TL;DR
Recent filings and infrastructure deals cited by Thorsten Meyer AI show frontier AI companies renting large-scale compute from rivals, cloud providers and chip suppliers rather than owning all of the machines they use. The confirmed pattern is a tight loop of GPU leases, supplier financing and multi-year commitments; the open question is whether demand can support the scale of those promises.
SpaceX’s IPO materials and recent AI infrastructure announcements have put new focus on a circular compute market: frontier AI labs are renting GPU capacity from rivals and suppliers while those same suppliers help finance parts of the buildout. The pattern matters because access to Nvidia-heavy compute now shapes which companies can train and serve leading AI models, and because the money, debt and chip orders are increasingly tied together.
The sharpest example is Anthropic’s reported lease of SpaceX/xAI Colossus capacity. MarketWatch, Business Insider and Axios, citing SpaceX filing materials, reported in May 2026 that Anthropic agreed to pay about $1.25 billion a month for access to Colossus data-center capacity, with terms that could run through May 2029 but include cancellation rights. Elon Musk later said the arrangement was a shorter 180-day lease with a 90-day mutual cancellation notice, leaving the exact practical duration dependent on how the parties use those terms.
Thorsten Meyer AI’s Control Series frames the deal as part of a wider neocloud shift. Neoclouds are AI-focused cloud providers that rent GPU clusters to labs and enterprises. CoreWeave is the best-known public example, while firms such as Nebius, Crusoe, Lambda, Together, Fireworks, Nscale and IREN are also competing to sell access to Nvidia-based systems. The analysis says CoreWeave has a contracted backlog above $55 billion and that major customers, including Meta and OpenAI, have made large multi-year commitments.
The financing loop extends beyond leasing. Nvidia announced in September 2025 that it could invest up to $100 billion in OpenAI, while OpenAI committed to buy Nvidia systems for new data centers. AMD’s OpenAI deal included warrants that could make OpenAI a major AMD shareholder. Microsoft and Nvidia later announced up to $15 billion of investment tied to Anthropic’s planned purchase of $30 billion in Azure compute. These are reported commitments and strategic arrangements, not proof that all cash has already changed hands.
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 Deals Shape AI Prices
The compute market now affects far more than the AI labs themselves. If a small group of chipmakers, hyperscalers and neocloud landlords controls access to the machines needed for frontier models, the cost of AI products, the pace of model releases and the ability of smaller firms to compete can all be shaped by capacity allocation.
The circular structure also creates financial risk. Supplier investment can help customers buy more chips, and customer contracts can help landlords raise debt for more data centers. That can add capacity quickly, but it also means one canceled lease or delayed order can become another company’s missing revenue. Falling H100 rental rates, cited by the Thorsten Meyer AI analysis as down 60% to 75% from peak levels, add pressure to a model built around expensive hardware and long-term demand forecasts.
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 Built Neoclouds
The neocloud market grew out of the 2024 and 2025 GPU shortage, when even well-funded AI labs faced delays getting enough Nvidia accelerators. Renting capacity became faster than building new data centers, securing power, buying racks and hiring infrastructure teams.
That shortage made firms such as CoreWeave valuable because they could assemble and finance large GPU clusters quickly. It also pushed labs to diversify across Microsoft, Oracle, AWS, Google, Nvidia, AMD and specialist providers. The new development is that some AI companies are no longer just renters. They are also landlords, investors or counterparties in deals with direct competitors.
“Almost no one racing to build AI owns the machine it runs on.”
— Thorsten Meyer AI Control Series
AI training GPU rental services
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Lease Terms Remain Opaque
Several material details remain unclear. Public reports describe large monthly lease figures, multi-year windows and cancellation rights, but the practical minimum payments, utilization guarantees, power-cost exposure and performance obligations are not fully public.
It is also uncertain how much of the reported trillion-dollar-scale compute demand will convert into actual spending. The Thorsten Meyer AI analysis cites OpenAI commitments of about $1.15 trillion over the next decade, but those figures are described as reported multi-year commitments, not cash on hand. The durability of consumer and enterprise AI revenue remains a live test for the entire financing chain.
reComputer J3011 – Edge AI Computer with NVIDIA Jetson Orin Nano 8GB (Support Super Mode
Brilliant AI Performance for production: The reComputer J3011 is equipped with the same NVIDIA Jetson Orin Nano 8GB…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Investors Track Cancellations And Demand
The next signals will come from filings, earnings reports and customer announcements. Investors will watch whether Colossus and other large clusters stay highly utilized, whether OpenAI, Anthropic and Meta keep expanding compute commitments, and whether rental prices keep falling as more GPU capacity comes online.
Regulators may also examine whether supplier investments and compute commitments limit competition or reinforce Nvidia’s position in the AI stack. For buyers of AI services, the near-term issue is practical: pricing, uptime and model access may depend on infrastructure deals that remain only partly visible from the outside.
cloud GPU leasing for AI development
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, often Nvidia-based clusters, to companies building or running AI models. It differs from a general cloud provider by centering its business on AI compute rather than broad cloud services.
Is the word cartel a legal allegation here?
No. In this article, cartel refers to Thorsten Meyer AI’s market-structure critique of a small, closely linked group of compute buyers, landlords and suppliers. It is not a legal finding of collusion.
Who is renting compute from whom?
Reports cited in the analysis say Anthropic is leasing Colossus capacity from SpaceX/xAI. OpenAI, Anthropic, Meta and others also have large compute commitments with cloud providers, chipmakers or neocloud firms such as CoreWeave.
Why is Nvidia central to the story?
Nvidia supplies the dominant GPUs used in large AI training and inference clusters. The company also has investment or financing links to major AI infrastructure customers, which places it near both the supply and funding sides of the market.
What could break the compute loop?
A demand slowdown, canceled leases, lower GPU rental prices, financing strain or faster adoption of rival chips could weaken the loop. The main unknown is whether AI revenue grows fast enough to support the scale of infrastructure commitments already announced.
Source: Thorsten Meyer AI
