MTIA 300

Meta just shipped their third-generation custom silicon for GenAI inference, which is basically them going all-in on the vertical integration play — if you’re running inference at the scale they are for billions of users, depending on Nvidia chips for everything gets expensive fast. This is their answer: build the silicon, own the stack, cut out the middleman. The signal here is clear — Meta is betting big on controlling their own hardware destiny, and the third gen means they’re past the experimental phase and into actual production deployment.