Justin Sullivan/Getty Images
Apollo Global Management and Blackstone have anchored a $35 billion loan secured against the chips that train and run AI, the first of what is likely to be a series of financings secured against hardware produced by Broadcom.
The first batch of chips will go toward fulfilling an already-announced plan by Anthropic to add more than a gigawatt of computing capacity at sites run by data center operator Fluidstack, according to a statement.
Broadcom says the financing platform, named AI XPV, could produce more than 20 gigawatts of computing power through 2028 for builders of large language models such as Anthropic and OpenAI. This is equivalent to the power generated by 20 new nuclear power plants.
The initial $35 billion will be syndicated to institutional investors, including insurance companies and banks, with Apollo’s insurance arm, Athene, accounting for a sizable portion, according to a person familiar with the matter.
The senior debt tranches of the financing have been rated investment grade, the person added.
The deal treats computing power as an asset to lend against.
A special-purpose vehicle raises the debt, buys the AI chips—including Google’s tensor processing units, which Broadcom co-designs—and leases them to Anthropic. The lease payments then repay the loan.
Holding the hardware in the SPV also keeps it off Anthropic’s balance sheet—useful for a company preparing to list on a market where companies with high debt loads are often punished by investors.
For five years, firms such as Blackstone bought data centers; the latest ventures mark the pivot to financing the hardware inside them.
In May, Blackstone put $5 billion in equity into a joint venture with Google to sell TPU capacity. In March, it anchored CoreWeave‘s $8.5 billion loan, secured against Nvidia chips—plowing capital into three chip manufacturing ecosystems in three months.
“The demand for compute has created an unprecedented opportunity to invest at scale across the AI infrastructure ecosystem, including providing financing through our credit and insurance business,” said Blackstone president and COO Jon Gray in a statement.
Morgan Stanley estimates the AI buildout will need about $1.5 trillion in outside financing through 2028, most of it private credit.
The collateral is the catch. Lenders know how to value land and power lines; chips are harder to put a number on and can lose value in months. Google has rolled out four generations of its TPUs in about three years, and Amazon, in regulatory filings in early 2025, cut the assumed life of a subset of its servers to five years from six, citing the increased pace of technology development, particularly in AI.
The use of off-balance-sheet financing has also set alarm bells ringing among regulators. The Bank for International Settlements, which counts more than $40 billion lent to AI companies last year against about $3 billion in 2010, warns that such structures can hide leverage that does not vanish when sitting outside public markets.
Moody’s Ratings estimates the five largest US cloud companies hold about $662 billion of data-center lease commitments off their balance sheets—a sum its analysts peg at 113% of the hyperscalers’ most recent adjusted debt.
Learn about our editorial standards.