TL;DR
A new analysis says the AI industry’s compute market is increasingly built on leases, supplier financing and cross-investment among a small group of labs, chipmakers and cloud providers. The reported deals include xAI renting capacity to Anthropic and Google, Nvidia backing major buyers, and OpenAI signing vast multi-year commitments. The central risk is whether revenue, valuations and GPU demand depend too heavily on the same closed circle of companies.
A new analysis from Thorsten Meyer AI says the AI industry’s compute market has become a tightly linked network of leases, investments and purchase commitments, with frontier labs renting GPU capacity from rivals and suppliers financing customers who buy their hardware. The report matters because compute access is one of the main limits on AI model development, and the money behind that access now appears concentrated among a small group of labs, chipmakers, cloud providers and GPU landlords.
The report’s main claim is that leading AI companies increasingly do not own the infrastructure used to train and serve their models. Instead, they rent from specialized “neocloud” providers such as CoreWeave and from other AI companies with excess capacity. Thorsten Meyer AI cites reported 2026 leases under which xAI rented its Colossus 1 supercomputer to Anthropic for about $1.25 billion a month and to Google for about $920 million a month after the cluster was said to be underused.
The analysis also points to large, multi-year compute and hardware commitments by OpenAI, including reported deals with Broadcom, Oracle, Microsoft, Nvidia, AMD, AWS and CoreWeave. The report says those commitments add up to roughly $1.15 trillion over the next decade, while stressing that these are reported commitments rather than cash already spent or guaranteed revenue.
Thorsten Meyer AI frames the structure as circular: chipmakers and cloud suppliers invest in AI labs, the labs commit to buying or renting compute from those suppliers, and those commitments support supplier valuations and financing. The report cites Nvidia’s agreement to invest up to $100 billion in OpenAI, Nvidia’s equity exposure to CoreWeave and other infrastructure firms, AMD warrants granted to OpenAI, and combined Microsoft and Nvidia commitments to Anthropic as examples of that loop.
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 development matters because AI progress is tied not only to model design and talent, but also to who can access enough GPUs at the right time. If a small set of chip suppliers, cloud platforms and neocloud operators controls that access, smaller AI companies may face higher barriers to entry and less bargaining power.
The report also raises financial questions. If one company’s infrastructure revenue depends on another company’s model growth, and that buyer is backed by the supplier it pays, the health of the market can become harder to judge from headline revenue or backlog figures alone. A cancelled order, delayed data center or weaker-than-expected AI demand could affect several firms at once.
For customers, the issue is practical as well as financial. Businesses building on AI services may need to know whether their providers depend on a single GPU supplier, a single cloud contract or a rival’s leased compute. That could affect pricing, availability and long-term reliability.
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.
How Neoclouds Gained Ground
Neoclouds grew quickly during the 2024-25 GPU shortage, when many AI labs faced long waits for high-end chips and data center capacity. The model is simple: specialized providers buy large quantities of AI hardware and rent access to labs and enterprises that need training or inference capacity.
CoreWeave is the best-known example in the source material. Thorsten Meyer AI says the company has been publicly traded since 2025 and has a contracted backlog above $55 billion, with major reported commitments from Meta and OpenAI. Other named providers include Nebius, Crusoe, Lambda, Together, Fireworks, Nscale and IREN.
The report argues that xAI’s role as a landlord marks a shift. xAI is not only an infrastructure company; it is also a frontier model developer competing with other labs. If the reported Anthropic and Google leases are accurate, they show how ownership and use of AI compute have split apart, even among companies trying to build their own full-stack AI systems.
“Almost no one racing to build AI owns the machine it runs on.”
— Thorsten Meyer AI report
Kill Command
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Debt And Demand Remain Open
Several points remain uncertain. The reported commitments are often multi-year and may depend on buildout schedules, financing conditions, chip supply, customer demand and contract terms that are not fully public. It is also unclear how much of the headline value will become actual cash revenue.
The report cites falling H100 rental rates from peak levels and says only a small share of consumers pay directly for AI services. Those figures point to possible pressure on the rental market, but they do not by themselves prove that demand is too weak to support the current buildout.
The word “cartel” in the report is an interpretation of market structure, not a confirmed finding by regulators. No antitrust authority is cited in the source material as having made such a determination.
enterprise GPU leasing solutions
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Contracts Face Market Tests
The next test is whether AI demand grows fast enough to support the reported compute commitments, data center construction and supplier financing. Investors will watch whether companies convert backlog into revenue, keep utilization high and avoid renegotiating large leases if GPU rental prices keep falling.
Regulators may also look more closely at supplier financing, exclusive access, chip allocation and cross-ownership if these arrangements appear to limit competition. For AI customers, the near-term takeaway is to track provider dependence on a few hardware suppliers and keep backup options for inference and model deployment.
MINISFORUM MS-02 Ultra Mini Workstation PC, Intel Core Ultra 9 285HX (24C/24T, up to 5.5GHz), PCIe 5.0 x16, 4× DDR5 Slots, 4× M.2 Slots, USB4 v2 80Gbps, Dual 25GbE+10GbE+2.5GbE, Wi-Fi 7, 350W PSU
High-Performance AI Processor: The MS-02 Ultra features an Intel Core Ultra 9 285HX (24C/24T, up to 5.5 GHz,…
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 a specialized AI infrastructure provider that rents GPU capacity, usually for model training or inference, without offering the full range of services found in older general-purpose cloud platforms.
Did the report prove an illegal cartel exists?
No. The report uses “cartel” as a description of a concentrated, circular market structure. The source material does not cite a legal finding by regulators.
Why is Nvidia central to the story?
The report says Nvidia captures much of the value in AI data center buildouts because its GPUs remain central to large-scale AI training. It also cites Nvidia investments and financing links with several companies that buy or rent AI compute.
Why would one AI lab rent compute from a rival?
According to the report, scarcity, speed and utilization can make leasing practical. If one lab has idle capacity and another needs GPUs quickly, both sides may have a financial reason to make a deal, even if they compete in model development.
What should readers watch next?
Watch GPU rental prices, data center utilization, debt financing, contract revisions and whether AI revenue grows fast enough to support the large compute commitments reported across the industry.
Source: Thorsten Meyer AI
