Intel and Broadcom chief executives warned that capital and its cost could become the largest constraint on artificial intelligence infrastructure. Their concern is not a claim that demand is weakening today. Both said demand for computing capacity still exceeds supply, while questioning how long investors can fund projects before customers prove adequate returns.
Why capital is becoming the bottleneck
AI data centers require accelerators, networking, memory, buildings and power connections. Many projects need large upfront spending years before revenue is certain. Higher bond yields raise financing costs and reduce the present value of future cash flows, especially for projects with long construction periods.
Broadcom chief executive Hock Tan said supply bottlenecks in memory, wafers and power can ease as capacity expands, while capital may remain scarce. Intel chief executive Lip-Bu Tan added that customers will eventually need to demonstrate return on investment through productivity or new business models.
The balance-sheet exposure is uneven
Broadcom reported 16.7 billion dollars of fiscal third-quarter AI semiconductor revenue, up 221% year over year. It also generated 14.2 billion dollars of operating cash flow and spent 0.5 billion dollars on capital expenditures. Broadcom can maintain relatively low capital intensity because manufacturing and much of the data-center construction sit elsewhere in the supply chain.
Intel carries a different burden because it operates semiconductor factories. Its second-quarter filing showed negative adjusted free cash flow for the first half of 2026 and continuing capital commitments, even as product revenue improved. Government incentives, partner contributions and financing terms can reduce net cash spending, but they do not eliminate execution risk.
Demand strength does not settle project economics
The executives expect compute demand to remain strong through the next several years. That outlook can coexist with a financing constraint. A project may have customers and still fail to earn an acceptable return if power, construction and borrowing costs rise faster than revenue.
Investors should watch hyperscaler capital spending, project-finance terms, data-center lease yields, utilization and evidence that AI applications generate revenue or savings. For Intel, foundry utilization and net capital expenditure are critical. For Broadcom, custom accelerator and networking growth must remain diversified across customers.
BTI's conclusion is that capital is becoming a selection mechanism rather than proof that the AI cycle is ending. Suppliers with strong free cash flow and asset-light economics can withstand tighter financing more easily than manufacturers and developers carrying large construction commitments.