TL;DR
A Thorsten Meyer AI report argues that the AI compute market has become increasingly circular, with labs, chipmakers and GPU cloud providers financing one another while renting capacity across competitive lines. The report says the pattern matters because multiyear commitments, supplier financing and falling GPU rental rates could expose the sector if projected AI demand does not keep pace.
A new Thorsten Meyer AI analysis says the AI industry’s compute market has become increasingly circular, with frontier labs, chipmakers and GPU cloud providers renting capacity from one another and financing purchases inside a small group of firms. The report matters because it frames compute, not models or apps, as the chokepoint shaping who can compete in advanced AI.
The report, titled The Neocloud Cartel, says many leading AI companies do not own the full infrastructure they rely on for training and inference. Instead, they rent large blocks of GPU capacity from so-called neocloud providers, a category of AI-focused infrastructure companies that rent Nvidia-heavy clusters to labs and enterprise customers.
Thorsten Meyer AI identifies CoreWeave as the largest example in the category, citing a contracted backlog above $55 billion and major reported commitments from Meta and OpenAI. It also lists Nebius, Crusoe, Lambda, Together, Fireworks, Nscale and IREN among companies competing to rent similar GPU infrastructure.
The report’s sharpest claim is that the market now contains a self-reinforcing loop: AI labs commit to huge compute purchases, suppliers and investors help finance those customers, and the resulting orders support the valuations of the companies providing the capacity. The analysis says Nvidia sits at the center because its chips remain the core input for most advanced AI buildouts, while the company also holds or has agreed to investments tied to several major buyers and infrastructure providers.
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 Shapes AI Competition
The report argues that access to compute has become a gatekeeping force in AI. If the largest labs must rent from a small group of infrastructure providers, and those providers depend on a narrow set of chip suppliers, competitive power may concentrate even when many companies appear to be building models.
For readers and businesses using AI products, the issue is not only who has the best model today. It is whether the companies behind those systems can afford the infrastructure needed to keep improving them, serve customers reliably and avoid being locked into costs they cannot sustain.
The analysis also points to a financial risk. It says reported compute commitments, including OpenAI’s multiyear deals with suppliers such as Broadcom, Oracle, Microsoft, Nvidia, AMD, AWS and CoreWeave, add up to figures far beyond current revenue levels. Those commitments are reported as multiyear obligations or planned spending, not cash already paid.
PNY NVIDIA RTX A6000
- CUDA Cores: Double-speed FP32 processing for workflows
- RT Cores: Up to 2X throughput for ray tracing
- Tensor Cores: Up to 5X AI training throughput
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
GPU Shortages Created New Landlords
The neocloud model grew out of the 2024 and 2025 GPU shortage, when AI labs faced long waits for advanced chips and large clusters. Renting capacity offered a faster route than building new data centers from scratch.
Thorsten Meyer AI says the pattern expanded in 2026 when xAI reportedly leased capacity from its Colossus 1 supercomputer to Anthropic for about $1.25 billion a month and to Google for about $920 million a month. According to the report, the cluster had low utilization after xAI moved some Grok training elsewhere, turning a frontier AI lab into a compute landlord for other major AI companies.
The report describes this as a shift in ownership and use: companies may own or finance infrastructure, but the capacity can be resold, leased or redirected depending on demand, utilization and contract terms.
“Almost no one racing to build AI owns the machine it runs on.”
— Thorsten Meyer AI
Risks Depend On Future Demand
Several details remain uncertain. The report relies on reported commitments and financing arrangements, many of which span years and may depend on future milestones, customer demand, chip availability and data center construction.
It is not yet clear how much of the reported spending will become actual cash outlay, how much capacity will be used profitably, or whether AI revenue will grow quickly enough to support the infrastructure being financed. The report also says H100 rental rates have fallen from their peak, which could pressure providers that raised capital assuming higher pricing.
The term “cartel” is the report’s characterization. The source material does not establish an illegal agreement among companies, and it frames the structure as a result of scarcity, capital intensity and supplier dominance.
Orders, Utilization And Revenue Tests
The next test is whether labs can convert compute access into enough paying usage to justify the scale of their commitments. Investors will watch data center utilization, customer concentration at neocloud providers, chip delivery schedules, and whether large AI labs can fund their obligations without relying on more supplier-backed financing.
The report says the same circular structure that supports rapid buildouts could become fragile if one major buyer delays or cancels orders. In that case, one company’s reduced spending could become another company’s missing revenue.
Key Questions
What is a neocloud?
A neocloud is an AI-focused cloud provider that rents GPU capacity, often for model training or inference, without operating as a broad general-purpose cloud like AWS, Microsoft Azure or Google Cloud.
Is the report alleging illegal collusion?
No. The report uses “cartel” as a description of market concentration and circular financing. It does not present evidence of an illegal agreement among the companies named.
Why is Nvidia central to the report?
The analysis says Nvidia captures a large share of AI data center spending because its GPUs remain the preferred hardware for many advanced AI workloads. It also cites Nvidia investments and financing arrangements involving major AI and infrastructure companies.
What could weaken the compute loop?
Lower rental prices, unused capacity, customer concentration, construction delays or slower-than-expected AI revenue growth could all weaken the economics described in the report.
Source: Thorsten Meyer AI
