The Three Musketeers

So who are these three OpenAI engineers defecting to Meta?

Peter Hoeschele is an expert in strategy and operations with previous experience at Deloitte, and a track record of creating data center network designs.

Shamez Hemani was Head of Infrastructure Strategy and Finance since 2022, and integral in OpenAI’s trajectory.

Anuj Saharan is an engineer working on, according to his Linkedin: “superduperintelligence infra” with previous ties to Nvidia.

All three of these fellows were senior leaders in OpenAI’s compute/infrastructure group, especially around Stargate.

Hoeschele helped launch and lead the Stargate project, while Hemani worked on compute strategy and business development for related efforts, and Saharan built systems and infrastructure for AI models in designing what he’s on record as describing as “the world’s largest computer.”

What does their departure mean for OpenAI?

The Plan Going Forward

Well, first of all, many view the changes as a signal that OpenAI may be pivoting on StarGate.

Specifically, the Bloomberg writers penning the report, Shirin Ghaffary and Dina Bass, write:

“Stargate was announced last year at the White House as a $500 billion venture between OpenAI, Oracle Corp. and SoftBank Group Corp. The project has more recently morphed into a catchall brand for all of OpenAI’s data center plans.”

But will the company still be taking the same route toward data center expansion, without three key members of its former team?

I thought that assertion, “morphed into a catchall brand,” was interesting. It sort of suggests a listlessness in advancing Project StarGate, which hasn’t really been talked about much in the White House this year, either.

Meanwhile, Meta is on a roll.

Zuckerberg’s Vision

Famously, the Zuck is plugging over $135 billion in capital expenditures into the AI wing of his newly renamed company, while laying off thousands of people to “align the company with AI.” Then there’s the Meta Ray-Bans pioneering a hands-free interface. That’s a lot, and beyond that, Zuckerberg has made major investments in infrastructure and hardware.

“Before the end of the decade, Zuckerberg has promised to spend hundreds of billions more on AI infrastructure projects,” Ghaffary and Bass write.

And then there’s Muse Spark which Meta claims is “purpose-built to prioritize people.”

“Muse Spark will power a smarter and faster Meta AI, and over time, unlock new features that cite recommendations and content people share across Instagram, Facebook, and Threads,” spokespersons write on the Meta web site. “Muse Spark is our most powerful model yet. It currently powers the Meta AI app and website, and will be rolling out to WhatsApp, Instagram, Facebook, Messenger, and AI glasses in the coming weeks. We will also be offering the model in private preview via API to select partners.”

So it’s going to be in the glasses, too.

Maybe one way to see this is as a tectonic shift that just means StarGate won’t be the object of focus that some of us thought it would be. Another way to see it is that companies will continue in the data center race by any other name. In either case, people in the communities chosen for many of these facilities are less than enthused, so that’s another wrinkle in trying to get big data center initiatives off of the ground.

But the news of the staffing changes seems to indicate that we’re less likely to hear a lot about StarGate outside of the infrastructure world.

Obviously, OpenAI hasn’t abdicated its role as a big player in models. Its relationships to Anthropic are front and center as we look at who’s who in trying for AI safety and ethics, for example. As for public/private partnerships, the rift between Anthropic and the Department of Defense has made big headlines this year, and is probably a bigger point of interest than what press releases are going out around StarGate or any given data center project. But that could change. Stay tuned.