Broadcom's $60B AI Debt Deal for Anthropic Chips

The bill for the AI buildout is increasingly being paid with borrowed money — and the sums are getting larger. On August 20, 2026, Bloomberg reported that Broadcom is in talks with a group of lenders to raise more than $60 billion in debt to finance custom AI chips and computing infrastructure for Anthropic and other customers. People familiar with the discussions said the total financing package could ultimately approach $100 billion, making it one of the largest single fundraising efforts yet assembled for the AI infrastructure boom.

Broadcom shares climbed on the report, which lands as the semiconductor company positions itself as the primary alternative to Nvidia for the custom accelerators — application-specific chips built to a single customer’s design — that the largest AI labs increasingly want to own rather than rent.

What Broadcom is raising, and how

According to the reports, the financing is being structured through a special-purpose vehicle (SPV) — a separate legal entity created to hold the debt and the assets it funds, keeping the obligations off Broadcom’s own balance sheet. That structure is standard for large infrastructure projects, and it has become the default plumbing of the AI compute era.

The package under discussion is split across tranches. People familiar with the talks described a roughly $30 billion junior debt tranche, sitting alongside a larger senior-secured tranche in the range of $60 billion to $70 billion that Broadcom would partly guarantee. Layered together, the numbers being floated push the total toward $100 billion. The debt would be repaid over time from the revenue the chips generate once they are deployed in data centers, rather than drawn from the AI labs’ cash on hand.

The mechanics matter because they are what make the deal possible. Borrowing arrangements of this kind let AI firms lock in chip supply without carrying the full cost on their own balance sheets — a crucial workaround for companies whose revenue is growing fast but whose spending ambitions are growing faster. For Broadcom, guaranteeing a slice of the senior debt is a way to make its custom silicon financeable at scale, turning a chip order into a long-lived, funded asset.

The Anthropic connection

The new debt talks build directly on a partnership announced earlier this year. In June 2026, Broadcom teamed with private-capital giants Apollo and Blackstone to finance a $35 billion expansion of Anthropic’s computing capacity using Broadcom’s custom chips and networking gear. That initial commitment was expected to add roughly one gigawatt of computing capacity, with the broader partnership aiming to enable more than 20 gigawatts of compute for leading AI labs by 2028.

Anthropic has been assembling compute from every direction. We have tracked the company’s $10 billion compute deal with Meta, its Volta compute arrangement, the Macquarie and GIC financing behind its Theseus data centers, and its strategic memory supply deal with Micron. It has also moved to control its own silicon roadmap, working with Samsung on a custom AI chip. The Broadcom debt package is the financing layer beneath that hardware strategy — the money that turns signed capacity commitments into installed racks.

The timing is notable. Anthropic is widely reported to be preparing for a public listing; readers can revisit the details in our coverage of the Anthropic IPO filing and the company’s reported $65 billion run rate. A lab heading toward the public markets has strong incentives to fund its compute through off-balance-sheet vehicles rather than raw equity or on-book debt, keeping its own financial statements cleaner as investors scrutinize them.

Why custom chips, and why now

Broadcom’s rise as an AI kingmaker rests on a simple thesis: the biggest buyers of AI compute no longer want to depend solely on merchant GPUs. Designing a custom accelerator — an application-specific integrated circuit, or ASIC — lets a lab tune the silicon to its own models and, crucially, escape the pricing power of a single dominant supplier. Broadcom provides the design expertise, intellectual property, and advanced networking to stitch thousands of those chips together; the customer provides the workload and, increasingly, the balance-sheet appetite.

That is a different model from Nvidia’s, but the two are converging on the same problem: the AI buildout now costs more than its customers can pay for out of pocket, so the supply chain is being financed. We covered Nvidia’s parallel move — its $500 billion Wall Street financing platform with six of the world’s largest asset managers, and its $250 billion data-center backstop tied to OpenAI. Broadcom’s SPV structure is a variation on the same theme, with the chip vendor guaranteeing debt to keep demand flowing.

The scale of these commitments has raised persistent questions about circularity and leverage. A growing share of AI infrastructure is now funded through off-balance-sheet obligations — vehicles, leases, and supply commitments that do not always appear as debt on a company’s headline financials but represent real, fixed liabilities. Broadcom’s package would add tens of billions more to that pile.

Market backdrop

The debt talks arrive during a stretch of extraordinary demand for AI chips. Global semiconductor sales reached a record $120.6 billion in May 2026, up sharply year over year and marking the fifteenth consecutive monthly record, and second-quarter industry sales topped $400 billion. Broadcom’s own guidance has called for AI semiconductor revenue to grow more than 200% year over year to roughly $16 billion in the current quarter.

Yet the same week the Broadcom news broke, chip stocks slipped — not on chip news, but on macro pressure, as the 30-year Treasury yield climbed to a 19-year high on fiscal-deficit and inflation worries. That tension sits at the heart of the debt-financed AI buildout: the entire model assumes capital stays cheap enough that borrowing against future chip revenue makes sense. Rising long-term rates make every one of these multi-year financing structures more expensive to service.

What it means

Broadcom’s push to raise more than $60 billion — potentially close to $100 billion — is not just a big number. It is confirmation that the AI compute race has fully crossed into a credit story, where the constraint is no longer chip design or fab capacity but who is willing to finance the hardware and on what terms.

Who wins if it works: Broadcom, which cements its position as the go-to custom-silicon partner and the financeable alternative to Nvidia; Anthropic and the other labs, which secure guaranteed chip supply without draining cash or bloating their balance sheets ahead of public listings; and the private-credit firms — Apollo, Blackstone, and the lenders now circling the new tranches — which gain early positions in what is becoming a vast new asset class. The economics of the AI data center get easier to sustain when the capital stack is deep and the chip vendor backstops part of the risk.

Who is exposed if it doesn’t: the same lenders, and ultimately Broadcom through its guarantees. The structural risk is duration mismatch — financing rapidly depreciating accelerators on multi-year terms. AI chips are on a punishing upgrade cycle, and if a newer generation renders the funded hardware uncompetitive before the debt is repaid, someone absorbs the gap. The SPV structure spreads that risk but does not erase it; it moves off the headline balance sheet, not out of the system.

What to watch next: whether the full package actually closes at the reported size, and at what spread over a rising rate environment; how much of the senior tranche Broadcom ends up guaranteeing; and whether other labs — OpenAI among them — line up behind Anthropic for similar Broadcom-backed vehicles. The broader AI capex boom has been financed on the assumption that demand is structural rather than cyclical. Deals of this magnitude are a bet that the assumption holds.