Broadcom’s agreement to lend Anthropic up to $42 billion marks a significant new phase in the artificial-intelligence infrastructure race: the companies building AI systems increasingly need not only chips and data centers, but also sophisticated financing structures to pay for them.
According to Anthropic’s IPO prospectus, the financing could support roughly one-third of the AI company’s $125.2 billion commitment to lease tensor processing unit (TPU) computing capacity over five years.
The arrangement places Broadcom in an unusually powerful position because it is simultaneously involved in supplying compute, leasing equipment and financing the infrastructure Anthropic needs. The numbers illustrate the extraordinary capital intensity of frontier AI.
Register for the next Tekedia Mini-MBA.
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
Training and operating increasingly capable models requires enormous amounts of computing power, while demand for inference is also expanding as businesses integrate AI into software, customer service, research and automated workflows.
For companies such as Anthropic, access to computing capacity can therefore become as strategically important as access to capital. Broadcom’s financing helps address that problem.
Instead of Anthropic funding its entire infrastructure expansion through traditional equity or conventional borrowing, part of the spending can be supported through a financing relationship with a critical technology supplier.
The notes can potentially be converted into Anthropic shares, adding another layer to the relationship as the company prepares for a potential public offering. The arrangement also demonstrates how deeply intertwined the AI supply chain has become.
Broadcom works closely with Google on the development of TPUs, the specialized processors that Anthropic plans to use for future computing capacity. Anthropic has said that its expanded partnership with Broadcom and Google will provide access to next-generation TPU capacity beginning in 2027.
That makes Anthropic an increasingly important customer for Broadcom. Reuters reports that Anthropic is expected to become Broadcom’s largest customer in its chip-design business in 2027.
Broadcom, meanwhile, projects substantial growth in AI semiconductor revenue, reflecting how much the company expects demand for specialized AI infrastructure to expand. But the financing also highlights a central question surrounding the current AI boom.
How much of the industry’s extraordinary growth is being supported by genuine end-user demand, and how much depends on increasingly complex financial relationships between suppliers, customers and investors?
Wall Street has increasingly examined what analysts describe as circular financing. Semiconductor companies can provide funding or financial support to AI developers, which then use the capital to purchase computing infrastructure from those same technology ecosystems.
Such arrangements can accelerate deployment and create enormous revenue opportunities, but they can also increase interconnectedness if future AI revenues fail to match infrastructure commitments. Anthropic’s own filings acknowledge risks.
The company has warned that Broadcom’s dual role as supplier and financier could create potential conflicts involving pricing, hardware availability and access to computing capacity. Certain defaults could also accelerate lease obligations and restrict access to the financing facility.
For the broader technology market, the $42 billion agreement is therefore more than another AI partnership. It is evidence that the next stage of the AI race is becoming a contest of capital structure as much as computational power.
The companies capable of securing chips, energy, data centers and financing at enormous scale may determine how quickly the next generation of AI reaches the market. The question now is whether the revenue generated by that infrastructure will grow quickly enough to justify the commitments being made today.