Broadcom’s $120 Billion Seat at the AI Financing Table

Broadcom makes the networking chips that move data inside AI data centers. Increasingly, it also helps pay for the data centers themselves. The company has participated in financing for artificial intelligence infrastructure special purpose vehicles totaling more than $120 billion, according to people familiar with the arrangements, a scale of involvement that makes the chipmaker one of the most important financial actors in the AI buildout.

The most visible of those vehicles is the AI XPV Platform, created in June with Apollo Global Management and Blackstone. The platform’s first deal was a $35 billion special purpose vehicle that bought Google TPUs and Broadcom networking hardware and leased it to Anthropic, the AI lab that needs billions of dollars of compute but has no credit rating and cannot borrow the money itself. The structure solves a problem that no bank would touch directly: how to finance tens of billions of dollars of chips for a startup customer.

The debt was arranged in three tranches. About $6 billion of senior notes were sold to banks at Treasuries plus 100 basis points, $24 billion of notes at a 5.75% yield went to institutional investors, and $4.5 billion of junior notes priced around 8%. Broadcom’s role went beyond selling hardware. The company agreed to provide residual value support covering roughly $30 billion of the $35 billion financing, meaning it would cover any shortfall if Anthropic stopped paying and the vehicle could not sell the chips for enough to make senior investors whole. Broadcom’s exposure declines as Anthropic makes its lease payments, according to people familiar with the deal.

The arrangement mirrors the financing playbook that built the AI chip boom into a financial industry of its own. Morgan Stanley estimates that U.S. AI capital markets financing will reach $400 billion, potentially surpassing $1 trillion by 2028, to match roughly $1.8 trillion in capital expenditure needs over the next two years. Bank of America has sized the class of vehicles behind Broadcom’s AI chip business at potentially $370 billion of senior debt by 2029, including roughly $150 billion of new issuance in 2027 alone.

Broadcom’s position in this machinery is structural. Its networking silicon, the Tomahawk and Jericho families, sits in virtually every large AI cluster, and its custom ASIC business designs chips for hyperscalers, most notably the TPUs it co-designs with Google. BMO Capital Markets calls Broadcom the second-largest AI chip supplier after Nvidia. When a customer like Anthropic needs chips, networking and financing, Broadcom can supply all three, and the financing is what closes the deal.

The wave is broader than one platform. Meta completed a $27.3 billion SPV transaction around its Louisiana data center campus. Amazon raised about $10 billion in Canadian dollar bonds. Alphabet is guaranteeing data centers for Anthropic as part of the same web of deals. Special purpose vehicles have become the preferred tool because they keep the assets off the balance sheets of everyone involved: the chipmakers do not carry the hardware, the lenders do not carry the customer risk directly, and the startup customers get the compute without the debt.

That is also the source of the concern. In these structures, the chip vendor, the equity holder and the financing guarantor are, to a meaningful degree, the same cluster of firms. The pattern echoes the telecom buildout at the turn of the century, when equipment makers lent customers the money to buy their equipment, a practice that flattered demand until customers could not pay and vendors absorbed the losses. The AI version is better collateralized, with guarantees layered in to protect senior lenders, but the underlying mechanic is the same.

The scale has drawn attention from rating agencies and regulators, who are watching whether the guarantees migrate risk onto the balance sheets of companies that have so far kept it off. For Broadcom’s own shareholders, the financing business is a double-edged sword: it locks in demand for the company’s chips and networking gear, but it also ties the company’s fortunes to the creditworthiness of customers it does not control. The company has said the structures are designed so that its exposure declines as customers pay, and the five-year guarantee periods are shorter than the life of the hardware. Still, the size of the commitments, measured in tens of billions, has made Broadcom’s financing activities a topic on every AI investor’s earnings call.

The exposure is concentrated. Roughly $200 billion in contracts depends on Anthropic continuing to make its lease payments, and the largest vehicle in the market is built around a single customer. If Anthropic’s growth stalls, or if credit markets tighten before the next tranches price, the structure will be tested in ways it has not been so far. The numbers keep growing as new deals are added, from the $35 billion platform in June to reports of additional vehicles in the tens of billions.

For now, the machine works. Broadcom’s involvement gives the company a claim on the AI boom that does not depend on any single product cycle: even if individual chip generations fade, the financing relationships endure, and the customers stay locked into Broadcom’s networking and custom silicon. The company has moved beyond selling components into the business of arranging the capital that buys them. In an industry that has become as much about financial engineering as about physics, that position, at the center of a $120 billion web of vehicles, is the one that matters most.

Related Posts

  • September 6, 2026
  • 10 views
Anthropic Moves Its IPO Filing to Late September

The bankers and lawyers running Anthropic’s initial public offering had told investors to expect the company’s registration documents as soon as this week. The calendar has moved. Anthropic now plans…

  • September 6, 2026
  • 11 views
OpenAI Quietly Revises GPT-6 Astra Scores After Launch

When OpenAI released GPT-6 Astra on Sept. 3, the launch post carried the usual furniture of a modern model debut: coding results, speed comparisons and a figure for how often…