TL;DR
A Thorsten Meyer AI report says frontier AI companies are relying on rented GPU capacity, sometimes from rivals, while chip suppliers and cloud firms help finance the buildout. The strongest confirmed pieces are reported SpaceX/xAI leases with Anthropic and Google, CoreWeave’s customer concentration and Nvidia’s announced investment links. What remains unclear is how much cash has changed hands and whether the largest commitments can be funded.
Thorsten Meyer AI has published a June 2026 Control Series report arguing that frontier AI compute has become a rental market in which labs, chip suppliers and cloud firms increasingly finance one another’s capacity. The report points to reported SpaceX/xAI leases for Anthropic and Google, CoreWeave’s large contracted backlog and Nvidia-linked financing as evidence that control over GPUs now sits with a small group of suppliers and landlords rather than with most model developers.
The report’s clearest new example is the Colossus leasing arrangement. MarketWatch, citing a SpaceX filing, reported that Anthropic agreed to pay about $1.25 billion per month for capacity at SpaceX/xAI’s Colossus data centers. Business Insider separately reported that Google agreed to pay SpaceX about $920 million per month for AI compute beginning in October 2026. Elon Musk said the Anthropic deal was short-term, with a 180-day term and 90-day mutual cancellation notice, so the headline value should not be read as cash already received.
Thorsten Meyer AI frames those deals as part of a wider loop. It says OpenAI has reported compute and hardware commitments of about $1.15 trillion across Broadcom, Oracle, Microsoft, Nvidia, AMD, AWS and CoreWeave. The report stresses that these figures are multi-year commitments, not current cash balances.
Nvidia is described as the common supplier at the center of the system. The Guardian reported in September 2025 that Nvidia agreed to invest up to $100 billion in OpenAI as part of an AI infrastructure partnership. The Thorsten Meyer AI report also cites Nvidia stakes or backstops involving CoreWeave, Nebius, Applied Digital and xAI, while OpenAI’s AMD agreement gives OpenAI warrants tied to future AMD chip purchases.
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 Becomes Market Risk
For readers, the issue is less the loaded label than the dependency it describes. If model developers rent most of their training and inference capacity, their costs, product road maps and bargaining power are shaped by lease terms, chip allocation and credit markets outside their direct control.
The structure also matters for investors and enterprise buyers because revenue can look strong while it rests on a narrow chain of counterparties. A cancelled lease, delayed data center or cut in GPU rental rates can become another company’s missing revenue. The report says H100 rental rates have fallen 60 percent to 75 percent from peak levels, a sign that some scarce assets may be turning into lower-margin capacity faster than expected.
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.
GPU Shortage Created Neoclouds
Neoclouds are AI-focused infrastructure companies that rent GPU capacity without the broad legacy cloud business of Amazon, Microsoft or Google. The category expanded during the 2024 and 2025 GPU shortage, when even well-funded AI labs faced long waits for Nvidia accelerators and power-ready data centers.
CoreWeave is the best-known public company in the category. Thorsten Meyer AI says its contracted backlog is above $55 billion, with major commitments from Meta and OpenAI. The report also cites Nebius, Crusoe, Lambda, Together, Fireworks, Nscale and IREN as part of the wider AI compute rental market.
“The cartel isn’t a conspiracy; it is the endpoint of extreme capital intensity, real scarcity, and one dominant supplier.”
— Thorsten Meyer AI, The Control Series
AI training GPU rentals
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Deal Terms Still Need Proof
Several central details are still not settled. It is not yet clear how much of the reported trillion-dollar-plus compute commitment can be financed, how many supplier investments will close on announced terms, or how durable demand will be if consumer AI subscriptions remain limited. Thorsten Meyer AI says roughly 3 percent of consumers pay for AI, but that figure depends on market definitions and may shift quickly.
The legal meaning of the word cartel is also unproved here. The report uses it to describe circular financing, concentration and mutual dependence, not a confirmed finding of unlawful collusion. Regulators, lenders and public-market investors may treat those questions differently as more filings arrive.
Mini Data Center: Build & Profit From AI at Home – No Experience Needed
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Watch Orders And Capacity
The next checkpoints are public filings, earnings calls and delivery milestones for Colossus, CoreWeave, Oracle, Microsoft, AMD and Nvidia-linked projects. Investors will watch whether announced GPU orders become deployed capacity, whether cancellation rights are used, and whether customers keep paying for reserved clusters as new chips arrive.
For companies building on AI models, the practical step is supplier risk review. The report argues for owning some inference capacity, keeping open-weight fallback models, and avoiding reliance on a single chip, cloud or lab-controlled compute source.
Nimo AI NAS, Agentic Computer Mini PC and AI Server, AMD Ryzen 7 PRO 8845HS(up to 5.1 GHZ, beat i5-1235u) up to 132TB ZFS Hybrid Storage, Dual 10GbE for 24hr AI Agent
[Local AI Inference & 70B Model Ready] Equipped with the AMD Ryzen 7 PRO 8845HS processor, NEXUS is…
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 for training or running AI models. These companies sell access to large clusters, usually built around Nvidia chips, rather than offering the full range of services found in older cloud platforms.
Did the report prove illegal collusion?
No. The report uses cartel as a description of concentration, circular financing and mutual dependence. It does not present a confirmed legal finding that the companies colluded unlawfully.
Why would AI labs rent compute from rivals?
AI training and inference require large numbers of GPUs, power and data center space. Renting can be faster than building, especially when a rival or adjacent company has unused capacity and a lab needs access quickly.
Who benefits most from the rental loop?
The report points most directly to Nvidia, because many of the buildouts depend on Nvidia GPUs and related systems. Neoclouds and data center operators can also benefit if long-term contracts are delivered and paid.
What could weaken the compute loop?
Lower GPU rental prices, cancelled leases, delayed power connections, weaker AI revenue or tighter credit could all strain the model. Since many deals depend on other deals, one large cancellation could affect another company’s expected revenue.
Source: Thorsten Meyer AI
