TL;DR
A new analysis argues that the AI compute market has become a tightly linked rental and financing loop among labs, chipmakers and neocloud providers. The report says Nvidia remains the main chokepoint, while frontier AI companies increasingly rent capacity from specialized GPU landlords and, in some cases, from direct competitors.
A new analysis from Thorsten Meyer AI says the artificial intelligence industry’s race for compute is now dominated by a small circle of labs, chipmakers and GPU rental firms that buy, lease and finance capacity among themselves, raising questions about who controls access to the hardware needed to train and run advanced models.
The report, titled The Neocloud Cartel, describes a market in which major AI developers often do not own the infrastructure they use. Instead, they rent graphics-processing capacity from specialized “neocloud” providers such as CoreWeave and other GPU-focused firms, or from companies that also compete with them in AI model development.
The analysis says the most striking reported example is xAI leasing capacity from its Colossus 1 supercomputer to Anthropic for about $1.25 billion a month and to Google for about $920 million a month after Grok training moved elsewhere and utilization reportedly fell to 11%. Those figures are attributed in the source material to reporting and filings cited by Thorsten Meyer AI; the terms and operational limits of those leases remain partly opaque.
The piece also points to large multiyear commitments by OpenAI, including reported compute and hardware deals with Broadcom, Oracle, Microsoft, Nvidia, AMD, AWS and CoreWeave. The analysis says those commitments total roughly $1.15 trillion over the next decade, while stressing that they are reported commitments rather than cash on hand.
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 Control Shapes AI Power
The report matters because access to advanced GPUs has become one of the main limits on AI development. If a small set of firms controls chip supply, datacenter capacity and financing, then the practical ability to train frontier systems may depend less on model research alone and more on infrastructure contracts.
For customers and smaller AI companies, the risk is being locked into compute pricing and availability set by suppliers above them in the stack. The analysis warns that companies relying entirely on rented inference or training capacity could become price-takers if GPU supply tightens again or if large buyers absorb available capacity first.
The financing loop also matters for investors. Thorsten Meyer AI argues that supplier investments, customer warrants and capacity backstops can lift valuations across the same circle of companies. That can support rapid expansion, but it can also spread losses if one large buyer cancels or delays orders.
GPU cloud rental services
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Neoclouds Grew From GPU Scarcity
Neocloud providers emerged during the 2024 and 2025 GPU shortage, when demand for Nvidia accelerators outpaced supply and AI labs faced long waits for capacity. These firms built businesses around renting AI-focused GPU clusters rather than offering the broad services of a general-purpose cloud provider.
CoreWeave is presented in the analysis as the largest example, with a contracted backlog described as north of $55 billion and major reported commitments from Meta and OpenAI. Other named providers include Nebius, Crusoe, Lambda, Together, Fireworks, Nscale and IREN.
The report says Nvidia sits at the center of the system because its chips capture a large share of spending in AI datacenter buildouts. It also says Nvidia has invested in or financially supported several buyers and infrastructure providers, including OpenAI, CoreWeave, Nebius and Applied Digital, while also using capacity pre-purchase arrangements and other financing structures.
HPE NVIDIA Tesla V100 32GB HBM2 PCIe 3.0 x16 Passive GPU Computational Accelerator for AI Machine Learning HPC Deep Learning 699-2G500-0216-400 (Renewed)
NVIDIA Volta GV100 Architecture — 5,120 CUDA Cores, 640 1st-Gen Tensor Cores delivering 14 TFLOPS FP32 and 112…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Lease Terms Remain Opaque
Several parts of the compute market described in the report remain hard to verify from public information. The exact terms of private capacity leases, supplier financing agreements, utilization rates and cancellation rights are often not fully disclosed.
It is also not clear how much of the reported multiyear spending will turn into actual purchases. The analysis says many headline figures are commitments spread over years, not current cash reserves or guaranteed near-term revenue.
The report’s use of the word “cartel” is an interpretation of market concentration and circular financing. It does not present confirmed evidence of illegal price-fixing or formal collusion among the companies named.
msi GeForce RTX 5080 Triple Fan OC Graphics Card, 16GB GDDR7, 30 Gbps, 256-bit, Boost 2640 MHz, NVIDIA Blackwell, DLSS 4, AI Content Creation, Local LLM Inference, PCIe 5.0, DP 2.1b x3, HDMI 2.1b
[NVIDIA Blackwell, DLSS 4, 4th Gen Ray Tracing Cores] Built for high fidelity gaming and creator workloads with…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Orders Will Test The Loop
The next test is whether AI revenue grows fast enough to support the scale of compute commitments now being signed. The analysis points to projections of heavy datacenter spending from 2025 through 2028, while also citing pressure from falling H100 rental rates and limited consumer willingness to pay for AI services.
Readers should watch for changes in GPU rental prices, canceled or delayed datacenter orders, new chip supply from AMD and custom silicon, and disclosures from major neocloud providers about customer concentration. Any weakening in one part of the loop could affect suppliers, landlords and AI labs at the same time.
neocloud GPU leasing platforms
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is a neocloud in this report?
A neocloud is described as an AI-focused cloud provider that rents GPU capacity, usually Nvidia-based clusters, to companies training or running AI models.
Does the report prove an illegal cartel?
No. The source uses “cartel” as a description of concentrated market power and circular financing. It does not claim to prove illegal collusion or price-fixing.
Why is Nvidia central to the analysis?
The report says Nvidia captures much of the spending in AI datacenter buildouts because its GPUs remain the dominant hardware for frontier AI workloads. It also points to Nvidia investments and financing links with several buyers and compute providers.
Why would AI rivals rent compute from each other?
The analysis says the reason is capacity. If a company has an underused cluster and another lab needs GPUs quickly, leasing can turn idle hardware into revenue even when the parties compete elsewhere.
What is the main risk for smaller AI companies?
The main risk is dependence on rented capacity controlled by larger suppliers and landlords. If prices rise, capacity tightens or contract terms change, smaller firms may have fewer options for training and serving models.
Source: Thorsten Meyer AI
