TL;DR
Thorsten Meyer AI’s latest Control Series installment reports that leading AI labs are relying on rented GPU capacity, including from rivals, while chipmakers and cloud suppliers finance some of the same customers buying their hardware. The report frames the pattern as a compute bottleneck whose strength depends on continued demand, financing and chip access.
Thorsten Meyer AI has published a new installment of its Control Series arguing that the AI industry’s compute market is being reshaped by circular rental and financing deals, with frontier labs renting GPU capacity from specialized cloud firms, suppliers and even rivals while much of the spending flows back to Nvidia.
The report says many leading AI companies do not own much of the compute they use for training and inference. Instead, they rent from so-called neocloud providers: GPU-focused cloud firms built around AI workloads. CoreWeave is identified as the largest example, with the report citing a contracted backlog above $55 billion and major commitments from Meta and OpenAI.
The most striking development cited in the report is xAI’s reported lease of its Colossus 1 supercomputer to Anthropic for about $1.25 billion a month and to Google for about $920 million a month. Thorsten Meyer AI says the cluster had fallen to roughly 11% utilization after Grok training shifted elsewhere. Those claims are attributed to the report’s source base, which it lists as including SpaceX filings, TechCrunch, The Register, Bloomberg, CNBC, Reuters, SemiAnalysis, McKinsey, Morgan Stanley and the Financial Times.
The report also describes a wider financing loop around OpenAI, Nvidia, AMD, Microsoft, Anthropic, Oracle, AWS, CoreWeave and other firms. It says OpenAI has made reported multi-year compute and hardware commitments totaling about $1.15 trillion, while Nvidia has agreed to invest up to $100 billion in OpenAI. The report frames those arrangements as reported commitments rather than cash currently on hand.
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 Power Is Concentrating
The report matters because access to advanced chips has become one of the main constraints on AI development. If a small group of chip suppliers, cloud landlords and frontier labs controls most high-end compute capacity, smaller companies may face higher costs, weaker bargaining power and less reliable access.
Thorsten Meyer AI’s central interpretation is that the market now behaves less like a broad commodity market and more like a concentrated capital loop. That is a claim about market structure, not a legal finding. The report does not present evidence of illegal coordination; it argues that scarcity, cost and supplier financing have pushed the industry into a small circle of interdependent deals.
For readers outside the AI sector, the stakes are practical. Compute costs shape which AI tools get built, who can compete, how fast model capability grows and whether current investment levels are financially durable. If rental prices fall or demand slows, one company’s canceled order could become another company’s missing revenue.
GPU cloud computing services
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Neoclouds Filled The GPU Gap
The report defines neoclouds as AI-focused cloud providers that rent GPU capacity without the broader business lines of legacy cloud platforms. It says the category grew quickly during the 2024 and 2025 GPU shortage, when even well-funded labs faced long waits for hardware and used rental deals to reach scale faster.
CoreWeave, Nebius, Crusoe, Lambda, Together, Fireworks, Nscale and IREN are among the companies named in the report as part of this market. They are described as backed by venture capital, private equity, sovereign money or large strategic buyers, while mostly renting out similar Nvidia-based infrastructure.
The report also places Nvidia at the center of the system. It says Nvidia captures a large share of each data center buildout through GPU sales, holds equity in several buyers and has influence over chip allocation during shortages. It also cites Nvidia’s investments in OpenAI, Intel, CoreWeave, Nebius and Applied Digital, along with a $6.3 billion CoreWeave capacity backstop.
“Almost no one racing to build AI owns the machine it runs on.”
— Thorsten Meyer AI report
AI training GPU rental
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Financing Risks Remain Unclear
Several key details remain unconfirmed from the provided material. It is not clear how much of the reported $1.15 trillion in OpenAI-related commitments is binding, how much is conditional, and how much depends on future revenue growth, financing or chip availability.
The durability of the xAI, Anthropic and Google arrangements also remains uncertain from the report alone. The source material states monthly lease figures and utilization data, but it does not provide contract terms, duration, termination rights or confirmation from all named companies.
It is also unclear whether lower H100 rental prices, cited in the report as down 60% to 75% from their peak, signal healthy supply growth, weaker demand, or both. That distinction matters for investors, cloud providers and AI companies planning multi-year spending.
supercomputer leasing for AI
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Orders Will Test The Loop
The next test is whether AI revenue, cloud demand and financing can keep pace with the scale of reported compute commitments. The report points to projected data center spending of about $3 trillion from 2025 to 2028, with private credit expected to fund a large share.
Readers should watch for new disclosures from public companies such as CoreWeave, Nvidia, Microsoft, Oracle and AMD, along with any confirmed contract details from OpenAI, Anthropic, xAI and Google. The most important signals will be utilization, rental pricing, canceled or delayed orders, and whether suppliers keep financing their own customers.
cloud GPU providers for AI
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is the actual news in this report?
The development is Thorsten Meyer AI’s publication of a 2026 analysis arguing that AI compute has become a concentrated rental and financing system linking labs, neoclouds, chipmakers and cloud providers.
Does the report prove an illegal cartel?
No. The report uses “cartel” as a description of concentrated market dynamics. It does not present a legal finding or evidence of unlawful collusion.
What is a neocloud?
A neocloud is an AI-focused cloud provider that rents GPU capacity, usually for training or running AI models, rather than offering a broad general-purpose cloud business.
Why is Nvidia central to the story?
The report says Nvidia supplies much of the hardware behind these buildouts, invests in several companies buying or renting GPU capacity, and benefits when new compute commitments translate into chip demand.
What should readers watch next?
Watch whether reported multi-year commitments become actual spending, whether rental prices keep falling, and whether large AI labs can generate enough revenue to support their compute plans.
Source: Thorsten Meyer AI
