Broadcom's AI Debt Bet Triples to $100 Billion

What Broadcom Is Asking Lenders For

Broadcom is in talks with lenders, including the private-credit firms Blackstone and Apollo Global Management, for a debt financing package that could total up to $100 billion, Bloomberg reported on 20 August 2026, citing people familiar with the discussions. The money is meant to fund AI chip infrastructure that benefits Anthropic and reportedly other AI companies, potentially including OpenAI.

The structure under discussion splits the raise into two pieces: a senior-secured tranche of roughly $60 billion to $70 billion, which gets paid first if anything goes wrong, and a junior tranche of about $30 billion, which sits behind it and carries more risk. Broadcom is expected to guarantee part of the senior tranche itself, the same backstop role it played in an earlier, smaller version of this deal.

From $35 Billion To $100 Billion In Ten Weeks

This is not the first time Broadcom, Blackstone and Apollo have done this. On 9 June 2026 the same three parties completed a $35 billion debt financing built to buy custom Google chips that Anthropic then leases through a special-purpose vehicle, part of what Broadcom chief executive Hock Tan called the AI XPV platform. That deal aimed to bring more than 1 gigawatt of computing capacity online in 2026, rising past 20 gigawatts by 2028.

Barely ten weeks later, the same relationship is being asked to nearly triple in size. The table below lines up what changed.

Why AI Compute Is Turning Into A Credit Instrument

The pattern across both deals is the same: instead of Broadcom, Google or Anthropic paying for chips and data centres out of cash or equity, a special-purpose vehicle borrows the money, buys or leases the hardware, and repays lenders from the revenue the compute eventually generates. Broadcom expects more than $100 billion in AI chip revenue next year, with Anthropic alone accounting for over 40 percent of that figure, exactly the kind of concentrated, long-duration cash flow that private-credit lenders like Blackstone and Apollo are built to underwrite.

That keeps the debt off the operating companies' own balance sheets, which is precisely why it can scale so fast: nobody has to ask a board or shareholders to approve a $100 billion loan, because the loan does not sit on their books. It also means the debt appetite for AI infrastructure is now being tested and re-tested in real time, at a pace no single quarterly earnings call would ever reveal.

The Upstream Risk Behind Every Compute Invoice

None of this shows up on an invoice. A European business renting AI compute or API access that ultimately runs on Anthropic's Google-chip or Broadcom-silicon infrastructure sees a monthly bill, a usage dashboard and an uptime SLA, not the leveraged financing structure sitting several layers upstream of that bill. If a senior or junior tranche were ever repriced, extended or restructured, the practical effect would land on capacity and pricing long before it appeared in any public disclosure a customer would see.

The useful move is to add one question to vendor due diligence alongside the uptime and security ones: how is the underlying compute capacity financed, and what happens to service continuity if that financing needs to be renegotiated. Most vendors will not have a rehearsed answer yet, which is itself useful information about how exposed a given supply chain is. Converted at roughly today's rates, the sums under discussion run to about EUR 92 billion or GBP 79 billion, scale most operators only ever meet through a vendor's pricing, never its balance sheet.

Read next: Google Hedges Its AI Chip Bet With $12 Billion  |  A Forced Seller Just Put A Real Price On Anthropic