The numbers keep going up. Cloud providers are spending more on data centres and chips than they earn from the services those assets support. Revenue from AI inference and training is growing fast, but not as fast as the capex curve. The gap widens every quarter.
GPU lead times have collapsed. A year ago you waited 12 months for an H100 cluster. Now you can get next generation hardware in weeks. That is the market clearing. Supply is catching up. Oversupply is the logical next step. The signs are already there in falling spot prices for cloud compute and the quiet push from vendors to shift more capacity.
This is not a bubble that will pop overnight. It is a slow correction that plays out over quarters. Companies that overbought will take writedowns. Projects that assumed infinite demand will be shelved. The capital that was poured into new fabs and server farms will take years to earn a return. Some of it never will.
The thesis that AI demand follows a J curve is being tested. If the applications don't materialise at scale, the infrastructure built today becomes a stranded asset. That risk is real. It is not priced in. The spending continues because nobody wants to be the one who stops first. But the math is simple. Supply grows faster than demand. Something has to give.
Paul