific Excl. China, and CALA). To purchase this report, please contact by email at [email protected].
References:
https://www.lightreading.com/6g/6g-momentum-is-building
6G Capex Ramp to Start Around 2030, According to Dell’Oro Group
https://www.lightreading.com/6g/6g-course-correction-vendors-hear-mno-pleas
https://www.lightreading.com/6g/what-at-t-really-wants-from-6g
Nvidia pays $1 billion for a stake in Nokia to collaborate on AI networking solutions
This is not only astonishing but unheard of: the world’s largest and most popular fabless semiconductor company –Nvidia– taking a $1 billion stake in a telco/reinvented data center connectivity company-Nokia.
Indeed, GPU king Nvidia will pay $1 billion for a stake of 2.9% in Nokia as part of a deal focused on AI and data centers, the Finnish telecom equipment maker said on Tuesday as its shares hit their highest level in nearly a decade on hope for AI to lift their business revenue and profits. The nonexclusive partnership and the investment will make Nvidia the second-largest shareholder in Nokia. Nokia said it will issue 166,389,351 new shares for Nvidia, which the U.S. company will subscribe to at $6.01 per share.
Nokia said the companies will collaborate on artificial intelligence networking solutions and explore opportunities to include its data center communications products in Nvidia’s future AI infrastructure plans. Nokia and its Swedish rival Ericsson both make networking equipment for connectivity inside (intra-) data centers and between (inter-) data centers and have been benefiting from increased AI use.
Summary:
- NVIDIA and Nokia to establish a strategic partnership to enable accelerated development and deployment of next generation AI native mobile networks and AI networking infrastructure.
- NVIDIA introduces NVIDIA Arc Aerial RAN Computer, a 6G-ready telecommunications computing platform.
- Nokia to expand its global access portfolio with new AI-RAN product based on NVIDIA platform.
- T-Mobile U.S. is working with Nokia and NVIDIA to integrate AI-RAN technologies into its 6G development process.
- Collaboration enables new AI services and improved consumer experiences to support explosive growth in mobile AI traffic.
- Dell Technologies provides PowerEdge servers to power new AI-RAN solution.
- Partnership marks turning point for the industry, paving the way to AI-native 6G by taking AI-RAN to innovation and commercialization at a global scale.
In some respects, this new partnership competes with Nvidia’s own data center connectivity solutions from its Mellanox Technologies division, which it acquired for $6.9 billion in 2019. Meanwhile, Nokia now claims to have worked on a redesign to ensure its RAN software is compatible with Nvidia’s compute unified device architecture (CUDA) platform, meaning it can run on Nvidia’s GPUs. Nvidia has also modified its hardware offer, creating capacity cards that will slot directly into Nokia’s existing AirScale baseband units at mobile sites.
Having dethroned Intel several years ago, Nvidia now has a near-monopoly in supplying GPU chips for data centers and has partnered with companies ranging from OpenAI to Microsoft. AMD is a distant second but is gaining ground in the data center GPU space as is ARM Ltd with its RISC CPU cores. Capital expenditure on data center infrastructure is expected to exceed $1.7 trillion by 2030, consulting firm McKinsey, largely because of the expansion of AI.
Nvidia CEO Jensen Huang said the deal would help make the U.S. the center of the next revolution in 6G. “Thank you for helping the United States bring telecommunication technology back to America,” Huang said in a speech in Washington, addressing Nokia CEO Justin Hotard (x-Intel). “The key thing here is it’s American technology delivering the base capability, which is the accelerated computing stack from Nvidia, now purpose-built for mobile,” Hotard told Reuters in an interview. “Jensen and I have been talking for a little bit and I love the pace at which Nvidia moves,” Hotard said. “It’s a pace that I aspire for us to move at Nokia.” He expects the new equipment to start contributing to revenue from 2027 as it goes into commercial deployment, first with 5G, followed by 6G after 2030.
Nvidia has been on a spending spree in recent weeks. The company in September pledged to invest $5 billion in beleaguered chip maker Intel. The investment pairs the world’s most valuable company, which has been a darling of the AI boom, with a chip maker that has almost completely fallen out of the AI conversation.
Later that month, Nvidia said it planned to invest up to $100 billion in OpenAI over an unspecified period that will likely span at least a few years. The partnership includes plans for an enormous data-center build-out and will allow OpenAI to build and deploy at least 10 gigawatts of Nvidia systems.
…………………………………………………………………………………………………………………………………………………………………………………………………………………………………
Tech Details:
Nokia uses Marvell Physical Layer (1) baseband chips for many of its products. Among other things, this ensured Nokia had a single software stack for all its virtual and purpose-built RAN products. Pallavi Mahajan, Nokia’s recently joined chief technology and AI officer recently told Light Reading that their software could easily adapt to run on Nvidia’s GPUs: “We built a hardware abstraction layer so that whether you are on Marvell, whether you are on any of the x86 servers or whether you are on GPUs, the abstraction takes away from that complexity, and the software is still the same.”
Fully independent software has been something of a Holy Grail for the entire industry. It would have ramifications for the whole market and its economics. Yet Nokia has conceivably been able to minimize the effort required to put its Layer 1 and specific higher-layer functions on a GPU. “There are going to be pieces of the software that are going to leverage on the accelerated compute,” said Mahajan. “That’s where we will bring in the CUDA integration pieces. But it’s not the entire software,” she added. The appeal of Nvidia as an alternative was largely to be found in “the programmability pieces that you don’t have on any other general merchant silicon,” said Mahajan. “There’s also this entire ecosystem, the CUDA ecosystem, that comes in.” One of Nvidia’s most eye-catching recent moves is the decision to “open source” Aerial, its own CUDA-based RAN software framework, so that other developers can tinker, she says. “What it now enables is the entire ecosystem to go out and contribute.”
…………………………………………………………………………………………………………………………………………………………………………………………………………………………………
Quotes:
“Telecommunications is a critical national infrastructure — the digital nervous system of our economy and security,” said Jensen Huang, founder and CEO of NVIDIA. “Built on NVIDIA CUDA and AI, AI-RAN will revolutionize telecommunications — a generational platform shift that empowers the United States to regain global leadership in this vital infrastructure technology. Together with Nokia and America’s telecom ecosystem, we’re igniting this revolution, equipping operators to build intelligent, adaptive networks that will define the next generation of global connectivity.”
“The next leap in telecom isn’t just from 5G to 6G — it’s a fundamental redesign of the network to deliver AI-powered connectivity, capable of processing intelligence from the data center all the way to the edge. Our partnership with NVIDIA, and their investment in Nokia, will accelerate AI-RAN innovation to put an AI data center into everyone’s pocket,” said Justin Hotard, President and CEO of Nokia. “We’re proud to drive this industry transformation with NVIDIA, Dell Technologies, and T-Mobile U.S., our first AI-RAN deployments in T-Mobile’s network will ensure America leads in the advanced connectivity that AI needs.”
……………………………………………………………………………………………………………………………………………………………………………………
Editor’s Notes:
1. In more advanced 5G networks, Physical Layer functions have demanded the support of custom silicon, or “accelerators.” A technique known as “lookaside,” favored by Ericsson and Samsung, uses an accelerator for only a single problematic Layer 1 task – forward error correction – and keeps everything else on the CPU. Nokia prefers the “inline” approach, which puts the whole of Layer 1 on the accelerator.
2. The huge AI-RAN push that Nvidia started with the formation of the AI-RAN Alliance in early 2024 has not met with an enthusiastic telco response so far. Results from Nokia as well as Ericsson show wireless network operators are spending less on 5G rollouts than they were in the early 2020s. Telco numbers indicate the appetite for smartphone and other mobile data services has not produced any sales growth. As companies prioritize efficiency above all else, baseband units that include Marvell and Nvidia cards may seem too expensive.
……………………………………………………………………………………………………………………………………………………………………………………….
Other Opinions and Quotes:
Nvidia chips are likely to be more expensive, said Mads Rosendal, analyst at Danske Bank Credit Research, but the proposed partnership would be mutually beneficial, given Nvidia’s large share in the U.S. data center market.
“This is a strong endorsement of Nokia’s capabilities,” said PP Foresight analyst Paolo Pescatore. “Next-generation networks, such as 6G, will play a significant role in enabling new AI-powered experiences,” he added.
Iain Morris, International Editor, Light Reading: “Layer 1 control software runs on ARM RISC CPU cores in both Marvell and Nvidia technologies. The bigger differences tend to be in the hardware acceleration “kernels,” or central components, which have some unique demands. Yet Nokia has been working to put as much as it possibly can into a bucket of common software. Regardless, if Nvidia is effectively paying for all this with its $1 billion investment, the risks for Nokia may be small………….Nokia’s customers will in future have an AI-RAN choice that limits or even shrinks the floorspace for Marvell. The development also points to more subtle changes in Nokia’s thinking. The message earlier this year was that Nokia did not require a GPU to implement AI for RAN, whereby machine-generated algorithms help to improve network performance and efficiency. Marvell had that covered because it had incorporated AI and machine-learning technologies into the baseband chips used by Nokia.”
“If you start doing inline, you typically get much more locked into the hardware,” said Per Narvinger, the president of Ericsson’s mobile networks business group, on a recent analyst call. During its own trials with Nvidia, Ericsson said it was effectively able to redeploy virtual RAN software written for Intel’s x86 CPUs on the Grace CPU with minimal changes, leaving the GPU only as a possible option for the FEC accelerator. Putting the entire Layer 1 on a GPU would mean “you probably also get more tightly into that specific implementation,” said Narvinger. “Where does it really benefit from having that kind of parallel compute system?”
………………………………………………………………………………………………………………………………………………….
Separately, Nokia and Nvidia will partner with T-Mobile U.S. to develop and test AI RAN technologies for developing 6G, Nokia said in its press release. Trials are expected to begin in 2026, focused on field validation of performance and efficiency gains for customers.
References:
https://nvidianews.nvidia.com/news/nvidia-nokia-ai-telecommunications
https://www.reuters.com/world/europe/nvidia-make-1-billion-investment-finlands-nokia-2025-10-28/
https://www.lightreading.com/5g/nvidia-takes-1b-stake-in-nokia-which-promises-5g-and-6g-overhaul
https://www.wsj.com/business/telecom/nvidia-takes-1-billion-stake-in-nokia-69f75bb6
Analysis: Cisco, HPE/Juniper, and Nvidia network equipment for AI data centers
Nvidia’s networking solutions give it an edge over competitive AI chip makers
Telecom sessions at Nvidia’s 2025 AI developers GTC: March 17–21 in San Jose, CA
Nvidia AI-RAN survey results; AI inferencing as a reinvention of edge computing?
FT: Nvidia invested $1bn in AI start-ups in 2024
The case for and against AI-RAN technology using Nvidia or AMD GPUs
Highlights of Nokia’s Smart Factory in Oulu, Finland for 5G and 6G innovation
Nokia & Deutsche Bahn deploy world’s first 1900 MHz 5G radio network meeting FRMCS requirements
Will the wave of AI generated user-to/from-network traffic increase spectacularly as Cisco and Nokia predict?
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
