When I say that “AI compute is rising exponentially”, I get these weird looks. It’s easy to dismiss that.

But think about it!

• Global AI compute is doubling roughly every 7 months. How to break that down?

• That sounds abstract, so let’s ground it:- Doubling every 7 months ≈ 3.3× per year

• Sustained over 5 years, that’s ~400× more compute

That what makes the AI Infrastructure buildout feel so urgent. That’s why Neo Clouds are rushing to pivot from Bitcoin mining to HPC. Neo Cloud are popping up all over the place.

The pie is getting bigger, the energy needed is getting much higher, fast! The AI bubble can’t pop easily so long as this is the case.

TSMC literally can’t make AI chips fast enough for Nvidia any longer. The demand for TSMC's advanced chip manufacturing and packaging services for AI applications is currently three times higher than its supply capacity. Think about what that means for a second.