Today, Meta announced that it hired Broadcom and Intel veteran, Dr. Alexis Björlin, as its VP of Infrastructure Hardware Engineering. She will report directly to Meta veteran and infrastructure leader, Jason Taylor. While this seems on the surface like a very straightforward hire, I believe it represents something more important than that. It begs the question, why would Meta need such an industry-heavy-hitter to lead hardware engineering? I had the chance to sit down virtually with Taylor to ask him about the hire, Meta’s datacenter semi strategy, and more.
Taylor first explained to me that Meta has a set of unique datacenter requirements, given the incredible diversity of its business. The company performs an incredible amount of video streaming, which it addresses with an in-house designed video transcoding chip. If you spend any time on Instagram or Facebook, you can personally see this in action.
Meta also heavily utilizes technologies in areas like computer vision, text to speech, and language translation where training has moved from CPUs to NVIDIA GPUs. Taylor complimented the industry at providing the right silicon for those workloads.
One of the more interesting parts of my conversation with Taylor was around recommendation engines. We have all experienced these in action on large-scale e-commerce sites like Amazon, where we look at or buy something and it recommends the next product we should look at. We see this on Netflix as well where we watch a certain video, and it recommends two or three related videos. Meta’s properties like Facebook and Instagram use this extensively and heavily leverage machine learning.
According to Jason Taylor, Meta’s silicon strategy’s primary focus is working with industry-leading vendors like AMD, Broadcom, Intel, and NVIDIA and that won’t change in the near-term. Meta also appears to be working with Marvell as well. That said, there is now such incredible diversity in workloads, particularly in areas like AI, that a new approach is required. Specifically, Taylor looks at the rapidly growing need for memory or networking bandwidth to provide the best recommendation capabilities in the future as an area where new and custom silicon approaches are needed. This growing need is being addressed through aggressive investments in people, partners and in-house silicon engineering.
This, in part, is one of the drivers for Dr. Alexis Björlin coming into the picture. Meta is bringing Björlin into the company to meet its rapidly growing datacenter needs. These needs will be a mix of general-purpose silicon from its CPU, GPU and networking partners and in-house special purpose processors for custom workloads. Unlike Apple who has strong-armed its semiconductor partners, I applaud Meta for continuing to invest in them. Taylor was very complimentary in my conversation and I don’t believe it’s window dressing.
It is great to see Meta likely creating more of its own silicon as I have said for a decade that “silicon is eating the world”. I look forward to learning more about Meta’s in-house silicon and will report back in the future.