[aurareturn]

  It seems that using hardware acceleration would decrease inference costs and they could make it up on unit economics over time.

Nvidia GPUs are accelerators too. The reason they can do this so fast is because they're storing entire models in SRAM.

[rpdillon]

There are degrees of acceleration. My understanding, limited as it is, is that groq and cerebras are using highly optimized acceleration to achieve their token generation rates, far beyond that in a regular GPU, and this leads to lower costs per token.

Is this incorrect?

[aurareturn]

Yes, they're called ASICs on Grog. But Cerebras has more general cores that can do more complex things. Inference is mostly limited by bandwidth though.