TL;DR
A Thorsten Meyer AI report says frontier AI companies increasingly rent GPU capacity from specialized cloud providers and, in some cases, from one another. The report argues that supplier financing, equity stakes and long-term compute commitments have created a circular market centered on Nvidia hardware, though several figures remain based on reported multi-year commitments rather than cash already spent.
A new Thorsten Meyer AI report says the AI industry’s race for computing power is now being shaped by a small set of GPU suppliers, neocloud operators and frontier labs that increasingly rent capacity from one another, a pattern the report argues has turned compute into the main chokepoint in the sector.
The report, published as Part 2 of Thorsten Meyer AI’s Control Series, says many leading AI companies do not own most of the machines they use for training and inference. Instead, they rent GPU capacity from specialized providers such as CoreWeave and other so-called neoclouds, and in some cases from direct competitors.
According to the source material, 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 its own Grok training work moved elsewhere. The report says the cluster was running at about 11% utilization, making rental revenue more attractive than leaving capacity idle. Those terms have not been independently detailed in the supplied material beyond the cited reporting.
The report also points to large reported compute and hardware commitments by OpenAI, including deals tied to Broadcom, Oracle, Microsoft, Nvidia, AMD, AWS and CoreWeave. It says those commitments total roughly $1.15 trillion over the next decade, while warning that the figures reflect reported multi-year obligations and supplier arrangements, not cash already available or spent.
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.
The Risk in Circular Compute
The report matters because compute capacity has become one of the main constraints on advanced AI development. If a handful of suppliers, lenders and landlords control access to high-end GPUs, they can shape which labs train larger models, how quickly rivals can scale, and what customers pay for AI services.
The central concern is not a proven conspiracy. The report describes the pattern as the result of extreme capital needs, chip scarcity and one dominant supplier. Still, it argues that circular financing can make the market more fragile: a canceled compute order can become lost revenue for a neocloud, weaker demand for a chipmaker and pressure on investors that funded the buildout.
For enterprise AI buyers, the practical risk is dependence. Companies that build products on rented compute may face price changes, capacity limits or contract terms set by vendors whose interests do not fully match their own. The report recommends owning more inference capacity where possible, keeping open-weight model fallbacks and diversifying silicon exposure.
Nvidia GPU cloud computing services
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How Neoclouds Became Central
The neocloud category grew out of the 2024 and 2025 GPU shortage, when even well-funded AI labs faced long waits for high-end Nvidia chips. These companies offer AI-focused GPU rental services without the broader business lines of general-purpose cloud providers.
CoreWeave is described in the report as the largest example, with a contracted backlog above $55 billion. The source material says Meta has committed about $35 billion across two deals and OpenAI about $22 billion. It also names Nebius, Crusoe, Lambda, Together, Fireworks, Nscale and IREN as part of the wider field.
The report’s broader claim is that ownership and use have separated. A company may finance or own a cluster, rent it to a rival, buy chips from a supplier that also owns equity, and depend on future customer growth to support the entire chain.
“Almost no one racing to build AI owns the machine it runs on.”
— Thorsten Meyer AI report
AI training GPU rental services
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Figures Still Need Verification
Several figures in the source material are described as reported commitments, multi-year arrangements or projected spending rather than cash already paid. It is not yet clear how much of the reported compute spending will be completed, renegotiated or canceled if AI revenue growth falls short of expectations.
The report cites outside reporting from outlets including Bloomberg, CNBC, Reuters, TechCrunch, The Register, the Financial Times, SemiAnalysis, McKinsey and Morgan Stanley, but the supplied material does not include the underlying documents or full contract terms. Details such as utilization rates, lease rights, chip allocation terms and financing conditions may change as new filings or company statements emerge.
high performance GPU cloud providers
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Watch the Next Contracts
The next test is whether AI companies can turn rented compute into revenue fast enough to support the scale of reported commitments. Investors and customers will be watching new lease announcements, supplier financing deals, cloud revenue concentration and any signs that rental prices for high-end GPUs keep falling from earlier peaks.
More public filings from neocloud providers and AI infrastructure suppliers could show whether the market is becoming more diversified or more dependent on the same small group of chipmakers, landlords and model developers.
AI compute rental platforms
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 GPU capacity, usually for training or running AI models, without operating like a broad general-purpose cloud platform.
Why does the report focus on Nvidia?
The report says Nvidia sits upstream of much of the AI buildout because its GPUs remain central to large-scale training and inference, and because it has also taken investment or financing roles in several companies that buy or rent its hardware.
Is the report alleging an illegal cartel?
No. The report uses the word as a market critique. It says the pattern comes from capital intensity, chip scarcity and supplier dominance, not from a confirmed illegal agreement among companies.
Why would AI rivals rent compute from each other?
If one company has unused GPU capacity and another needs immediate access, a rental deal can generate revenue for the owner and speed up training for the customer. The report says this has happened even among firms that compete in AI models.
What is still unknown?
The full contract terms, cancellation rights, actual cash flows and long-term economics remain unclear from the supplied material. The scale of future demand for paid AI products is also still developing.
Source: Thorsten Meyer AI
