This is starting to sound like startup scope creep. Instead of making the AI model it’s now custom silicon, web browsers, and consumer electronics?
Definitely has that smell... At the same time though, they NEED inference cost to drop substantially, and even better for them if it only happens for their models on their hardware.
I assume they're doing everything they can to make that happen model-side, but coming at it from the other end makes sense too if they can swing it.
Nearly all those initiatives have failed though
Maybe, but they’re also a massive company. At some point Google stopped being a startup and become a massive company with margins to look after