“I recently shared that, by the end of this year, we'll have about 350,000 H100s, and including other GPUs, that will be around 600,000 H100 equivalents of compute.”

In context

“of the tools were social-, commerce- or maybe media-oriented that it might be possible to deliver these products by solving only a subset of AI's challenges. But now it's clear that we're going to need our models to be able to reason, plan, code, remember and many other cognitive abilities in order to provide the best versions of the services that we envision. We've been working on general intelligence research and FAIR for more than a decade. But now general intelligence will be the theme of our product work as well. Meta has a long history of building new technologies into our services, and we have a clear long-term playbook for becoming leaders. And there are a few key aspects of this that I want to take some time to go through today. The first is world-class compute infrastructure. I recently shared that, by the end of this year, we'll have about 350,000 H100s, and including other GPUs, that will be around 600,000 H100 equivalents of compute. We're well positioned now because of the lessons that we learned from Reels. We initially underbuilt our GPU clusters for Reels. And when we were going through that, I decided that we should build enough capacity to support both Reels and another Reels-sized AI service that we expected to emerge so we wouldn't be in that situation again. And at the time, the decision was somewhat controversial, and we faced a lot of questions about CapEx spending, but I'm really glad that we did this. Now going forward, we think that training and operating future models will be even more compute-intensive. We don't have a clear expectation for exactly how much this will be yet, but the trend has been that state-of-the-art large language models have been trained on roughly 10x the amount of compute each y”