Anthropic, the artificial intelligence laboratory preparing for a landmark public offering, is moving from renting computing power to owning the infrastructure that runs it — a shift that places the company inside the same capital-intensive world once reserved for utilities and telecommunications giants. Backed by $35 billion in chip-collateralized financing from Apollo and Blackstone, and committing $50 billion toward purpose-built facilities in Texas and New York, the company is betting that control over physical infrastructure is inseparable from control over the frontier of intelligence it
Anthropic's IPO Signals Strategic Shift to Data Centre Ownership
The infrastructure expansion is expected to create 800 permanent jobs and 2,400 construction jobs in the US.
Power has emerged as the physical bottleneck that no amount of capital can easily overcome.
Mark
Why does Anthropic need to own its own data centres? Couldn't it just keep renting from Amazon and Google?
Mimi
Renting works until it doesn't. When you're pushing the frontier of AI, you can't wait for someone else's capacity to free up. You need certainty—guaranteed power, guaranteed compute, on your timeline. Ownership gives you that control.
Mark
But that $35 billion from Apollo and Blackstone—why structure it that way, with chips as collateral?
Mimi
Because chips are now worth something as financial assets. Nvidia and the investment firms realized GPUs could back a loan the same way real estate does. It's a new asset class. For Anthropic, it means cheaper financing than traditional debt.
Mark
The $50 billion commitment to Texas and New York—that's enormous. What happens if power grids can't keep up?
Mimi
That's the real risk nobody's talking about yet. You can build the buildings and buy the chips, but if the grid can't deliver the megawatts, none of it matters. Anthropic will have to disclose this in its IPO filing.
Mark
So power is the actual constraint?
Mimi
It's becoming one. Grid interconnection takes years. You can finance chips in months. That mismatch is going to force hard choices about where and how fast to build.
Mark
What does it mean that all this capital is circulating among the same six or seven companies?
Mimi
It means the AI infrastructure boom isn't as decentralized as it looks. Anthropic pays SpaceX for power, SpaceX has deals with Google, Google has deals with Anthropic. The money moves in circles. When it comes time for IPO disclosures, investors will want to know how much of this is real growth versus capital shuffling.
Mark
Will other AI companies follow Anthropic's path?
Mimi
They probably have to. Once one frontier lab owns its infrastructure, the others can't afford to be dependent on someone else's capacity decisions. Ownership becomes competitive necessity.
El Pulso
- Anthropic has abandoned its reliance on rented hyperscaler capacity, choosing instead to own and operate data centres — a strategic rupture with the cloud-leasing model that defined the first wave of AI companies.
- A $35 billion financing deal, treating computer chips as collateral for the first time at this scale, signals that AI infrastructure is being recast as a hard-asset investment class, not merely a software story.
- The web of interlocking compute-for-cash and compute-for-equity agreements among Anthropic, Google, Amazon, SpaceX, and Nvidia raises urgent questions about whether the AI boom is built on genuinely new capital or on the same money circling a closed system.
- Power and grid interconnection have emerged as the physical ceiling that no financing structure can dissolve — analysts expect Anthropic's IPO filing to formally price electricity access as a business risk for the first time in frontier AI.
- With CoreWeave carrying $35 billion in debt after a similar infrastructure pivot, investors and regulators are watching closely to see whether Anthropic's ownership ambitions lead to sustainable capacity or to a leveraged race against the grid.
Anthropic, the artificial intelligence laboratory preparing for a landmark public offering, is moving from renting computing power to owning the infrastructure that runs it — a shift that places the company inside the same capital-intensive world once reserved for utilities and telecommunications giants. Backed by $35 billion in chip-collateralized financing from Apollo and Blackstone, and committing $50 billion toward purpose-built facilities in Texas and New York, the company is betting that control over physical infrastructure is inseparable from control over the frontier of intelligence itself. The move surfaces a deeper question about the AI era: whether the capital fueling this transformation is genuinely expansive, or whether it is the same money cycling among the same few hands.
Anthropic is no longer content to rent its way to the frontier of artificial intelligence. The company, preparing for a public offering that could value it at up to $2 trillion, has begun building and owning its own data centre infrastructure — a pivot that redraws its relationship with the cloud giants it once depended on entirely.
The turning point came in June, when Apollo and Blackstone structured a $35 billion financing vehicle using computer chips themselves as collateral — an arrangement without precedent at this scale. Rather than routing that capital through Amazon, Google, or Microsoft, Anthropic deployed its first gigawatt of capacity through Fluidstack, a UK-based neocloud provider. The choice was deliberate: a preference for control over convenience. The company has since committed $50 billion to purpose-built facilities in Texas and New York, work expected to create 800 permanent jobs and 2,400 in construction. Chief executive Dario Amodei has been clear that infrastructure capable of supporting frontier AI cannot be left to someone else's capacity schedule.
Yet the financing architecture surrounding this shift raises harder questions. SpaceX's IPO documents revealed that Anthropic will pay $1.25 billion monthly to rent capacity at xAI's Memphis data centres through 2029. Nvidia has formalized a $500 billion financing platform with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR, treating GPUs as an investable asset class. The overlapping compute-for-cash and compute-for-equity relationships among Anthropic, Google, Amazon, SpaceX, and Nvidia prompt a question regulators and investors will soon have to answer: how much of the AI infrastructure boom is financed by capital genuinely entering the system, and how much is the same money moving among the same few players?
Beyond capital, power has become the constraint that no financing structure can easily resolve. Analysts expect Anthropic's IPO prospectus to list grid interconnection as a formal risk factor — likely the first time a frontier AI company has priced electricity access into a public filing. CoreWeave, which followed a similar path from renting to owning GPU infrastructure, ended the second quarter carrying $35 billion in debt. As Anthropic deepens its commitment to ownership, the industry will watch whether the company traces the same arc — and whether the physical limits of the power grid ultimately force a reckoning with the pace of the entire AI buildout.
Anthropic is no longer content to rent computing power from the cloud giants. The artificial intelligence lab, preparing for what could be a multi-trillion-dollar public offering, has begun a deliberate pivot toward owning and operating its own data centre infrastructure—a move that reshapes not just the company's balance sheet but the entire ecosystem of AI compute financing.
Until recently, Anthropic's strategy was straightforward: lease capacity from Amazon, Google, Microsoft, and SpaceX as needed. That changed in June when Apollo and Blackstone structured a $35 billion financing vehicle backed by computer chips themselves, treating processors as collateral in a way that has never been done before at this scale. Rather than funnel this capital through Anthropic's established relationships with hyperscalers, the company chose to deploy the initial gigawatt of capacity through Fluidstack, a UK-based neocloud provider. The decision signals something deliberate: a preference for certainty and control over the convenience of existing partnerships.
The scale of Anthropic's ambition is now visible. The company has committed $50 billion toward building purpose-designed computing facilities in Texas and New York, work that will generate 800 permanent jobs and 2,400 construction positions. Dario Amodei, Anthropic's chief executive, framed this not as optional but as necessary—infrastructure capable of supporting the frontier of artificial intelligence development cannot be outsourced or delayed by someone else's capacity constraints. Layered atop this are existing commitments: a $150 billion chip supply agreement with Google, an $11 billion AWS campus in Indiana, and multi-cloud access to Google's tensor processing units and Microsoft's Azure. Yet the recent decisions point unmistakably toward ownership.
The financing architecture underlying this shift reveals how capital now circulates among a small cluster of technology and investment firms. SpaceX's IPO documents disclosed that Anthropic will pay $1.25 billion monthly to rent roughly 300 megawatts of capacity at xAI's Colossus data centres in Memphis through May 2029—a commitment that sits alongside a separate $30 billion compute agreement between SpaceX and Google. Nvidia, meanwhile, has formalized a $500 billion financing platform with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR, treating graphics processing units as an investable asset class for the first time. The web of compute-for-cash and compute-for-equity relationships among Anthropic, Google, Amazon, Microsoft, SpaceX, and Nvidia raises a question that regulators and investors will soon confront: how much of the AI infrastructure boom is actually financed by capital moving in circles among the same handful of players?
But ownership and financing are only part of the constraint. Power has emerged as the physical bottleneck that no amount of capital can easily overcome. Analysts expect Anthropic's IPO prospectus to list compute and power availability as formal risk factors—likely the first time a frontier AI company has priced grid interconnection risk into a public filing. Anthropic's Fluidstack-built facilities and its capacity commitments with Amazon and Google all depend on power and grid timelines outside the company's direct control. Nvidia's $3 billion stake in Texas developer Lancium, which is involved in the Stargate campus in Abilene, extends the company's exposure to power infrastructure. The neocloud sector offers a cautionary preview: CoreWeave's total debt reached $35 billion by the end of the second quarter, driven by a $9.4 billion capital expenditure program for specialized GPU cloud infrastructure. As Anthropic transitions from renting to owning, the industry will watch whether the company follows a similar path into asset-backed debt, and whether power constraints force a recalibration of the entire AI infrastructure buildout timeline.
Citas Notables
This expenditure is necessary for infrastructure capable of supporting continued development at the frontier of AI.— Dario Amodei, Anthropic CEO
GPUs have become an 'investable asset class' for the first time, with chips now serving as collateral.— Jensen Huang, Nvidia CEO (paraphrased)
Quieres la nota completa?
Lee el original en ET Datacenters ↗
Technology