00:00Ed this just seems like the move that a lot of these hyperscalers are making. Look for homegrown chips rely
00:06less on the NVIDIA's of the world. Yeah remember that meta is still one of the biggest buyers of both
00:11NVIDIA and AMD chips but there is some merit in having custom silicon. So they basically have four generations of
00:17new accelerator. They're called the meta training and inference accelerator MTIA and it starts quite simply like right now in
00:25production is MTIA 300 and that is used to train the AI
00:29behind ranking and recommendations. So if you open up your Instagram feed Danny and you see a targeted ad and
00:36you're like wow that's oddly specific to me. Meta sees merit in using its custom silicon to train that algorithm.
00:43But bigger picture as they train more models principally on NVIDIA and AMD gear running those models the inference phase
00:51the economics of doing that they see a better opportunity to do their own chips. So now through the end
00:57of 2027 they will release four generations
00:59of it and we were able to go inside the chip lab and check it out and to all intents
01:04and purposes it's real and it disputes a little bit recent reporting that meta had kind of given up on
01:10its own proprietary silicon. Yeah it just puts more pressure on NVIDIA. Ed I look forward to your reporting on
01:18this. Your program Bloomberg Tech kicks off in just about one hour's time and we'll hear more about meta's chips.
01:24Entering the chat after we already saw a little bit of a wave being made by Google chips and of
01:33course Amazon has its own chips so it's getting to be a big fight.
Comments