TL;DR
A Thorsten Meyer AI report says frontier AI companies are increasingly renting GPU capacity from neoclouds and, in some cases, from competitors. The report argues that supplier financing, equity stakes and long-term compute commitments have created a circular market centered on Nvidia hardware, though many figures remain reported commitments rather than cash already spent.
A new Thorsten Meyer AI report says the race to build frontier AI systems is increasingly being run on rented infrastructure, with major labs buying access to GPU clusters from neocloud providers, established cloud firms and, according to the report, even direct competitors.
The report, titled The Neocloud Cartel and framed as Part 2 of the Control Series on AI power chokepoints, focuses on compute: the large-scale GPU infrastructure used to train and serve advanced AI models. Its central finding is that many AI companies do not own much of the hardware they rely on, instead signing large rental, cloud and capacity deals with a small group of suppliers.
Thorsten Meyer AI describes neoclouds as AI-focused GPU rental companies that grew during the 2024 and 2025 GPU shortage, when labs faced long waits for high-end chips. The report identifies CoreWeave as the largest player in the category, citing a contracted backlog above $55 billion and reported commitments from Meta and OpenAI. It also names Nebius, Crusoe, Lambda, Together, Fireworks, Nscale and IREN as part of the broader market.
The most striking claim in the source material concerns xAI. The report says xAI leased capacity from its Colossus 1 supercomputer to Anthropic for about $1.25 billion a month and to Google for about $920 million a month after Grok training moved elsewhere and the cluster was underused. Those figures are attributed to the report and remain described as reported lease terms, not independently confirmed in the source material provided.
The Neocloud Cartel
Almost no one racing to build AI owns the machine it runs on. They rent — increasingly from each other — and the money loops back to one chip maker that’s also an investor in nearly everyone at the table.
The cartel isn’t a conspiracy — it’s the endpoint of extreme capital intensity, real scarcity, and one dominant supplier. But the same circularity that makes it powerful makes it a fuse: each cancelled order is someone else’s missing revenue. Don’t be a price-taker at the bottom of a loop you don’t control — own your inference, keep an open-weight fallback, diversify silicon.
Compute Dependence Shapes AI Power
The development matters because compute access now affects which companies can train the largest models, how fast they can ship products and how much leverage suppliers hold over AI labs. If the report’s account is correct, the AI industry’s infrastructure layer is becoming more concentrated even as the number of customer-facing AI products grows.
The report argues that Nvidia sits at the center of this market because its GPUs remain the main hardware used in high-end AI buildouts. It says Nvidia captures a large share of every data center buildout dollar, holds equity in several buyers and can influence allocation during supply shortages. That combination, the report argues, gives the chipmaker unusual power across both the supply chain and the financing chain.
For readers, the practical concern is not only corporate concentration. A circular funding structure can make the market look stronger than it is if future revenue depends on customers whose spending is partly enabled by suppliers. The report warns that canceled orders or lower-than-expected AI demand could move through the same loop quickly, turning one company’s delayed buildout into another company’s missing revenue.
As an affiliate, we earn on qualifying purchases.
Shortages Built A Rental Market
The neocloud category expanded after demand for Nvidia GPUs outpaced supply in 2024 and 2025. Rather than wait years to build their own data centers, AI labs and enterprises could rent capacity from companies that specialized in buying, financing and operating large GPU clusters.
The report says that model has since evolved. OpenAI is described as having made roughly $1.15 trillion in long-term compute and hardware commitments across Broadcom, Oracle, Microsoft, Nvidia, AMD, AWS and CoreWeave. The report stresses that these are reported multi-year commitments, not cash sitting on balance sheets.
Supplier financing is a major part of the story. According to the source material, Nvidia agreed in 2025 to invest up to $100 billion in OpenAI, took a $5 billion stake in Intel, held equity in CoreWeave, Nebius and Applied Digital, and pre-purchased $6.3 billion of CoreWeave capacity as a backstop. The report also says OpenAI’s AMD agreement includes warrants for up to 160 million AMD shares, while Nvidia and Microsoft committed up to $15 billion to Anthropic.
“Almost no one racing to build AI owns the machine it runs on.”
— Thorsten Meyer AI
SLURM FOR AI AND DEEP LEARNING: GPU CLUSTER MANAGEMENT AND DISTRIBUTED TRAINING: SCHEDULE PYTORCH, TENSORFLOW, AND MULTI-NODE LLM WORKLOADS WITH JOB QUEUING AND RESOURCE OPTIMIZATION
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Reported Deals Need Verification
Several key numbers in the source material are presented as reported commitments or lease terms, and the provided material does not include underlying contracts. It is not yet clear how much of the cited spending is binding, how much can be delayed or canceled, and how much will convert into actual revenue for suppliers.
The xAI lease claims are especially material because they would show a frontier AI lab renting major capacity to direct rivals. The source material says Anthropic and Google used xAI capacity, but the details, duration and operational limits of those arrangements are not fully established in the provided material.
It is also unclear how durable the neocloud business model will be if GPU rental prices continue falling. The report cites H100 rental rates down 60% to 75% from their peak, which could weaken margins for companies that financed clusters at high prices.
upHere GPU Support Bracket,Graphics Card GPU Support, Video Card Sag Holder Bracket, GPU Stand, M( 49-80mm / 1.93-3.15in ),GB49K
- Material: All-aluminum construction for durability
- Adjustment: Tool-free telescopic height adjustment
- Compatibility: Fits GTX, RTX, Radeon cards
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Orders, Prices And Utilization
The next test is whether AI demand grows fast enough to support the scale of contracted compute. Investors, customers and regulators are likely to watch utilization rates, rental prices, debt financing and whether the largest labs continue signing supplier-backed infrastructure deals.
The report’s practical recommendation is that companies buying AI services should avoid becoming dependent on a single compute loop. It advises users to own or control inference where possible, keep open-weight model options available and diversify across silicon providers when the workload allows.
THE CLOUD REDOUBT : Sovereign Data Centers, The Semiconductor Sandbox, and the Geopolitics of AI Compute Infrastructure
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is a neocloud?
A neocloud is an AI-focused cloud provider that rents access to GPU clusters, often without offering the wider range of services found in traditional cloud platforms.
Why does the report call the market circular?
The report says suppliers invest in AI labs and infrastructure companies, those companies commit to buying compute or chips, and the resulting deals raise the value of the same firms inside the loop.
Is the report alleging illegal collusion?
No. The source material says the structure is not a conspiracy. It frames the concentration as the result of scarce GPUs, high capital costs and Nvidia’s central role in AI hardware.
What remains unconfirmed?
The exact terms of some reported leases and commitments remain unclear from the provided material, including how binding the deals are and how much money has already changed hands.
Why should AI customers care?
Customers could face higher dependency risk if their AI products rely on a narrow group of compute suppliers. Falling rental prices or canceled orders could also affect service availability, pricing and vendor stability.
Source: Thorsten Meyer AI
