Altman seeks to raise billions for network of AI chip factories
bloomberg.com
bloomberg.com
So something isn't very clear about this story.
What matters is FLOPS/dollar (production cost) and FLOPS/kilowatt (energy use of resulting chips).
If you can use an older tech fab to get production cost down enough, it might be worth living with bigger transistors that use a little more energy.
AI stuff is massively parallel, unlike most other tech stuff, so they don't actually need super low latency and high clock speeds.