TL;DR
Thorsten Meyer AI’s latest Control Series report says frontier AI compute is increasingly rented through a small set of neoclouds, labs and chip suppliers. The report frames the pattern as a circular financing loop, while stressing that many figures are reported multi-year commitments rather than cash already spent.
Thorsten Meyer AI’s new Control Series report says the AI industry’s compute layer is increasingly built on a narrow web of rental, financing and chip-purchase deals linking frontier labs, neocloud providers and Nvidia, a pattern the report argues could make the sector more exposed if demand or funding slows.
The report, titled The Neocloud Cartel, focuses on companies that rent GPU capacity to AI labs rather than selling general-purpose cloud services. It identifies CoreWeave, Nebius, Crusoe, Lambda, Together, Fireworks, Nscale and IREN among the firms competing to supply Nvidia-based infrastructure to labs that need large training and inference clusters.
Thorsten Meyer AI says the strongest recent example is xAI, which the report says leased 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 fell. Those figures are attributed in the source material to prior reporting and filings, but the full terms of the leases have not been independently detailed in the provided material.
The report also cites OpenAI’s reported compute and hardware commitments of roughly $1.15 trillion over the next decade across suppliers including Broadcom, Oracle, Microsoft, Nvidia, AMD, AWS and CoreWeave. It says Nvidia has agreed to invest up to $100 billion in OpenAI, holds stakes in several infrastructure companies, and has used capacity commitments and financing arrangements that support buyers of its own chips.
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 Dependence Shapes AI Power
The report matters because compute access now affects which AI companies can train frontier models, serve users at scale and negotiate from strength. If leading labs mostly rent capacity, the owners and financiers of that infrastructure can influence pricing, supply access and the pace of model development.
Thorsten Meyer AI argues that the arrangement is not evidence of an illegal conspiracy. Its central claim is that high capital costs, real GPU scarcity and Nvidia’s market position have produced a tightly linked market in which suppliers, customers and investors increasingly overlap.
That overlap could also create risk. The report says each major order becomes revenue for another participant in the same circuit, meaning a delayed model rollout, cancelled lease or weaker end-user demand could affect multiple companies at once.
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 Filled GPU Shortages
The report traces the rise of neocloud providers to the 2024 and 2025 GPU shortage, when even well-funded labs faced long waits for large Nvidia clusters. Renting from specialist providers became a faster route to capacity than building data centers and procurement pipelines from scratch.
CoreWeave is presented as the category’s largest example, with a contracted backlog described as north of $55 billion and major commitments from Meta and OpenAI. The report says the broader neocloud group is backed by venture capital, private equity and sovereign money, but often rents out similar Nvidia hardware.
The source material frames this as the second installment of a six-part Control Series on AI power centers. Its first part identified compute as one of the main chokepoints in the AI stack; this report narrows in on how GPU access, cloud leases and supplier financing interact.
“Almost no one racing to build AI owns the machine it runs on.”
— Thorsten Meyer AI report
AI training GPU rental services
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Deal Terms Remain Opaque
Several numbers in the report are described as reported commitments, often spread across years, rather than current cash spending. That means the headline totals may change if companies revise buildout plans, renegotiate contracts or fail to raise enough capital.
It is also unclear how much of the reported compute demand will be matched by paid customer use. The report cites estimates that only about 3% of consumers pay for AI, while also pointing to large projected operating losses and falling H100 rental rates from prior peaks.
The degree of control created by lease terms is another open issue. The report refers to an xAI lease provision that reportedly lets Elon Musk reclaim compute if Claude “harms humanity,” but the supplied material does not include the full contract text or confirmation from all parties.
ASUS Ascent GX10 AI Supercomputer, DGX Spark, NVIDIA GB10 Superchip, 128GB LPDDR5x, 1TB PCIe Gen4 NVMe SSD, Wi-Fi 7 & BT5.4, Agentic AI Ready, Supports OpenClaw, NemoClaw, Stackable Chassis
Extreme AI Performance: Powered by NVIDIA GB10 Grace Blackwell Superchip delivering 1 petaFLOP of AI performance and 128GB…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Orders Will Test Demand
The next test is whether labs can turn reported compute commitments into sustainable revenue while infrastructure providers keep clusters filled. Investors and customers will be watching utilization rates, rental pricing, financing terms and whether large suppliers keep backing their own buyers.
If AI usage and enterprise adoption rise fast enough, the compute loop may keep funding new clusters. If demand lags, the report’s warning is that one cancelled or delayed order could become another company’s missing revenue.
cloud compute leasing for AI labs
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 June 2026 report arguing that AI compute is increasingly controlled through a small network of rental, chip-purchase and financing deals among labs, neoclouds and suppliers.
What is a neocloud?
In the report’s usage, a neocloud is an AI-focused cloud provider that rents GPU capacity, usually for training or inference, without operating as a broad general-purpose cloud like AWS, Azure or Google Cloud.
Is the report alleging illegal cartel behavior?
No. The report uses “cartel” as a market-structure description, while saying the pattern stems from capital intensity, GPU scarcity and supplier dominance rather than proven illegal coordination.
Why is Nvidia central to the report?
The report says Nvidia captures a large share of AI data center spending through GPU sales, holds stakes in several buyers or infrastructure firms, and can influence allocation during periods of chip scarcity.
What remains unconfirmed?
Many deal totals, lease terms and utilization figures come from reported commitments and outside reporting cited by the source material. The final spending, contract details and long-term demand for the compute remain uncertain.
Source: Thorsten Meyer AI
