Will AI Chips Get Faster and Cheaper?

George Kurtz guarantees that chips will get faster and cheaper, though he does not know the exact trajectory. Open weight models, certainly the U.S. models, still have a lot of room to grow, and big advances are coming in shrinking them so they run on a phone.

Chips used to sit out of fashion. They are in fashion now, and more of them will be needed, along with faster chips and more specialized ones. Breakthroughs are coming in the chip world. He treats Moore's law as a bit dated, yet new chips have already made computing faster and cheaper. Specialized chips aimed at particular areas, plus new science around the mathematics, can create step function increases in performance and use less power. A football field-sized room of computers once launched the first rocket. An iPhone today holds more power than the entire room. The same shift is coming for AI: make it faster, cheaper, and more efficient, then bring it to the desktop and the phone, at the edge.

The setup will probably be a hybrid model, and a lot of the work can already be done locally. TPUs and NPUs already sit on systems people have today, and a lot of the time they go to waste.