In the run-up to 2008, a feedback loop formed: rising home prices justified more mortgage lending, which pushed prices higher, which justified even more lending. The loop didn't require anyone to lie — it just needed each participant to rationally respond to the step before it.
The AI version of the loop
A structurally similar loop is worth watching in AI infrastructure:
Capital markets
→ AI CapEx growth
→ AI revenue growth
→ Higher valuations
→ More capital raised
→ More AI spendingThe loop isn't inherently a problem — durable technologies do attract more capital as they prove themselves. The question that separates a healthy cycle from a circular one is:
Are revenues driven by real end-user demand, or by continuous capital expansion itself?
What to actually watch
A few concrete tells:
- Vendor financing — is a chipmaker financing its own customers' purchases of its chips, in a way that shows up as both revenue and an asset?
- Related-party revenue — is a meaningful share of "AI revenue" coming from other AI companies buying services from each other, funded by the same pool of venture and debt capital?
- Revenue concentration — is growth broad-based across many independent enterprise buyers, or concentrated in a handful of well-funded AI labs spending investor capital?
None of these, on their own, prove a bubble. But together, they're the same pattern the Enron and 2008 cycles shared: complexity that makes it hard to tell whether growth is organic — right up until financing conditions tighten and the loop runs in reverse.