Anthropic's $35 billion chip financing shows how the AI race is moving from venture rounds into structured credit. The next frontier is not just better models, it is who can finance enough compute to keep them running.
Anthropic has found a new way to pay for the most expensive part of artificial intelligence: let Wall Street finance the chips, then lease the compute back as the business scales. Apollo Global Management and Blackstone have finalized a $35 billion private credit package tied to Google tensor processing units for Anthropic, turning AI infrastructure into one of the largest structured debt stories in the market.
This is not a normal startup financing. It is closer to the way airlines finance aircraft or logistics companies finance warehouses. The hardware sits inside a special structure, Anthropic gets access to the capacity it needs, and lenders get claims supported by assets, leases and guarantees rather than relying only on the company's equity value. That matters because frontier AI is no longer a software business in the clean, old sense. It is a capital expenditure race with model labs attached.
According to Bloomberg reporting carried by Investing.com, the finalized financing includes $6 billion of A1 notes and $24 billion of A2 notes supported by Broadcom through a residual value agreement, with the balance sitting in a more junior tranche. In plain English, Broadcom is helping reassure lenders that the chips will still have meaningful value if something goes wrong. That is the unusual part, and it is why this deal deserves more attention than another giant AI funding headline.
For the past two years, the AI market has been described mainly through equity rounds, cloud commitments and chip shortages. Anthropic's latest financing suggests that view is already incomplete. The company closed a $65 billion Series H round in May at a reported $965 billion post-money valuation, but even that kind of fundraising does not solve the infrastructure problem by itself.
Training and serving advanced models requires vast amounts of compute, and compute is not abstract. It means chips, racks, networking, cooling, power contracts and data center capacity. Equity is too expensive to use for every layer of that buildout, especially when the useful life of the equipment can be modeled and financed. Private credit has been waiting for exactly this kind of opening.
Apollo and Blackstone are not behaving like tourists here. Both firms have spent years moving deeper into infrastructure, data centers and asset-backed finance. Blackstone announced in May that it was working with Google on a new U.S.-based TPU cloud venture, while Apollo has been pushing further into large-scale private credit structures. AI gives them a new borrower class with enormous capital needs and, in the best case, long-term lease payments attached to mission-critical hardware.
That is why the Anthropic deal feels bigger than one company's balance sheet. If it works, it gives other AI labs a financing map. Instead of raising equity every time they need more chips, they can build vehicles around hardware, lease obligations and vendor support. The risk does not disappear. It moves into a different container.
Broadcom's role changes the chip conversation
Broadcom is the other important name in this story. The company helps Google develop and supply TPUs, and Anthropic has already expanded its partnership with Google and Broadcom for multiple gigawatts of next-generation compute. In April, Anthropic said that arrangement would support its growing Claude customer base and its push toward frontier model development.
Now Broadcom is doing more than selling silicon into the boom. By supporting the senior debt through a residual value arrangement, it is helping make the chips financeable. That is a very different role from the usual semiconductor supplier position. It effectively says the vendor has enough confidence in the hardware platform, demand and resale economics to help lenders get comfortable.
Broadcom chief executive Hock Tan has also told investors that the company is working with Apollo, Blackstone and other investors on an AI XPV platform expected to deploy more than 20 gigawatts of compute capacity through 2028. The first $35 billion tranche tied to Anthropic gives that ambition a concrete shape. This is not just chip demand. It is chip demand wrapped in credit engineering.
The practical implication is that custom AI chips may become easier to finance than general-purpose hardware if vendors, customers and lenders can agree on value, utilization and backstop terms. That would strengthen the position of companies with deep chip roadmaps and committed AI lab customers. It could also widen the gap between the largest model developers and everyone else, because these structures reward scale, predictable usage and trusted counterparties.
There is still a clear risk. AI infrastructure finance depends on the assumption that demand keeps rising fast enough to justify the debt. If enterprise adoption slows, pricing falls too quickly, or newer chips make older hardware less valuable, these structures will be tested. Residual value agreements help, but they do not make technology cycles stand still.
For now, the signal is straightforward. The AI buildout has moved beyond venture capital theater and into the machinery of global credit markets. Anthropic gets the compute it needs without treating every chip purchase like an equity event. Apollo and Blackstone get a new asset class. Broadcom gets deeper into the economics of the platforms it helps build. The next thing to watch is whether this becomes a one-off trophy deal or the standard financing model for frontier AI.
Also read: Anthropic turns to private credit for its $35 billion chip bill • Britain’s AI supercomputer bet still needs American chips • Google makes AI Plus cheaper as the subscription fight turns practical
