Blackstone is quietly shopping a second giant debt package, potentially topping $36 billion, to keep Anthropic supplied with Google's custom AI chips.
Blackstone has opened early talks with investors on a fresh mega debt package, at least $36 billion by one initial proposal, to finance Anthropic's use of Google's custom TPU chips across its data centers, according to Bloomberg. The move comes roughly two months after Apollo Global Management and Blackstone arranged a $35 billion debt deal for the exact same purpose, one of the largest private credit transactions ever assembled. Details on size, structure, and whether Blackstone leads this round are still being worked out.
The mechanics mirror the first deal almost exactly. Borrowed funds buy Google's tensor processing units, then lease them back to Anthropic for use at data centers in New York, Texas, Louisiana, and Indiana. Anthropic gets the compute. The debt stays off its own balance sheet. Broadcom, which helps Google design and build the TPUs, backstops the largest slices of the transaction through a residual value support agreement, meaning if Anthropic ever defaults on its lease payments, Broadcom is on the hook to make noteholders whole.
That's the trick behind these deals. Anthropic needs staggering amounts of compute to train and run Claude, but building or buying that capacity outright would balloon its balance sheet right as it prepares to go public. Leasing it through a private credit vehicle lets the company scale its infrastructure while keeping the debt technically someone else's problem.
Two months ago would have felt like plenty for one such deal. Now Blackstone wants a second one of similar size, before the ink has even dried on comparisons to the first. That pace says something about how fast Anthropic's compute needs are growing, and how comfortable Wall Street's largest alternative asset managers have become writing enormous checks against AI infrastructure they'll never touch or use themselves. Apollo and Blackstone aren't buying chips because they want to run large language models. They're buying yield, wrapped in leases and default protections that push the actual technology risk onto Broadcom and, ultimately, onto Anthropic's future revenue.
It's a pattern that's spreading well beyond Anthropic. Private credit firms have become the default financiers of the entire AI buildout, stepping in where bank lending is too slow or too cautious for the scale AI labs now require. Frankly, the fact that a $35 billion deal from May barely counts as unusual anymore tells you how normalized this kind of financing has become in a matter of months.
The IPO clock is ticking #
The timing isn't incidental. Anthropic confidentially submitted a draft S-1 registration statement to the SEC on June 1, 2026, according to the company's own announcement, less than a week after closing a $65 billion Series H round that valued it at $965 billion. As of mid-July, no public S-1 had appeared on EDGAR, and Anthropic hasn't set a date, price range, or exchange. OpenAI followed with its own confidential IPO filing on June 8, according to TechCrunch, setting up a race between the two biggest names in generative AI to reach public markets first.
Locking in compute now, before an IPO, lets Anthropic walk into that process with its infrastructure roadmap already funded rather than dependent on offering proceeds. It also means public investors will eventually be underwriting a company whose growth already runs on tens of billions in private debt they had no say in structuring.
None of this is small money moving quietly in the background anymore. Two $36 billion-scale debt packages in three months, both tied to the same customer and the same chip supplier, is a bet that Anthropic's compute appetite keeps climbing at a pace few companies in history have ever tested credit markets to match. If Blackstone closes this second deal anywhere near its opening ask, Anthropic's leased TPU fleet will have been financed almost entirely by private credit before a single public shareholder owns a piece of the company.
Also read: JPMorgan and Accenture Back a New Group to Standardize AI Token Costs • White House Flip-Flops on AI Regulation as California Law Takes Effect • Michael Burry's Palantir Bet Is Paying Off but His Nvidia Short Isn't