TL;DR
Thorsten Meyer AI’s new Control Series report says frontier AI companies are relying on leased GPU capacity from neoclouds, suppliers and, in reported cases, direct rivals. The report frames that structure as a circular financing loop centered on Nvidia, while stressing that many figures are reported multi-year commitments rather than cash already spent.
Thorsten Meyer AI has published a report saying the AI compute market is increasingly built on rented GPU capacity, with frontier labs, neocloud providers, chipmakers and financiers tied together through leases, equity stakes and large future commitments.
The report defines neoclouds as AI-focused GPU rental providers that grew out of the 2024-25 shortage in advanced Nvidia chips. It says CoreWeave has become the largest company in that category, with a contracted backlog above $55 billion and major commitments from Meta and OpenAI. It also lists Nebius, Crusoe, Lambda, Together, Fireworks, Nscale and IREN among companies renting out similar Nvidia-based capacity.
The most striking reported development is xAI’s move into the landlord role. According to the report, 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 its own Grok training moved elsewhere and the cluster was underused. Those terms are presented as reported lease figures, not public filings verified in the article.
The report also says OpenAI has made roughly $1.15 trillion in future compute and hardware commitments across suppliers including Broadcom, Oracle, Microsoft, Nvidia, AMD, AWS and CoreWeave. It frames those commitments as part of a loop in which suppliers help finance customers who then buy or rent capacity from the same supplier network.
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 Scarcity Reshapes AI Power
The report matters because compute access now helps decide which companies can train, deploy and sell frontier AI systems. If GPU supply is concentrated among a few chipmakers, cloud providers and neocloud landlords, smaller AI developers may face higher costs, fewer choices and contract terms set by companies that also compete in the AI market.
For investors and customers, the risk is that headline demand may depend on circular commitments rather than broad end-user revenue. The report cites estimates of about $3 trillion in data center spending from 2025 to 2028, a projected OpenAI operating loss of about $74 billion in 2028, and consumer AI payment rates near 3%. If demand falls short, one canceled order could become another company’s missing revenue.
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.
Neoclouds Grew From GPU Shortages
Neoclouds emerged because large AI labs needed clusters quickly and could not always wait for conventional cloud buildouts. The report says renting became the practical path to scale while GPUs were scarce, power was hard to secure and data centers took years to build.
Nvidia sits at the center of the reported structure because most advanced AI clusters depend on its GPUs. The report says Nvidia captures a large share of each new buildout dollar, holds equity in several buyers or capacity providers, and has arranged financing or investment links with OpenAI, CoreWeave, Nebius, Applied Digital and xAI. It also cites AMD’s warrant deal with OpenAI as another example of suppliers turning customers into financial partners.
“Almost no one racing to build AI owns the machine it runs on. They rent.”
— Thorsten Meyer AI report
HHCJ6 Dell NVIDIA Tesla K80 24GB GDDR5 PCI-E 3.0 Server GPU Accelerator (Renewed)
Dell Nvidia Tesla K80 GPU (Nvidia Part Number: 900-22080-0000-000)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Private Deal Terms Stay Opaque
Several core facts remain hard to verify independently from the source material. The report relies on reported lease terms, projected commitments and private financing arrangements, many of which may not be fully disclosed in public filings. It is also unclear how much of the cited spending will become actual cash outlays.
The legal meaning of the word ‘cartel’ is also uncertain. The report uses it to describe market structure and circular dependence, not to assert a proven unlawful agreement among companies. It remains unclear whether regulators, creditors or public investors will treat these arrangements as ordinary infrastructure finance or as a concentration risk.
supercomputer leasing for AI training
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Lease Commitments Meet Demand Tests
The next test is whether AI revenue, cloud demand and model usage can support the scale of the reported commitments. Investors will watch public filings from neoclouds, chipmakers and cloud providers for backlog quality, customer concentration, utilization rates and financing exposure.
Further disclosures from companies such as Nvidia, CoreWeave, OpenAI, Anthropic, xAI, Oracle, AMD and Microsoft could clarify how much of the compute boom is backed by durable demand and how much depends on the same companies funding one another’s growth.
ASRock Radeon AI PRO R9700 Creator 32GB Professional Graphics Card, 2920 MHz Boost Clock, GDDR6, AMD RDNA 4, AI-Accelerators, DisplayPort 2.1a, PCIe 5.0, Blower Cooler
Professional AI & Creator Workstation: AMD Radeon AI PRO R9700 GPU with 32GB GDDR6 is engineered for AI…
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 model training or inference, without the full range of older general-purpose cloud services.
Is the report alleging an illegal cartel?
No. The report uses ‘cartel’ as a description of concentrated market power and circular financing. It does not present a legal finding that companies formed an unlawful agreement.
Why is Nvidia central to the report?
The report says most large AI clusters still depend on Nvidia GPUs, giving the company influence through chip allocation, supplier economics and equity links with customers and compute providers.
Why would an AI lab rent compute to a rival?
According to the report, unused clusters can become expensive idle assets. Renting that capacity can generate revenue even when the customer is also a competitor.
What could break the compute loop?
The report points to falling rental rates, weak paid consumer adoption, customer concentration and canceled orders as risks that could strain the system if AI revenue does not keep pace with infrastructure spending.
Source: Thorsten Meyer AI
