Broadcom’s $60B AI debt push shows the chip race is becoming a credit race

The deal on the table

Broadcom’s AI story has moved from the semiconductor aisle to the credit desk. The chip designer is in talks with lenders to raise more than $60 billion in debt for an AI chip financing deal that would benefit Anthropic and other companies, according to a Reuters report citing Bloomberg News. The proposal under discussion could include about $30 billion of junior debt, while Broadcom would guarantee part of a senior-secured tranche that could range from roughly $60 billion to $70 billion. If the moving pieces are combined, the financing under discussion could reach as much as $100 billion. [1]

The structure matters as much as the headline number. Reuters reported that the new debt would be issued by a special-purpose vehicle and could resemble the $35 billion financing framework Broadcom, Apollo Global Management and Blackstone previously used to support Anthropic’s computing-capacity expansion. Blackstone and Apollo are again in talks to participate, while Blackstone declined to comment and Broadcom and Apollo did not immediately respond to Reuters requests. [1]

In plain language, the model shifts part of the AI buildout away from a simple “customer buys chips” transaction and toward a project-finance structure. Investors fund racks and systems built around Broadcom technology; AI labs or hyperscalers lease or consume the capacity; Broadcom, in some cases, helps backstop the economics so lenders are willing to write very large checks. That is why a semiconductor order now reads like a private-credit transaction.

Why Broadcom is central

Broadcom sits at a particularly valuable junction in the AI supply chain: custom accelerators, networking silicon and the software infrastructure inherited from VMware. Reuters described the company as playing a crucial role in helping firms such as Alphabet and Meta design custom chips, while also noting Broadcom chip-supply deals with Anthropic and OpenAI. [1]

The financing follows a previously disclosed Broadcom-Apollo-Blackstone platform designed to finance Anthropic capacity using Broadcom custom chips and networking products. Reuters’ fresh report says that earlier commitment was expected to add one gigawatt of computing capacity, while the wider partnership aims to enable more than 20 gigawatts of compute power for leading AI labs by 2028. [1]

That scale explains the debt appetite. AI labs do not merely need chips; they need packaged capacity: accelerators, networking, memory, power, cooling, data-center shells and long-term operating contracts. A frontier model company may want compute now but prefer not to carry the entire data-center capital burden on its own balance sheet. Broadcom’s proposition is therefore not only “we can design the silicon,” but “we can help assemble the financing stack that gets the silicon deployed.”

The credit risk investors are watching

The new talks land at a delicate moment. Broadcom shares have been under pressure as investors debate whether these AI-financing vehicles create hidden risk for the company. Investing.com reported on August 19 that Broadcom stock slid 4.2% in midday trading to $364.13, citing concerns including Google’s expanded relationship with Marvell, Broadcom’s AI financing exposure, a VMware security issue and insider selling. The same report said Bank of America estimated that a related Broadcom financing vehicle could eventually reach $370 billion in senior debt by mid-2029. [2]

That $370 billion figure should not be confused with debt sitting directly on Broadcom’s own balance sheet. TS2.tech, in an August 19 analysis, stressed that the estimate is not Broadcom revenue guidance and may not reflect direct Broadcom debt. But the figure has still become a shorthand for the market’s concern: if Broadcom’s AI growth depends on financing vehicles, investors need to understand who ultimately bears lease, residual-value and customer-payment risk. [3]

The market is trying to price two truths at once. First, financing can accelerate Broadcom’s AI revenue by removing a funding bottleneck for customers. Second, backstops and guarantees can blur the boundary between a chip supplier and a credit sponsor. The more successful the platform becomes, the larger the absolute exposure investors will scrutinize.

A more crowded custom-chip field

The debt talks also come as competition in custom AI silicon is intensifying. Reuters reported on August 19 that Marvell Technology will help develop Google custom chips and gave Google the option to buy a stake worth as much as $12.2 billion. Marvell shares jumped nearly 8%, while Broadcom, described in the report as Google’s main custom-chip partner so far, fell more than 5%. [4]

The Marvell deal does not necessarily mean Google is abandoning Broadcom. TS2.tech noted that Google’s expanded Marvell relationship does not indicate that any current Broadcom projects have been discontinued. But it does show that hyperscalers are diversifying suppliers and using equity-linked incentives to lock in custom silicon road maps. [3]

For Broadcom, the strategic message is mixed. On one hand, the company remains one of the few suppliers capable of delivering both custom accelerators and high-end networking at hyperscale. On the other, Google’s Marvell agreement signals that large AI buyers are negotiating hard, spreading risk and extracting more alignment from vendors. Financing support may become part of that negotiation, just like chip performance or delivery schedules.

Supply-chain inflation adds urgency

The capital race is not limited to Broadcom’s own balance sheet. Reuters also reported that Samsung Electronics has raised prices for some advanced contract chipmaking services by as much as 15% amid an AI-driven demand spike, and that Samsung has announced an AI chip production deal with Broadcom. [5]

That matters because custom silicon is not an abstract design file. It requires foundry capacity, advanced packaging, memory and networking components. If foundry prices rise and leading-edge capacity tightens, the total cost of deploying AI compute rises too. The financing vehicle becomes a pressure valve: when hardware inflation meets urgent model-training and inference demand, credit markets are asked to absorb more of the shock.

What to watch next

The immediate question is whether the reported package closes near the lower end, above $60 billion, or expands toward the $100 billion level described in the report. The second question is how much of any senior-secured debt Broadcom guarantees, and under what triggers. The third is whether Anthropic remains the anchor beneficiary or whether other AI labs are explicitly folded into the structure.

Investors will also watch whether rating agencies and credit analysts treat these vehicles as remote project debt or as quasi-Broadcom exposure. That distinction can affect bond spreads, equity multiples and the willingness of Apollo, Blackstone and bank lenders to scale the model.

The larger lesson is clear: the AI race is no longer only about who has the fastest accelerator. It is about who can finance the largest, most reliable blocks of compute. Broadcom is trying to turn its silicon position into an infrastructure platform. If it succeeds, it becomes more than a chip supplier. If the credit market balks, the AI buildout may discover that engineering ambition can still be constrained by debt capacity.

Sources from the last 72 hours

1. [1]Broadcom seeks more than $60 billion in latest AI debt deal, Bloomberg News reportsAug 20, 2026, 8:20 PM UTC
2. [2]Why is Broadcom stock sliding today?Aug 19, 2026, 5:54 PM UTC
3. [3]Broadcom Shares Exposed to Google Supplier Concerns as $370 Billion in AI Funding ProjectedAug 19, 2026, 1:50 PM UTC
4. [4]Marvell gives Google option to buy $12.2 billion stake in custom AI chip dealAug 19, 2026, 12:38 PM UTC
5. [5]Selon certaines sources, Samsung augmenterait les prix de ses puces électroniques jusqu'à 15 % en raison d'une forte hausse de la demandeAug 19, 2026, 8:28 AM UTC

AI-generated article based on recent web research, then preserved as a dated editorial snapshot.