Fangran Xu has always been passionate about helping people connect with each other and the world around them. She believes that connection is the first step to helping people discover new opportunities and improve their lives.

“Growing up in China, I remember my Dad telling me that I should study abroad and decide my future for myself,” she recalls. “It never felt real until, one day, he showed me a magazine with a picture of MIT’s library and a group of students studying out front. Suddenly, I could see what my life would look like at an American university. That moment fundamentally changed my life, and it’s why I am so passionate about our mission of connection at Meta.”

Today, Fangran works as a custom silicon sourcing manager on the infrastructure supply chain and engineering team. She helps procure critical components for the Meta Training and Inference Accelerator (MTIA) program — custom-built silicon chips that support everything from the recommendation systems within Meta technologies to new generative AI products, services and advanced AI research.

MTIA is central to our mission of building the future of human connection and the technology that makes it possible. It helps deliver greater compute efficiency across our infrastructure and supports our software developers as they build advanced AI models — a critical focus as Meta works to bring the world’s first 1GW+ supercluster online. Fangran has been part of the MTIA journey since the very beginning.

Innovating without limits

Fangran remembers the early days of building MTIA as a challenging but informative experience, driven by a desire to push infrastructure capacity to new heights in support of growing company investments in AI. Her team’s work has only become more relevant as Meta turns its focus to developing superintelligence.

“We created MTIA after realizing existing silicon solutions could no longer satisfy our infrastructure deployment, so we decided to lean in and build custom silicon that could handle our specific generative AI workloads,” she explains. “Meta is constantly trying to push technology boundaries, and we dedicate a lot of resources to building advanced solutions that make our infrastructure more efficient.”

For Fangran, her team’s unrestrained access to compute resources is a big part of why they’ve been able to fearlessly pursue new technological innovations like custom silicon. They don’t have to worry about being limited by outside factors. Instead, they can simply focus on solving the immediate challenge at hand.

“Meta doesn’t have the same resource constraints that you often see at other companies — you’re not restricted to just one answer or approach. Instead, we start with the problem statement, understand all the possibilities, identify the most optimal solution for our needs and figure out how to work through any constraints along the way. This mindset is unique to Meta and has been crucial in enabling our technology vision.”

Collaboration leads to better hardware design

Another key differentiator to innovating at Meta? The ability to collaborate with cross-disciplinary experts across all levels of the company.

“Meta is distinct in that there’s no wall between software use cases and hardware deployment — we’re developing custom silicon to support our own AI workloads,” Fangran shares. “In a product company environment, you have to knock on hyperscalers’ doors to learn how they’re writing their software scripts and LLMs. But here, you can talk to the software team as your peers. This allows us to get very specific in how we design our ASICs, which drives greater efficiency, and we can quickly deploy them into production workloads across Meta technologies, so we see the direct impact of our work.”

Fangran believes this integrated model, combined with her team’s immense compute resources, is responsible for the rapid, industry-first innovation at Meta.

“And by the way,” says Fangran, “the first generation of our custom video transcoding silicon won the semiconductor Emmy last year.”