TL;DR
Reported 2026 compute deals show AI labs increasingly renting GPU capacity from specialized cloud firms and, in some cases, from direct rivals. The pattern matters because huge multi-year commitments, supplier financing and falling rental prices could leave the sector exposed if demand or funding slows.
AI companies are increasingly renting the computing power they need from specialized GPU providers and, in some cases, from direct competitors, a pattern highlighted by reported 2026 leases of xAI supercomputer capacity to Anthropic and Google and by large supplier-backed financing deals across the sector.
The source material says the AI compute market has shifted toward “neoclouds,” companies built mainly to rent GPU capacity for AI training and inference. CoreWeave is described as the largest player in that category, with a contracted backlog above $55 billion and reported commitments from Meta and OpenAI.
The most striking reported deal came in May 2026, when xAI 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. The source material says xAI’s own Grok training had moved elsewhere and the cluster was running at about 11% utilization.
The report also says OpenAI has made roughly $1.15 trillion in multi-year compute and hardware commitments across suppliers including Broadcom, Oracle, Microsoft, Nvidia, AMD, AWS and CoreWeave. Those figures are described as reported commitments, not cash already paid or guaranteed revenue.
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 Deals Concentrate Risk
The deals matter because AI capacity is becoming both the industry’s main input and one of its biggest financial dependencies. If frontier labs rent rather than own their most expensive infrastructure, the terms of access, pricing and allocation can shape who can train large models and who cannot.
The source material argues that the market is not a cartel in the legal sense but a tight loop created by scarce GPUs, high capital costs and supplier financing. Nvidia is described as the central upstream supplier, capturing a large share of each data center buildout dollar while also holding stakes or financing arrangements with major buyers.
That circular structure can support rapid expansion while demand is rising. It can also amplify stress if orders are delayed, rental prices fall or major labs reduce capacity plans. The source material cites reported H100 rental prices down 60% to 75% from peak levels, a sign that some once-scarce capacity may be becoming easier to obtain.
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.
How Neoclouds Became Landlords
Neoclouds grew out of the 2024 and 2025 GPU shortage, when companies needed large clusters quickly and could not wait years to build data centers themselves. Firms such as CoreWeave, Nebius, Crusoe, Lambda, Together, Fireworks, Nscale and IREN rent access to AI hardware, often based on Nvidia GPUs.
The source material says CoreWeave drew 77% of its revenue from two customers, a concentration that shows how closely the fortunes of some infrastructure providers are tied to a small group of AI buyers. It also says Nvidia has invested in or supported companies that buy or rent its chips, including OpenAI, CoreWeave, Nebius and Applied Digital.
AMD is also part of the financing loop described in the material. OpenAI’s AMD deal reportedly gives OpenAI warrants for up to 160 million AMD shares at a penny each, making the customer a potential major shareholder if the agreement’s conditions are met.
AI Data Center Infrastructure Engineering: Power Distribution, Liquid Cooling, High-Density Networking, and Energy Efficiency for GPU Training … Hardware & Compiler Engineering Series)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Demand Still Has Open Questions
It is not yet clear how much of the reported multi-year commitment total will become actual cash spending, how quickly the capacity will be used, or whether customer demand will support the scale of planned buildouts. The source material says figures are reported commitments and often multi-year, not cash on hand.
There is also uncertainty around utilization. The reported xAI lease is framed as a response to unused Colossus 1 capacity, but the longer-term use rate of new AI data centers remains unclear. If labs overbook capacity, neoclouds could face lower rental prices or weaker renewal terms.
The broader legal and regulatory meaning of the “cartel” label is also not established by the source material. The article uses the term to describe market concentration and circular financing, not a confirmed finding of unlawful collusion.
nVidia Tesla K80
New Fujitsu S26361-F222-L81 NVIDIA Tesla K80
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Capacity Contracts Face Tests
The next test is whether major AI labs can turn rented compute into enough revenue to support their obligations. Investors and customers will be watching utilization rates, financing terms, chip delivery schedules and whether large buyers keep expanding or begin canceling orders.
More scrutiny is also likely around supplier-backed deals, especially where chipmakers, cloud providers and AI labs finance one another. If rental prices keep falling, the strongest evidence will come from contract renewals, disclosed revenue concentration and any changes to planned data center spending.
GPU rental for AI inference
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 a cloud provider focused mainly on renting GPU capacity for AI workloads, without the broader legacy cloud business of companies such as Amazon, Microsoft or Google.
Did xAI rent compute to Anthropic?
According to the source material, xAI leased Colossus 1 capacity to Anthropic for about $1.25 billion a month and to Google for about $920 million a month in May 2026.
Why is Nvidia central to this story?
The source material describes Nvidia as the dominant upstream chip supplier, with revenue exposure to data center buildouts and investments or financing ties to several buyers of AI compute.
Is this financial advice for AI investors?
No. The figures are reported historical commitments and market estimates, not guarantees of future revenue, profit or investment returns.
Source: Thorsten Meyer AI
