TL;DR
Thorsten Meyer AI’s latest Control Series report says the AI industry is relying on rented GPU capacity, including deals between direct competitors. The report argues that supplier financing, neocloud contracts and chip scarcity have created a circular market whose risks are still hard to measure.
Thorsten Meyer AI’s second Control Series installment says the AI industry has shifted toward a circular compute-rental model in which major labs, neocloud providers and chip suppliers finance and rent capacity among themselves, raising questions about who controls the infrastructure behind frontier models.
The report identifies compute as a main chokepoint in the AI stack and says many leading labs do not own the machines they use at scale. Instead, they rent GPU capacity from AI-focused cloud providers, legacy cloud companies and, in some cases, competitors.
Thorsten Meyer AI points to CoreWeave as the largest example in the neocloud category, citing a contracted backlog above $55 billion and reported commitments from Meta and OpenAI. It also cites xAI’s reported lease of 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.
The report says these arrangements are tied to a wider financing loop. It cites OpenAI compute and hardware commitments of roughly $1.15 trillion over the next decade across Broadcom, Oracle, Microsoft, Nvidia, AMD, AWS and CoreWeave, while also noting that many of those figures are multi-year 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 GPUs can decide which companies can train and serve frontier AI systems. If rented capacity is concentrated among a small group of suppliers, labs and financing partners, then model builders may depend on terms set by companies that also have strategic stakes in their success or failure.
Thorsten Meyer AI argues that the structure is not proof of an illegal conspiracy. Its analysis frames the pattern as the outcome of extreme capital costs, real GPU scarcity and dependence on Nvidia hardware. The risk, according to the report, is that each large commitment may support the next supplier’s revenue forecast, valuation or financing plan.
NComputing RX440(RDP) Thin Client Built on Raspberry Pi4 for Microsoft AVD, Windows 365 Cloud PC, RDS, Verde VDI, vSpace Pro Enterprise
Requires connection license for specific virtualization platform you intent to use (Not Included)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Neoclouds Filled The GPU Gap
Neoclouds are AI-focused cloud providers that rent GPU clusters for training and inference. The report says the category grew quickly during the 2024 and 2025 GPU shortage, when even well-funded AI labs faced long wait times for capacity.
Companies named in the report include CoreWeave, Nebius, Crusoe, Lambda, Together, Fireworks, Nscale and IREN. Thorsten Meyer AI says they are often renting out similar Nvidia-based infrastructure, funded by venture capital, private equity or sovereign-backed money.
The report also says Nvidia occupies the strongest position in the stack because it supplies much of the hardware, holds stakes in some buyers and can influence chip allocation during shortages.
“Almost no one racing to build AI owns the machine it runs on.”
— Thorsten Meyer AI
AI training GPU rentals
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Real Exposure Is Unknown
Several parts of the picture remain unclear. The report cites reported commitments, leases and financing arrangements, but it says many figures are multi-year obligations and may not represent cash already spent.
It is also unclear how durable demand will be if AI revenue growth does not match infrastructure spending. Thorsten Meyer AI cites pressure points including falling H100 rental rates, heavy projected operating losses at OpenAI and limited consumer willingness to pay for AI products, but those figures depend on market conditions that can change quickly.
NVIDIA Video Card 900-22080-0000-000 Tesla K80 24GB DDR5 PCI-Express Passive Cooling Brown Box NCNR.
Colour: brown
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Contracts Will Test Demand
The next test is whether labs can convert rented compute into enough revenue to support long-term capacity commitments. Investors and customers will be watching utilization rates, rental pricing, supplier financing terms and any signs that large orders are delayed, reduced or canceled.
For companies that rely on AI systems, the report’s practical warning is to avoid dependence on a single compute provider. It recommends owning inference where possible, keeping an open-weight fallback and diversifying silicon choices.
neocloud GPU providers
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is the actual news development?
Thorsten Meyer AI published a new Control Series analysis arguing that AI compute is increasingly rented through a circular network of labs, neoclouds, cloud providers and chip suppliers.
Is the report alleging an illegal cartel?
No. The report uses the term to describe concentrated market structure and circular incentives. It says the pattern is driven by high capital costs, GPU scarcity and dependence on one dominant chip supplier.
What is confirmed versus claimed?
The report lists named companies, reported contracts and reported financing arrangements. Its broader claim that these deals form a cartel-like compute loop is an interpretation based on those reported facts.
Why does this matter to readers?
Compute access affects which AI companies can train large models, what customers pay for AI services and how much risk sits inside the infrastructure financing behind the sector.
What remains uncertain?
It is not yet clear whether demand for AI services will be strong enough to support the scale of the reported compute commitments, or how exposed suppliers and landlords would be if major customers cut orders.
Source: Thorsten Meyer AI
