AI Chip Boom's Fault Line: Supply Optimism Meets Financing Reality

The Divergence No One Is Talking About

Two CEOs told opposite stories about the AI chip boom last week. Christophe Fouquet, boss of ASML—the only supplier of extreme ultraviolet lithography machines—says demand for AI infrastructure will stretch beyond 2028. Hock Tan, chief of Broadcom, warned that financing may become the industry’s biggest obstacle.

That tension is the sharpest fault line in AI capex right now. English-language coverage has flattened it into one narrative: everyone agrees AI is eating the semiconductor industry, so growth will continue. But the real story is more complicated. The equipment makers are seeing orders booked through 2028. The chip designers are seeing borrowing costs threaten to choke off demand before those orders can be filled.

Who Wins, Who Loses

ASML wins. Its backlog is already full. Most new orders for advanced EUV tools are scheduled for delivery in 2028. The company plans to boost annual production capacity by 30% in 2027, with another 30% expansion under review for 2028. That’s not a company looking for demand—it’s a company waiting for supply to catch up.

Broadcom loses on margin. It’s not just building chips for the AI boom; it’s raising billions to help customers buy them. Talks are underway for more than $50 billion in financing so OpenAI can purchase its custom AI chips. A proposed $60 billion package for Anthropic includes $42 billion in debt backed by Broadcom. Tan isn’t complaining about weak demand—he’s saying the demand is so expensive to finance that it may stall.

Chip fabricators like TSMC and Samsung sit in the middle. Their capex depends on both sides: ASML’s equipment arriving on time, and hyperscalers finding money to pay for the silicon. If financing tightens, their utilization rates could fall even as orders stay booked.

The Numbers Behind the Warning

Tan’s concern is precise. He sees wafer shortages easing by 2028 or 2029, but demand for computing power exceeding supply through 2029 or 2030. That means the bottleneck shifts from physical output to financial capacity. Companies want chips they can’t get, but they also can’t pay for chips they can get.

The $50 billion and $60 billion figures illustrate the scale. Those are not startup budgets. They are sovereign-debt-level raises for private companies. Broadcom is effectively acting as a lender-of-last-resort for AI infrastructure. That’s a risky position if interest rates stay high or if project revenues fall short.

ASML’s timeline is equally concrete. Fouquet expects AI infrastructure construction to last two to five years. That’s why his company is scaling EUV production by 30% in 2027 and 2028. JPMorgan noted in September that most new ASML orders are booked for 2028 delivery. The machine tool maker isn’t guessing—it’s responding to purchase orders from TSMC, Intel, and Samsung that already have payment terms locked in.

The Hidden Risk: Whose Balance Sheet Holds?

The financing model Broadcom is pursuing shifts risk from chip designers to lenders. OpenAI and Anthropic don’t have the cash flow to buy $50 billion in custom silicon outright. They need debt. Broadcom is helping arrange it, but the debt sits on borrower balance sheets, not Broadcom’s.

That matters because if AI revenue projections fail to materialize, those balance sheets weaken. A distressed borrower defaults. A default ripples through lenders, then through Broadcom’s financing arms, then through the chip design ecosystem. ASML’s orders, by contrast, are paid by foundries with cash flows from existing businesses. Less direct exposure to AI revenue uncertainty.

The consequence: ASML is insulated. Broadcom is exposed. The market hasn’t priced that asymmetry yet.

What Happens Next

If financing tightens, we’ll see delays in cloud capacity expansions. Hyperscalers will push back on delivery timelines. Foundries will cut capital spending. ASML’s order book may stay full, but fulfillment could slip. Broadcom’s financing deals may stall or require restructuring.

If financing stays loose, demand continues to outstrip supply through 2030. Wafer prices climb. Fabrication capacity remains constrained. ASML keeps building machines. Broadcom keeps brokering debt.

The middle scenario—moderate tightening—creates a gap between booked orders and actual shipments. That’s where margin pressure hits foundries and chip designers. ASML still ships, but its customers’ ability to profit from the chips slows.

Why This Divergence Matters Beyond Semiconductors

The ASML–Broadcom split mirrors a larger pattern in AI investing: hardware and software are decoupling financially. Equipment makers are locked into long-term contracts with industrial clients. Chip designers are financing speculative revenue streams. The two models carry different risks, but both are labeled “AI capex” in market reports.

That labeling masks the divergence. An investment thesis built on “AI will drive semiconductor growth” ignores the financing risk Tan is highlighting. It assumes demand translates directly into revenue. It doesn’t account for the cost of capital, debt servicing, and the possibility that AI projects underdeliver on promised returns.

Investors watching this space should track two metrics: ASML’s order-to-ship ratio, and the spread on Broadcom’s financing deals. If ASML’s backlog grows while Broadcom’s debt terms deteriorate, the fault line is widening.

The Bottom Line

ASML sees a boom that outlasts 2028. Broadcom sees a boom that may run out of money before it hits 2030. Both are right in their own corner of the supply chain.

The real question isn’t whether AI demand is strong. It’s who pays for it, and whether the payers can keep paying. Until that answer emerges, the chip complex will remain split between equipment optimism and financing anxiety—and the gap between them will determine who wins the next phase of the AI buildout.