Broadcom is in discussions to arrange a massive debt package to finance AI chips and infrastructure for Anthropic and other AI companies, an initiative that demonstrates how tech vendors are moving beyond sales and are now helping finance deals with customers.
The proposed structure could include roughly $30 billion of junior debt and $60 billion to $70 billion of senior secured debt. Broadcom would guarantee part of the senior portion, while the borrowing would be issued through a special-purpose vehicle rather than directly by Broadcom. The negotiations remain underway, so the size and structure could change, but reports indicate the figure could reach $100 billion.
For the enterprise AI market, the sheer size of the potential deal is significant. AI infrastructure is moving beyond the traditional model in which tech companies simply purchase servers and chips. Hardware vendors, investment firms and AI developers are creating financing structures capable of supporting data centers that require tens of billions of dollars in capital. For example, NVIDIA’s recent partnership with Wall Street firms to channel more than $500 billion into building AI infrastructure.
“Broadcom needs to guarantee its own customers’ compute bills to keep pace with NVIDIA,” Brendan Burke, Research Director for The Futurum Group, told Techstrong Semi.
“Custom accelerators may perform well in single-customer deployments but lack the multi-generation, multi-customer track record and the liquid resale market that makes GPU collateral bankable. Broadcom has to absorb the risk lenders will not.”
Expanding Anthropic’s Compute Capacity
Apollo Global Management and Blackstone are discussing participation in the new Broadcom financing. The three companies already work together through the AI XPV Platform, which was launched in June to finance computing infrastructure.
That platform’s first transaction provided $35 billion to expand Anthropic’s compute capacity using Broadcom’s custom AI chips and networking technology. Investors financed the hardware, which is then leased to Anthropic. Broadcom supported most of the debt in that transaction, helping the senior debt earn investment-grade ratings and lowering borrowing costs.
The initial $35 billion commitment is expected to provide about one gigawatt of computing capacity. The larger plan is far more ambitious: the partnership is targeting more than 20 gigawatts of capacity for leading AI labs by 2028. Reaching that level could require hundreds of billions of dollars.
The strategy, of course, is based on creating a direct link between financing and Broadcom’s semiconductor business. Broadcom supplies custom AI silicon and networking technology, and has relationships with major AI players like Anthropic and OpenAI. It also develops custom chips for companies such as Alphabet and Meta.
Broadcom expects AI chip revenue to surpass $100 billion next year. If that forecast proves accurate, AI would become an even more important driver of the company’s business as hyperscalers and model developers search for alternatives to relying exclusively on NVIDIA GPUs.
Broadcom sits in a particularly useful position because it can participate in both the technology and financing sides of that expansion. Its guarantees can help investors finance infrastructure built with Broadcom hardware, while AI companies gain access to compute without funding the entire purchase themselves.
Bottom line, this flow of capital offers another indication that the AI infrastructure expansion has moved far past advancing models, and now relies on multi-year planning for networking equipment, data center capacity and power on an extraordinary scale.
