Takeaways
- Meta is significantly increasing our compute power by investing in two large-scale new AI data center initiatives.
- Advancing AI requires deep integration between hardware and software design.
- Meta teammates collaborate across disciplines to holistically solve infrastructure challenges.
There’s a saying that software eats hardware for lunch. But that’s where Open Hardware Ecosystem Lead Dharmesh Jani (‘DJ’) and Meta mechanical engineer Steve Mills will tell you you’re wrong.
According to DJ, the industry is coming full circle, and that excites him as a hardware engineer. “We are finally in a renaissance area of hardware, and suddenly, hardware engineering has become very cool.”
At the center of this disruption is rapid AI innovation. Meta chairman and CEO Mark Zuckerberg recently shared about our vision to bring superintelligence to people everywhere, giving them the freedom to use AI to add value to their own lives. According to Mark, getting there will take massive effort over the coming years. “We'll bring online ~1GW of compute in '25, and we'll end the year with more than 1.3 million GPUs. We're planning to invest $60-65B in capex this year while also growing our AI teams significantly, and we have the capital to continue investing in the years ahead.”
Alongside this vision, model sizes are scaling up rapidly, putting new and tougher demands on the infrastructure and teams that build them. “Software typically moves at a faster pace, whereas hardware takes time to develop. Anticipating that in the perfect way is a very, very hard task, and that makes things extremely challenging,” explains DJ.
Solving complex infrastructure problems requires fresh thinking
Advancing AI requires a new approach that calls for deep hardware and software team cross-pollination and integrated co-design disciplines. “No longer can you design something in your area and simply deliver it to somebody. It has to be designed together, and that is a new way of operating,” shares DJ, who has spent more than 20 years throughout his career reshaping how teams collaborate.
Like DJ, Steve believes that organization-wide engagement is essential for optimizing systems. “The technologies that we're developing and delivering are essential to enabling the software that's being built on top of it. Unlike many other technologies, the code development work required in AI systems to optimize the software and the hardware at the same time is much more integrated.”
“AI-driven workloads are growing and changing rapidly, impacting infrastructure in an unpredictable manner,” DJ explains. “What worked before will not work moving forward. Hardware engineers have to take a clean slate approach to solve these challenges. It’s very exciting to work on and has created a golden era for hardware engineering.”
Breakthroughs come from breaking down barriers
DJ advises anyone entering this line of work to redefine what they know from the ground up, wipe the slate completely clean of any bias and be prepared to collaborate. “No longer are we in an era where problems are solved in a siloed fashion. You have to think a lot broader than your particular domain. You have to learn how to work with colleagues with diverse backgrounds and skill sets in a collaborative fashion. For example, if you're trying to solve for networking, it also includes some solution and co-design on the software, accelerator, interconnect or thermal side.”
When teams come together and are given the space to problem solve, anything’s possible, adds Steve. “Most of my work at Meta involves collaborating with cross-functional teams. Whether it's a new process or a new product, it’s about finding the right people, getting teams set up and then letting them run with the support needed to operate independently. It’s exciting to see the results as our products, like air-assisted liquid cooling, hit data center floors and deliver higher power capabilities than previously achieved.”
“The mental model that I follow for myself is to be perennially curious. On a daily basis, I learn from diverse perspectives. I collect ideas people share because, oftentimes, solutions appear in a very serendipitous fashion through the connection of very disparate and unconnected ideas. I would encourage anybody walking into this space to be very open-minded.” — Dharmesh J., Open Hardware Ecosystem Lead
Working together both drives innovation in AI and strengthens the teams tackling some of the industry’s toughest infrastructure challenges, according to Steve.
“Unlike in some companies where you're developing one particular type of product over and over and over again, we have access to numerous products, teams and design constraints. This provides a lot of opportunity for engineers to branch out, learn various products and build on their careers.” — Steve Mills, Mechanical Engineer
What’s most personally rewarding for DJ is his ability to have a large-scale impact. “Meta serves 3.4 billion users on a daily basis. We need to build infrastructure that serves the needs of these users. At Meta, I have a much bigger forum to drive forward that vision.”
“Since day one, working at Meta has been one of the most gratifying work experiences of my life,” reflects DJ.
Stay connected.
- Are you interested in joining our team? Explore open roles.
- Listen to the Meta Tech Podcast for in-depth interviews on technical challenges.
