OpenAI Stargate Leaders Depart in Latest Shakeup to Data Center Strategy
The Takeaway
- Three senior OpenAI executives behind the Stargate initiative are departing.
- OpenAI shifted from building own data centers to renting compute capacity.
- Company aims for $600 billion in compute capacity over five years.
Three senior OpenAI executives who helped launch the company’s original Stargate data center initiative have left or are preparing to depart the ChatGPT maker in the coming days, according to several people with direct knowledge of the matter.
Peter Hoeschele, an OpenAI executive who played a key role in getting the Stargate effort off the ground, has already left the company, one of the people said. Two others—Shamez Hemani, who worked on compute strategy and business development, and Anuj Saharan, another leader in OpenAI’s compute organization—have announced their departures to colleagues, two other people said..
The departing executives are all going to the same new company, according to one of the people, though the name of that firm could not be learned.
The infrastructure team has been at the center of one of the biggest challenges that OpenAI and its rivals are contending with: how to ensure access to enough computing power to meet soaring demand for AI products at prices that the startup can sustain. OpenAI has said it plans to secure more than $600 billion in compute capacity over the next five years—spending plans that have prompted it to ink deals with an array of chip and cloud giants, from Oracle to Amazon Web Services.
The departures come after a broader reorganization of OpenAI’s infrastructure efforts. Late last year, the company hired former Intel executive Sachin Katti as head of compute and infrastructure.
“We’re grateful for the contributions Peter, Shamez, and Anuj have made to OpenAI and wish them the very best in what comes next,” an OpenAI spokesperson said in a statement. “Sachin Katti was recently hired to lead our Industrial Compute organization, which is making solid progress as we scale the infrastructure required for the next generation of AI systems.”
Last year, the team inked deals for 8 gigawatts of compute capacity in the coming years, short of the 10 gigawatt goal the firm set out to reach when it announced its $500 billion Stargate initiative in January 2025. That’s still a massive amount of capacity, however. OpenAI ended last year with access to around 2 gigawatts of capacity, the same amount of power that can be generated by two nuclear power plants.
Saharan posted a note in OpenAI’s Slack Thursday about his decision to depart, saying that “getting to help build literally the world’s largest computers over and over again with you all has been the ride of a lifetime.”
Original Stargate Leaders
Hoeschele, Hemani and Saharan were part of the original Stargate team, an ambitious initiative aimed at helping OpenAI build and operate its own large-scale data centers. The effort, announced at a televised White House event shortly after President Trump took office, was central to the company’s early push to secure long-term control over the infrastructure needed to train and run advanced AI models.
But OpenAI ultimately pulled back from that approach last year after struggling to line up financing and finalize the structure of a proposed joint venture with SoftBank and Oracle. In its place, the company has increasingly leaned on partnerships, opting to rent large amounts of compute capacity from cloud and infrastructure providers rather than directly owning the assets.
Following Katti’s arrival, several executives who had previously reported directly to OpenAI President and co-founder Greg Brockman—including Hoeschele—were moved under Katti, who reports to Brockman. Hoeschele went on to lead a group known internally as Ecosystems and Partnerships for Industrial Compute, or EPIC.
OpenAI does not plan to find a replacement for Hoeschele’s role, according to a person with direct knowledge.
Swipes at Anthropic
Though OpenAI’s Stargate strategy evolved last year, the firm believes that its aggressive spending strategy has given it a leg up against its competitors. In a recent memo OpenAI shared with investors, OpenAI took a swipe at rival Anthropic, which it claims is “operating on a meaningfully smaller curve” when it comes to compute capacity.
OpenAI said it believes Anthropic had 1.4 gigawatts of capacity at the end of last year, while OpenAI had 1.9 gigawatts. But OpenAI said it plans to ramp its capacity more steeply, with total gigawatts in the mid-single digit range at the end of this year and more than 10 gigawatts in 2027.
In contrast, it believes Anthropic will have 3 to 4 gigawatts in 2026 and 7 to 8 gigawatts by the end of 2027. (Anthropic hasn’t publicly disclosed its compute plans, but The Information has reported it has discussed securing at least 10 gigawatts of capacity over the next several years.) Bloomberg first reported on OpenAI’s investor memo.