The AI infrastructure spending spree keeps getting bigger. Hyperscalers are committing hundreds of billions to data centres, chips and power. The question isn't whether they can afford it. It's whether the demand side ever shows up to justify it.
Look at the actual revenue. Microsoft, Google, Amazon and Meta are reporting cloud growth, but the AI-specific uplift is hard to isolate. Meanwhile, the capex guidance keeps going up. At some point, the market starts asking when the earnings per share will follow. That moment is getting closer.
The energy angle is the real constraint. You can't build a 1GW data centre overnight. Grid connections take years. Gas plants are being fired up to cover baseload. That's not a tech problem, it's a planning and regulatory one. If the buildout slows for those reasons, the capex cycle stretches, not bursts.
But the bubble talk misses something. The hyperscalers have the balance sheets to absorb a write-down. They can fund this out of cash flow for years. The real risk sits lower down the stack, with the debt-funded developers and the chip startups with no product-market fit.
When the correction comes, it won't look like 2001. It'll be a quiet repricing. The money keeps flowing, but the marginal buyer gets pickier. That's not a burst. It's a hangover.