TL;DR
Reports cited by Thorsten Meyer AI say xAI leased capacity from its Colossus 1 supercomputer to Anthropic and Google in May 2026, turning a frontier AI lab into a compute landlord for rivals. The development adds to concern that AI infrastructure spending is becoming circular, with labs, cloud providers and chipmakers financing each other’s growth.
xAI reportedly leased major capacity from its Colossus 1 supercomputer to Anthropic and Google in May 2026, a development that shows how leading AI companies are increasingly renting scarce compute from rivals while much of the spending flows back to chip suppliers and specialized GPU cloud firms.
According to source material from Thorsten Meyer AI, xAI leased Colossus 1 capacity to Anthropic for about $1.25 billion a month and to Google for about $920 million a month after Grok training moved elsewhere and the cluster was running at about 11% utilization. The report frames the deals as a sign that ownership and use of AI infrastructure have split: even companies building their own frontier models may rent out machines when their internal workloads shift.
The same report describes a wider neocloud market built around AI-only GPU rental providers. CoreWeave, which went public in 2025, is cited as having a contracted backlog above $55 billion, with reported commitments of about $35 billion from Meta and about $22 billion from OpenAI. Other firms named in the category include Nebius, Crusoe, Lambda, Together, Fireworks, Nscale and IREN.
The article also points to large multi-year compute commitments by OpenAI, including reported or estimated deals across Broadcom, Oracle, Microsoft, Nvidia, AMD, AWS and CoreWeave. The figures are described as commitments rather than cash already spent, and the report says some of the arrangements rely on supplier financing, equity stakes, warrants or capacity backstops.
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.
Rivals Are Becoming Landlords
The reported xAI leases matter because they blur the line between AI competitors and infrastructure suppliers. If Anthropic trains on xAI-owned capacity, and Google rents capacity from the same cluster, the competitive map becomes harder to read: companies may compete on models while depending on each other for the machines needed to build them.
For customers, investors and policymakers, the issue is whether AI infrastructure spending reflects durable demand or a set of linked commitments that depend on each other holding. If one lab cuts orders, delays training runs or shifts chips, that can become lost revenue for the cloud provider, chipmaker, lender or investor tied to the same buildout.
The report also keeps attention on Nvidia’s position. The source material says Nvidia captures a large share of every gigawatt-scale AI buildout through GPU sales and also holds stakes or financing ties with several buyers. That is a reported business pattern, not proof of illegal coordination, but it shows why compute access remains one of the main constraints in advanced AI.
AI GPU cloud computing services
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How Neoclouds Filled The Gap
The neocloud sector expanded during the 2024 and 2025 GPU shortage, when AI labs faced long waits for high-end accelerators and needed faster access to training clusters. These companies offer GPU capacity as a service, often focused on AI workloads rather than broad cloud products.
Thorsten Meyer AI describes the market as a response to capital intensity and scarcity. Building large AI data centers requires chips, power, real estate, networking equipment and financing on a scale that few labs can absorb directly. Renting allows labs to scale faster, but it also puts them under the pricing, capacity and contract terms of landlords.
The report says several deals include circular elements: chipmakers invest in buyers, cloud providers finance customers, customers receive warrants, and suppliers pre-purchase capacity or provide backstops. Those structures can support rapid buildout, but they can also make headline demand harder to interpret.
“Almost no one racing to build AI owns the machine it runs on.”
— Thorsten Meyer AI
Nvidia GPU server rental
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Limits Of The Public Record
Several key details remain unverified from public materials alone. The exact terms of the xAI leases, including cancellation rights, capacity guarantees, performance requirements and restrictions on model use, have not been fully disclosed in the source material.
It is also unclear how much of the reported trillion-dollar-scale compute pipeline will become actual spending. The source material says figures include multi-year commitments, not cash on hand. That distinction matters because future AI revenue, financing conditions, chip availability and power constraints could all change how much infrastructure is built.
The term “cartel” is used by the source as an analytical label for concentration and circular financing, not as a confirmed legal finding. No evidence cited in the provided material establishes unlawful collusion among the companies named.
high performance supercomputer hardware
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Watch Orders And Utilization
The next markers are data center utilization, chip orders, lease renewals and whether large model labs keep expanding training budgets at the pace implied by current commitments. Public filings from CoreWeave, Nvidia, cloud providers and AI labs will help show whether the demand is sustained or pulled forward by financing structures.
Regulators and investors are also likely to keep examining supplier financing, equity-linked customer deals and concentration in GPU supply. If rental prices keep falling from their peak, as the source material says has already happened for H100 capacity, the sector may face a sharper test of which compute landlords can stay profitable.
enterprise AI compute infrastructure
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is the actual development in this story?
The reported development is that xAI leased capacity from its Colossus 1 supercomputer to Anthropic and Google in May 2026, adding to evidence that major AI companies are renting compute from rivals and specialized GPU clouds.
What is a neocloud?
A neocloud is a specialized cloud provider focused on renting GPU capacity for AI workloads, rather than offering a full general-purpose cloud platform.
Does this prove the AI industry is operating an illegal cartel?
No. The provided source uses “cartel” as a description of concentration, circular financing and dependency, not as a confirmed legal finding of collusion.
Why does Nvidia matter so much in this story?
Nvidia supplies the dominant GPUs used in many AI training clusters and, according to the source material, also has investment or financing ties with several companies buying or renting that capacity.
What remains unclear?
The full contract terms, actual future cash spending, long-term utilization rates and durability of AI compute demand remain unclear based on the provided material.
Source: Thorsten Meyer AI
