According to Reuters, the prospectus of IPO of Anthropic shows that Broadcom has agreed to provide up to $42 billion in financing, which is linked to the massive computing power commitments of this AI company. This funding can cover approximately one-third of the obligation to lease tensor processing units (TPU) with a capacity of $125.2 billion over a five-year period.

This arrangement effectively makes Broadcom both a semiconductor supplier and a lender, serving one of its largest future customers.

Broadcom is currently helping to raise funds for its own AI needs.

It is expected that by 2027, Anthropic will become Broadcom's largest custom chip customer.

As the AI laboratory seeks alternatives to NVIDIA GPU, the custom accelerators and network hardware of Broadcom are becoming increasingly important. The company has anticipated that its AI semiconductor revenue will reach approximately $115 billion in 2027 and may reach $230 billion in 2028.

This outlook stems from the broader growth in AI chip revenue. Broadcom reported AI order amounts of over $30 billion in its most recent quarter report.

However, this financing structure has brought about an unusual dynamic: some of the demand that supports these forecasts may itself be driven by Broadcom.

The infrastructure expenditure of Anthropic is enormous.

As disclosed in its IPO document, Anthropic has committed to a total of approximately $518 billion for future cloud, computing, and infrastructure initiatives.

This AI company had revenue of nearly $4.6 billion in 2025, but its expenditures on computing power and infrastructure amounted to about $7.33 billion, indicating that it is significantly expanding its production capacity while the AI laboratories have not yet fully caught up in terms of revenue. The prospectus of Anthropic also shows that it incurred operating losses of over $8 billion.

The financing of Broadcom may make this expansion easier to pursue, while also securing chip demand for the coming years.

This structure also aligns with the broader trend in the AI financing sector: chip manufacturers, hyperscale cloud service providers, and private credit investors are increasingly using loans, guarantees, and leasing to fund data centers.

Risks of Recurrent Requirements

For Broadcom, this arrangement may be very effective, but it also carries risks.

If the growth of Anthropic slows down, then providing financing to a single customer who simultaneously occupies an increasingly large share of Broadcom AI's revenue may amplify risks. Reuters pointed out that the agreement involves potential conflicts regarding pricing, hardware availability, and payment obligations.

As Wall Street assesses whether GPU is capable of supporting large-scale chip financing, NVIDIA also faces similar issues.