If you ask Omar B., at the end of the day, everything relies on the network.

As senior director of data center networking at Meta, Omar leads the teams that develop and operate our global data center network infrastructure. Their work supports all the underlying services and AI models running inside Meta data centers for Facebook, Messenger, Instagram, WhatsApp, ads and now Meta AI. For Omar, networking is the thread that ties it all together.

“Part of what drew me to networking was its intricacy and interconnectedness. Everything runs across the network. When I ask a question through my Ray-Ban Meta glasses, it goes from my glasses to my phone and across the internet to Meta data centers. There, our GPUs quickly formulate a response, and the information starts streaming all the way back to me. Our networking infrastructure must work together seamlessly to support a smooth back-and-forth conversation.

There are tons of challenging problems to do that well. For example, it takes light 5 microseconds to travel 1 kilometer, so we have to figure out creative ways to work around that limitation. If something goes down, my team has to work quickly to identify the offending component in a 100K+ GPU data center (which has millions of components). Our network is one of the largest distributed systems in the world with an incredible mix of hardware, software and systems, so we’re always learning something new.”

The Meta network spans the entire globe, providing high-speed networking for everyone around the world using our technologies. As an example, the network team at Meta recently announced Project Waterworth — the world’s longest subsea cable that, once complete, will span more than 50,000 km to reach five major continents and open three new oceanic corridors. The network team also works on networking for highly demanding AI models, building and optimizing the network for 100K+ GPUs and beyond.

Leveraging a company-wide focus on AI innovation

Omar has been with Meta since 2013. In that time, he’s worked on nearly every part of our network — from the edge where Meta connects to the rest of the internet, to the global backbone, to his most recent focus on data center networking.

The company-wide focus on AI drives Omar and his team to build networking for our immediate AI needs and to advance the state-of-the-art in networking for AI.

Meta Chairman and CEO Mark Zuckerberg recently made a series of announcements about our increased infrastructure investments to support personal superintelligence. Meta plans to bring online 1GW of new compute this year and have more than 1.3 million GPUs total by year-end. We’re also building two new data centers. The first is Prometheus, a multi-region data center campus in Ohio that’s expected to reach 1GW by the end of 2026. The second data center is Hyperion. Located in Richland Parish, Louisiana, Hyperion spans 11 buildings and 2,700 acres, with the ability to scale up to 5GW over several years. The push for this GW-scale AI capacity is driving all parts of the network infrastructure — beyond the data center, across the long-haul backbone and out to the edge of the Meta network.

Working across disciplines to build what’s next

While having a motivating company vision is important, Omar also believes a key principle at Meta is having one team that’s responsible for both building and operating the network. He feels this ownership has allowed his team to overcome technical hurdles and quickly pivot to what’s needed next for AI.

As Omar and his team work to build a data center networking infrastructure that can support the future of human connection, they source insights from a wide variety of disciplines across infrastructure. The team works with GPUs and custom accelerators from Meta, large-scale networking and server systems, host-based networking (including RDMA and TCP), and even the networking communication libraries used in PyTorch.

“We have experts in the group serving up Meta AI right now, as well as working on future tech. We can easily share with each other the operational challenges we’re solving, which helps us build more effective products for the next generation. It’s exciting and inspiring to work on these cutting-edge networks as they enable and define the future of AI for everyone!”