Karen Moscow

The strategy is buy compute at scale from NVIDIA and AMD, but also use custom silicon where Meta's workloads are uniquely its own.

Because in the AI race, it isn't just about the models.

It's about the compute behind them.

Okay, so we broke the story on this one this morning.

The reaction in Metashare is kind of muted.

Remember, Meta's still a big buyer of NVIDIA and AMD GPUs.

Let's get more on Meta's own chip plans, the in-house and custom silicon with Bloomberg's Riley Griffin, who came with me to Fremont.

And it was an interesting experience, right?

Like, this is both.

They want to do custom silicon because they have a lot of internal workloads, lots of AI inference, but they're going to have to keep buying from the big names as well.

Yeah, it's a strategy of more is more is more, and that is for the chips, that is for gigawatts, that is a total diversification approach to securing that compute, given the insatiable demand, Ed, right now for that.

Talk to us a little bit about the M&A that was involved, Riley, to beef up the team, because Ed hinted at it in his piece there, but they were pushed back by one Japanese company, but then were able to buy a local chip-making focus.

Yeah.

So early last year, Meta had made an approach for Furiosa AI, a Korean chip maker that rebuked it.

It turned down that $800 million offer we then reported in March.

Later towards the end of the year, we also reported that Meta was successfully able to acquire Revos.

And with that,

its bench of talent.

And that was really the key here.

They were able to bring in more than 400 employees who are now peppered across MTIA, this team, some of whom we met in Fremont, that are making these four different chips.