Broadcom's $80 Billion Gamble: The Biggest AI Debt Deal Ever Is Being Built for Anthropic's Chips

On Friday, August 21, CNBC’s David Faber confirmed what Bloomberg had first reported: Broadcom is in talks with a group of lenders to raise somewhere between $70 billion and $80 billion in debt for an AI chip financing deal — and the structure, executed through a Broadcom-backed special-purpose vehicle, would channel the capital into custom silicon and computing capacity that primarily benefits Anthropic, the maker of Claude. If the junior tranches are fully placed, the total package could approach $100 billion, making it the largest single financing ever assembled in the history of the AI industry — and one of the largest private debt deals of any kind.

What we know about the deal

The numbers are still moving, but the shape of the transaction has come into focus. According to Quartz and CNBC reporting, lenders are targeting a senior tranche of roughly $45 billion and a junior tranche of about $35 billion, though both figures could shift before anything is finalized. The vehicle raising the debt is backed by Broadcom, and the proceeds are earmarked for AI infrastructure — specifically the custom accelerators and TPU-based computing capacity that Broadcom designs and deploys with Google for its largest customers.

This is not Broadcom’s first trip to the private credit well for Anthropic. In June 2026, Apollo and Blackstone finalized a $35 billion private credit package to fund roughly one gigawatt of Broadcom’s TPU capacity dedicated to Anthropic’s workloads. That debt was structured into a $6 billion senior A1 tranche, a $24-25 billion A2 tranche, and $4.5 billion of Class B notes, with Broadcom backstopping only the senior layers — a detail that mattered enormously to the lenders’ risk models. By August, Blackstone was already pitching a follow-on mega package of at least $36 billion. The new $70-80 billion negotiation is the direct descendant of those earlier deals, scaled up to a size that few thought possible even a year ago.

The Anthropic connection

To understand why Broadcom is borrowing tens of billions on Anthropic’s behalf, you have to look at the compute chain Anthropic has assembled. In October 2025, Anthropic committed to deploying up to one million of Google’s Tensor Processing Units in a deal worth as much as 10 gigawatts of capacity — a staggering figure that reportedly made Broadcom, which co-designs and supplies Google’s TPUs, one of the biggest beneficiaries. In December 2025, Broadcom confirmed that its mystery $10 billion custom-chip customer was Anthropic. Then in April 2026, Anthropic, Google, and Broadcom signed an expanded agreement for multiple additional gigawatts of next-generation TPU capacity starting in 2027, with about 3.5 gigawatts of TPU-based computing accessible to Anthropic from Broadcom alone.

Anthropic, in other words, has promised to consume amounts of compute that no AI lab has ever consumed, and it does not have the balance sheet of a hyperscaler to prepay for it. The debt structure solves this: Broadcom’s creditworthiness and chip margins sit behind the vehicle, private credit funds supply the cash, and Anthropic gets its gigawatts without issuing the debt itself. It is vendor financing reborn at civilizational scale — closer to how Boeing finances aircraft purchases than how startups buy cloud GPUs.

Why the debt markets are nervous

The timing of the deal is what makes it genuinely dramatic rather than merely large. Goldman Sachs estimated in early August that nearly $500 billion of AI-related debt has already been issued, and BNP Paribas data shows hyperscaler debt issuance hitting $220 billion in 2026 alone. The Bank for International Settlements warned in its annual report that the five largest hyperscalers are on track to spend over a trillion dollars on AI capex across 2025-2026, much of it increasingly financed by credit rather than cash flow. Investors have begun pushing back: Bloomberg reported that after roughly $300 billion of AI debt flooded markets, demand for the riskier tranches started thinning, and analysts at PIMCO and Chicago Booth have publicly asked whether AI credit is becoming the next systemic risk.

Against that backdrop, a $70-80 billion ask is a genuine stress test. If the senior tranches clear the market smoothly, it signals that private credit still has unlimited appetite for AI infrastructure backed by long-dated chip supply contracts. If the junior tranches struggle — or Broadcom has to backstop more than it wants to — it would be the strongest signal yet that the AI financing cycle is approaching its limits. Wall Street is watching the tranche-by-tranche placement the way it once watched subprime mortgage bond auctions.

What it means

For Anthropic, the deal is existential insurance. Its rivalry with OpenAI and Google is increasingly a compute war, and locking in multi-gigawatt capacity through 2027-2028 — financed off its own balance sheet — lets it keep training frontier models without an IPO or a mega-dilutive raise. For Broadcom, every dollar of debt it helps arrange is a dollar of guaranteed demand for its custom silicon business, which analysts view as its structural advantage over merchant chip suppliers. For the broader market, the deal’s fate will answer the question that has hovered over the AI boom all year: whether the credit system can keep up with the industry’s appetite, or whether the financing itself becomes the constraint. Either way, when the history of the AI buildout is written, a $100 billion bond deal for one lab’s chips will be a hard paragraph to skip.