Broadcom and Intel CEOs Warn That Capital, Not Chips or Power, Could Become AI's Biggest Bottleneck

SINGAPORE — The executives who sell much of the world's AI silicon have a new worry, and it is not wafer capacity. Speaking at DBS Bank's Asia Future Forward Conference on Monday, Broadcom CEO Hock Tan said access to capital "might be the biggest constraint" on the AI buildout — potentially a bigger obstacle than shortages of semiconductors, memory chips or electricity. Intel CEO Lip-Bu Tan, sharing the stage, pointed to the same pressure: "If there's any challenge in the front, it's capital and the cost of capital, and then the other part is in the longer run, people want to see the ROI."

The warning lands just as the financing machine behind AI has visibly slowed. According to Financial Times reporting citing Morgan Stanley data on public bonds and private placements, global AI-related debt issuance fell to about $23 billion in September — less than half of August's total and well below the monthly peak of $113 billion reached in June. Morgan Stanley still expects close to $570 billion of AI-related debt issuance for 2026 as a whole, while JPMorgan has projected roughly $4.1 trillion of cumulative AI debt financing through 2030. Those forecasts assume the issuance curve recovers; September suggests it has not yet.

Neither executive is calling a top on demand. Tan said demand for compute infrastructure currently exceeds what the industry can supply, and that the imbalance could persist through 2029 and potentially into 2030. He argued that today's bottlenecks have a way of resolving themselves: "Memory shortage today might not be that way a year from now." Wafer shortages could ease by 2028 or 2029 as new capacity comes online, he said, with power availability improving in parallel as new infrastructure is energized.

Intel's Tan was similarly constructive on the near term, saying a significant slowdown over the next two to three years is unlikely given how broadly AI applications are being adopted. But both men drew the same line beyond that horizon. Tan cautioned that infrastructure spending could decelerate if adoption of advanced AI models disappoints or if improvements in model capability begin to plateau, while Intel's chief said companies will increasingly need to demonstrate tangible returns through efficiency gains and new business models rather than promises.

The remarks carry extra weight because both companies are among the largest prospective borrowers in the ecosystem. Broadcom is assembling roughly $60 billion in debt financing tied to its chip arrangements with Anthropic, which banks are structuring as about $42 billion of senior-secured debt plus an $18 billion junior tranche led by Blackstone. Under the lending commitment disclosed in Anthropic's IPO prospectus, Broadcom agreed to lend Anthropic up to $42 billion, with the amount convertible into equity. Bank of America analyst Tom Curcuruto has estimated the financing vehicle could support roughly $370 billion of senior debt by mid-2029 at a 20-gigawatt scale, including about $150 billion of new issuance in 2027 — while Broadcom's first transaction already exposes it to as much as $29 billion of customer lease obligations.

Intel has been raising on the equity side. In August it upsized a common stock offering to $20 billion, selling 210,526,315 shares at $95 each to fund its contract-manufacturing buildout. The company raised its 2026 capital expenditure target to $20 billion from $18 billion in July and has committed to reaching high-volume production on its 14A process node by 2028.

The shift in framing is the story. For two years the AI buildout was narrated as a race against physical constraints — HBM supply, packaging capacity, transformers, gigawatts. What the two CEOs said this week is that the binding constraint going forward may be the price of money rather than the supply of machines, and that beyond 2028 or so the whole stack depends on AI producing revenue that justifies the debt. That points toward more deals structured like vendor financing with equity kickers and special-purpose vehicles — arrangements that keep the buildout moving while quietly shifting risk from customers to suppliers. It is exactly the pattern that has made late-cycle infrastructure booms fragile.