Anthropic reportedly seeks about $60B in chip financing, offering Blackstone over 10% yield

Anthropic in talks for roughly $60 billion

Anthropic, the company behind the Claude models, is negotiating a chip financing package worth about $60 billion to lock in roughly 1.5 gigawatts of compute capacity on Google's custom TPU hardware, according to a Financial Times report relayed by dev.to. Blackstone-managed funds are in talks to subscribe around $9 billion of that package as subordinated debt carrying an all-in return above 10%. Nothing has been finalized, and the terms could change before any deal closes.

The pricing is the real signal

The yield is the most notable detail in the report. Comparable AI infrastructure financing priced at roughly 8.5% just four months earlier, meaning borrowing costs for frontier compute have climbed into territory usually associated with high-yield borrowers rather than top-tier AI labs, and they did so within the space of a single quarter.

The dev.to summary frames this as structural rather than temporary: competition among AI labs is shifting away from equity raises and headline valuations toward whoever can fund compute most cheaply, because financing cost now feeds directly into the economics of training and serving models.

Private credit becomes the AI financier

The reported structure also says something about who is paying for the buildout. With venture investors back to a discipline and public markets staying selective, private credit funds such as Blackstone are becoming the default financiers of AI capital expenditure, the analysis argues.

Structure matters as much as size here. Subordinated debt moves risk from Anthropic's equity holders to credit investors, creating a new class of AI-linked debt exposure. The package fits a broader pattern of AI labs borrowing against expected future revenue to fund data centers, and it lands at a notable moment for the company: the report also cites a potential record-setting IPO and reported third-quarter revenue of $16.8 billion.

A widening competitive spread

The analysis raises a longer-term risk. If Anthropic pays double-digit yields for compute while better-capitalized rivals, namely labs backed by hyperscalers that fund infrastructure internally at hurdle rates of roughly 6-8%, the gap could compound into a durable disadvantage over a multi-year buildout. A few percentage points of financing spread, applied across tens of billions of dollars of capacity, becomes a real cost disadvantage in a market where compute access already decides who can compete at the frontier.

The deal is also a test for private credit itself: whether funds like Blackstone can sustainably bridge the gap between AI revenue growth and the enormous upfront cost of compute, and whether this package is a one-off bridge or the first of many in a wider AI credit cycle.

Why it matters

The reported terms are a signal about AI buildout economics. Even a leading lab with billions in quarterly revenue and a hotly anticipated IPO is having to pay high-yield-level rates to secure chips, which suggests that cheap capital for AI infrastructure is no longer plentiful. If financing cost becomes a lasting differentiator, the compute race tilts toward companies that can fund it internally, chiefly the hyperscalers, and away from independent labs that must borrow. And if Anthropic's structure becomes a template, the industry's massive capital spending will increasingly sit on private credit balance sheets, tying the sector's fortunes to debt markets as much as to model quality. One caveat worth keeping in mind: the figures come from the Financial Times via a secondary summary, the deal is not final, and the numbers may shift before anything is officially announced.