Reuters reports “Artemis” will complement the hundreds of thousands of Nvidia H100 chips Meta bought. Similar to the MTIA chip Meta announced last year, Artemis is also focused on inference — the actual decision-making part of AI — and not training AI models, but it’s already a little late to the game.

Google introduced a second-generation TPU in 2017 that could do both, and so can Microsoft’s recently-announced Maia 100. And AMD claims its M1300X chip performs better than H100s on the inference side.