The Chip Boom Is Real. So Is the Financing Fog Around It.

If you judged the AI economy purely on this week's earnings reports, you'd conclude it has never been healthier. Seagate and Teradyne both posted the kind of beat-and-raise quarters that used to be rare even in boom times, with Seagate's data-center hard drive business growing revenue by more than half from a year ago and management extending its planning horizon on capacity commitments out to 2029. KLA and NXP Semiconductors both topped Wall Street's targets, and KLA raised its outlook for global wafer-fab equipment spending to a range in the low $150 billions for the year, up from roughly $140 billion just months ago. Amphenol, which makes the connectors and cables that wire together AI data centers, posted 30% organic growth.

And yet semiconductor stocks broadly fell even as the numbers came in strong, extending a monthlong slide that has erased tens of billions in AI-adjacent market value, with South Korea's memory-chip giants down more than 40% from June highs. That disconnect is the real story. Investors are no longer asking whether AI demand is real — the order books and lead times, some stretching into 2028, answer that question convincingly. They're asking who is actually paying for it, and increasingly, the answer looks circular.

Over the past several weeks, a pattern has become impossible to ignore: chip and infrastructure companies extending financing, equity stakes, or long-term purchase commitments to the very customers buying their products, who in turn use that capital to buy more chips. When a semiconductor supplier effectively bankrolls its own demand, revenue growth stops being a clean signal of end-market appetite and starts looking like leverage dressed up as commerce. It isn't fraud, and it isn't new — vendor financing has shown up in every capital-intensive tech buildout from telecom in the late 1990s to solar a decade ago. But it tends to flatter growth rates right up until the financing well runs dry, at which point the slowdown looks far sharper than the underlying demand ever was.

This matters well beyond Wall Street trading desks. The AI buildout has become a meaningful driver of U.S. capital expenditure, and by extension, of GDP growth, employment in construction and skilled trades tied to data-center buildout, and demand for electricity — all of which show up in the real economy Suffolk County residents live in, from Long Island Power Authority planning to local construction payrolls tied to data-center and warehouse projects. If the financing architecture underneath that capex is more fragile than it looks, a correction in AI infrastructure spending wouldn't stay confined to chip stocks. It would ripple into the broader capital-goods economy that has been a genuine bright spot this year.

There's a second, quieter tell in this week's results: gross margins. Skyworks, Qorvo, and NXP all flagged rising input costs pressuring margins even as revenue beat expectations — a sign that suppliers further down the chain are absorbing cost pressure that hasn't yet been passed through to end customers. That's the kind of friction that shows up before a cycle turns, not after.

None of this means the AI capex cycle is ending. Order backlogs, extended lead times, and raised capacity guidance from multiple companies this week all argue for continued strength well into 2027. But the gap between spectacular reported earnings and falling stock prices is the market's way of pricing a risk that doesn't show up on an income statement: that some meaningful share of today's AI revenue is being financed by the industry itself, and that when that financing tightens — whether because of higher rates, tighter credit, or simple fatigue — the reckoning could be abrupt. Voters and savers watching their 401(k)s ride this boom should treat headline earnings beats with a bit more skepticism than usual, and watch the financing structures behind them, not just the numbers on top.