Broadcom chief executive Hock Tan and Intel chief executive Lip-Bu Tan told a Singapore conference that the AI infrastructure boom’s most binding constraint is no longer silicon or power but the cost of capital, with investment-grade AI bond spreads widening sharply and OpenAI’s $50 billion annualized revenue run rate still failing to demonstrate a profitable model.
The casino analogy
Hock Tan compared many AI investors to casino gamblers who “set very extreme expectations only to find out that it’s almost like they are in a casino.” He said enterprises buying tokens today are like children given $10 at a video arcade: they spend it in ten minutes and learn nothing. The blunt assessment, delivered at the DBS Asia Future Forward Conference, frames the current spending wave as experimentation rather than investment.
OpenAI revenue versus losses
OpenAI reached an annualized $50 billion in revenue at the end of September, but Tan said the figure “doesn’t tell anything at this point except that there is a lot of potential revenue growth.” The company’s massive operating expenses and capital expenditures on model training and infrastructure are producing large losses, leaving the sustainability of its business model unproven. Tan argued model developers need customers who can generate returns from deploying AI, not users burning cash on tokens.
Bottlenecks shifting to capital
Tan said the infrastructure buildout could run through 2029 and into 2030, but that timeline depends on “things we are learning as we go along.” Memory shortages may ease within a year, wafer constraints by 2028 or 2029, and power availability as new capacity comes online. Intel’s Lip-Bu Tan agreed the front-end challenge is “capital and the cost of capital,” adding that over the longer run “people want to see the ROI.”
Bond market signals
Treasury yields have risen sharply this year and spreads on AI-linked investment-grade debt have widened substantially. SpaceX’s 10-year senior unsecured notes, issued in June with a 5.875 percent coupon, traded on Friday at a 7.20 percent yield. The move sets a template for future AI infrastructure financing and raises the hurdle rate for every data center and chip project in the pipeline. With compute demand still exceeding supply, the question is whether businesses adopt AI fast enough to service debt at these levels.