The Old Boom-Bust Signal Is Back, and It's Pointing at AI

I've covered tech long enough to remember when telecom capital spending looked like a one-way bet. Every analyst had a model, every model had more fiber going into the ground, and nobody wanted to be the person who asked what happens if the money stops. Then, over about eighteen months, the whole thesis came apart, and "overbuild" became the word of choice in every sad research note I read through 2002.

So when I saw Jonathan Weil's Wall Street Journal column reviving the idea of a "funding gap" as a boom-bust signal, and pointing it at AI, I sat up a little straighter. It's not a new indicator. It's just one most people ignore until it's too late.

What a funding gap really measures

The mechanic is less mysterious than the name suggests. A funding gap shows up when a sector's capital spending and operating needs outrun the cash it generates on its own. When that happens, everything depends on outside money. Equity, debt, whatever's cheap that quarter. As long as capital keeps flowing, the machine hums. When it stops, the machine doesn't slow down. It seizes.

Weil's track record with this one is decent. In 1999 and 2000, telecom and internet infrastructure companies were pouring money into fiber and equipment faster than they could earn it back. The gap widened, capital markets blinked, and the Nasdaq fell roughly 78% from its March 2000 peak of 5,048 to its October 2002 trough near 1,114. A few years later, mortgage lenders carried the same shape of problem, funded by securitization markets that worked beautifully right up until they didn't. Housing starts ran at about 2.27 million annualized in early 2006 and collapsed to under 500,000 by 2009.

Different decades. Same setup. That's the part that nags at me.

Why AI looks different, and why that's cold comfort

Proponents of the AI trade will tell you the math is sturdier this time. And, to be fair, the top layer of the market does look different. Microsoft, Alphabet, Amazon and Meta are funding most of their data center buildout from operating cash flow, not from a shaky credit structure. That's a real cushion, and I'll grant it.

But the second and third tiers don't have that luxury. The neoclouds, the model labs, the startups renting GPUs at scale, they run on outside capital. The four biggest hyperscalers guided toward combined 2025 capital spending north of $300 billion, up from roughly $230 billion a year earlier. That number only works if the funding keeps showing up.

And that's the tell. A funding gap doesn't cause the bust. It just decides who's standing when the music stops.

What I'd actually watch

Admittedly, I'm not entirely convinced this signal is actionable in real time. It's easy to spot in hindsight, harder to call while everyone's still buying. But a few things are worth tracking. Watch the spread between hyperscaler capex and their free cash flow. Watch debt pricing for AI-linked borrowers. Watch whether the circular deals, where chipmakers invest in the customers who buy their chips, keep getting bigger. Color me skeptical, but that last one has a certain 1999 energy to it.

The question worth asking: at what point does the market decide the gap is too wide to bridge?

Nobody rang a bell in 2000 or 2007. History suggests they won't this time either. Time will tell, though.