The biggest obstacle to AI's expansion may no longer be silicon. The CEOs of Broadcom and Intel on Tuesday said that rising capital costs are becoming the primary constraint on building out AI infrastructure, shifting the industry's focus from chip shortages to financial engineering and return on investment.
From fab floors to balance sheets
For the past two years, the AI boom has been defined by scarcity of graphics processing units and the specialized memory that feeds them. Companies that could get chips built data centers; those that couldn't waited. But according to the two chief executives, the conversation inside boardrooms has changed. The question is no longer whether enough accelerators can be sourced, but whether the capital required to deploy them at scale still makes financial sense.
Broadcom and Intel both sell critical components for AI systems—networking gear, custom accelerators, and server CPUs—and their leaders' comments carry weight because they sit at the center of the supply chain. Their message: the bottleneck has migrated from the factory to the financing.
The math that's getting harder
AI infrastructure is capital-hungry. A single large-scale training cluster can cost hundreds of millions of dollars, and the depreciation schedules on that hardware are short. As interest rates have risen, the cost of borrowing to fund those builds has climbed. That combination—expensive money and fast-depreciating assets—puts pressure on the projected returns that justify the spending.
Neither CEO offered specific figures or forecasts. But their framing suggests that the next phase of AI growth won't be won by whoever has the fastest chip, but by whoever can deploy capital most efficiently. That's a different kind of competition, and it favors companies with strong balance sheets and predictable revenue streams over those burning cash for market share.
If capital becomes the limiting factor, the ripple effects could be broad. Cloud providers may stretch out construction timelines. Enterprises may prioritize smaller, more targeted AI projects over sprawling general-purpose clusters. And chipmakers could see order patterns shift as customers optimize for total cost of ownership rather than raw performance.
The CEOs didn't announce any changes to their own capital plans. But their comments align with a growing unease among investors who have poured money into AI-themed stocks and now want to see tangible returns. The hardware story is well understood. The financing story is just beginning.
The return-on-investment clock
What happens next depends on whether the revenue from AI services can scale fast enough to cover the cost of the infrastructure being built to deliver them. If it can't, the constraint won't be chips or capital—it'll be patience. Broadcom and Intel haven't set a timeline for when they expect the capital environment to ease, and neither company has said it will slow its own AI-related investments. For now, the message from the top is that the easy part—building the technology—is giving way to the hard part: paying for it.
