TL;DR
Thorsten Meyer AI’s latest Control Series installment argues that frontier AI companies increasingly rent GPU capacity from neoclouds, chip suppliers and even rival labs. The report frames recent deals involving xAI, Anthropic, Google, OpenAI, CoreWeave, Nvidia and AMD as a circular financing system, while stressing that many figures are multi-year commitments rather than cash already spent.
Thorsten Meyer AI’s new Control Series report says the AI industry’s race for compute is increasingly built on rented GPU capacity, supplier-financed purchases and lease deals between direct competitors, a structure the report calls a “neocloud cartel” rather than a normal open market.
The report’s central claim is that many frontier AI companies do not own the machines they use for training and inference at scale. Instead, they rent capacity from AI-focused cloud providers, chip-linked partners and, in some cases, other AI labs.
The most striking reported example is xAI. The report says xAI leased 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’s utilization fell to about 11%. Those lease terms have not been independently detailed in the material provided, so they should be treated as reported figures rather than company-confirmed terms.
The report also says OpenAI has made roughly $1.15 trillion in long-term compute and hardware commitments across Broadcom, Oracle, Microsoft, Nvidia, AMD, AWS and CoreWeave. It adds a key caveat: these are reported commitments, often spread over years, not 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 Rents Concentrate AI Power
The report matters because compute access is now one of the main limits on which companies can build frontier AI systems. If labs depend on landlords, suppliers and rival-controlled clusters, pricing, allocation and contract terms can shape who can train large models and who gets pushed to smaller deployments.
The financing structure also creates risk. If one company’s compute contract is another company’s expected revenue, a cancelled order or lower utilization rate can move through the same circle quickly. The report cites falling H100 rental rates and low consumer payment rates for AI as pressure points that could weaken the assumptions behind large buildouts.
Local AI on Linux in Practice: Build Private LLM Servers, GPU Workstations, Ollama Apps, Dockerized AI Services, and Self-Hosted AI Infrastructure with CUDA, ROCm, vLLM, and Open WebUI
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Shortage Created GPU Landlords
The neocloud category grew out of the 2024 and 2025 GPU shortage, when even well-funded labs faced long waits for Nvidia hardware. Companies such as CoreWeave, Nebius, Crusoe, Lambda and others built businesses around renting Nvidia GPU capacity to AI customers that needed scale faster than they could build it themselves.
CoreWeave is described in the report as the largest player in the category, with contracted backlog above $55 billion and major commitments from Meta and OpenAI. The report says Nvidia sits upstream of much of this activity through chip supply, equity stakes and financing arrangements with companies that then buy or rent Nvidia-based capacity.
“almost no one racing to build AI owns the machine it runs on”
— Thorsten Meyer AI
AI training GPU rentals
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Private Terms Limit Certainty
Several key details remain unclear. The report does not establish that every contract size, cancellation clause or utilization figure has been confirmed by the companies involved. It also does not show unlawful coordination; its use of “cartel” describes concentration, circular financing and supplier power.
It is also unclear how much of the reported multi-year spending will become actual cash outflow, how much can be renegotiated, and whether falling rental prices will reduce revenue expected by neocloud providers.
Practical RHEL AI: Designing, Deploying and Scaling AI Solutions with Red Hat Enterprise Linux
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Watch Cancellations And Utilization
The next signals will come from filings, earnings reports, credit markets and new capacity announcements. Investors and customers will be watching whether AI labs keep signing large compute commitments, whether neocloud utilization stays high, and whether Nvidia’s allocation power weakens as more supply comes online.
The central question is whether circular AI compute financing can support real demand, or whether slower model revenue and falling GPU rental prices force labs and landlords to revise their plans.
High-Performance GPU Computing with Python: Unlock Massive Speedups in Data Science and ML Using CUDA and Numba
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 large AI models, without offering the full range of general cloud services.
Did the report prove an illegal cartel?
No. The report uses “cartel” to describe concentrated control, circular financing and scarce compute access. It does not present a legal finding of unlawful collusion.
Why is Nvidia central to the report?
The report says Nvidia captures a large share of AI data center spending through GPU sales while also holding stakes or financing ties with companies buying or renting its hardware.
What remains unconfirmed?
Exact lease terms, cancellation rights, utilization levels and the timing of many reported commitments remain partly private or based on cited reporting rather than full company disclosure.
Source: Thorsten Meyer AI
