Broadcom is in talks to raise more than $60 billion in debt — potentially as much as $100 billion — through a special-purpose financing vehicle built specifically to fund AI chips for Anthropic. It’s one of the largest AI infrastructure financing deals reported to date, and the way it’s structured says something real about how the entire industry is now funding its buildout. Here’s what’s actually happening, and why the structure matters as much as the number.

The real deal structure

Broadcom is negotiating with a group of lenders — reportedly including Blackstone and Apollo Global Management — to raise the financing through a special-purpose vehicle, structured with a junior tranche of roughly $30 billion paired with a senior-secured tranche of $60-70 billion, potentially bringing the total as high as $100 billion. Blackstone and Apollo are building on a financing partnership with Broadcom first formed in June 2026, when their initial deal raised $35 billion specifically to expand Anthropic’s computing capacity using Broadcom’s custom chips and networking equipment.

Why Anthropic isn’t just buying the chips itself

The real, structural detail that matters here: under this arrangement, Anthropic doesn’t purchase the AI chips directly. Investors finance the purchase, then lease the hardware to Anthropic — a real, meaningful shift in how AI compute gets funded. Rather than an AI lab spending its own capital or raising equity to buy hardware outright, financial firms are stepping in as the capital source, then earning a return by leasing the resulting infrastructure back to the lab that needs it.

Why this matters more than the headline number

Any single financing deal in the $60-100 billion range would have been the single biggest tech story of the year in almost any prior year — that scale, on its own, is worth noticing. But the real structural story is what this lease-based model does to who bears the risk. Under a direct-purchase model, an AI lab’s own balance sheet absorbs the risk if compute demand doesn’t materialize as expected. Under a lease-based special-purpose-vehicle model, that risk gets spread across the financial firms providing the capital — Blackstone and Apollo, in this case — rather than sitting entirely on Anthropic’s books. That’s a genuinely different risk allocation than how most of the tech industry financed infrastructure buildouts in prior cycles, and it’s worth watching whether this becomes the default model industry-wide as the capital requirements for AI compute keep climbing.

The broader pattern this fits into

This deal doesn’t exist in isolation. Nvidia has separately confirmed backing up to $105 billion in financing for a new OpenAI data center campus in Ohio, with OpenAI committing to exclusively use Nvidia GPUs at that site. Between the Nvidia-OpenAI Ohio deal and the Broadcom-Blackstone-Apollo-Anthropic financing, tens of billions of dollars in AI infrastructure financing are being arranged in parallel, each with its own capital structure, but all pointing at the same underlying reality: the compute AI labs say they need has outgrown what any single company can fund from its own revenue or equity raises alone.

What to actually watch next

A few real, concrete questions this deal raises that are worth tracking as it develops: whether the lease-based, investor-financed model spreads to other labs beyond Anthropic and OpenAI; whether the debt markets continue absorbing financing at this scale without pricing in more risk premium than they currently appear to be; and what happens to a special-purpose vehicle’s investors if AI compute demand growth slows before the leased hardware’s useful life is up — a real, underexamined tail risk in a financing structure this new and this large.

The honest takeaway

This isn’t just “another big AI funding number” — it’s a real, structural shift in how the industry is choosing to fund its infrastructure buildout, spreading capital risk across financial firms via chip-leasing arrangements rather than concentrating it on AI labs’ own balance sheets. Whether that turns out to be a genuinely smart way to fund a still-uncertain compute demand curve, or a new source of systemic risk if that demand curve doesn’t hold, is exactly the kind of question worth revisiting as more of these deals get finalized.

For the full, regularly updated picture of what else is happening in AI right now, see our daily AI news briefing.

More AI deep dives