Synopsis
Peter Hoeschele, a key OpenAI executive behind the early Stargate data centre effort, has left the company, according to a report by The Information. Two other senior leaders, Shamez Hemani and Anuj Saharan, are also set to exit in the coming days. All three are expected to join the same new startup.Additionally, two other senior figures, Shamez Hemani and Anuj Saharan, have also told colleagues they are exiting the company in the coming days. Hemani is focussed on compute strategy and business development, and Saharan is another leader at OpenAI’s compute organisation.
All three are expected to join a new startup, although its name has not been disclosed, the report said.
The three executives were part of OpenAI’s Stargate effort, which was meant to give the company greater control over large-scale data centre capacity instead of relying on external providers.
“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 to The Information. “Sachin Katti was recently hired to lead our Industrial Compute organisation, which is making solid progress as we scale the infrastructure required for the next generation of AI systems.”
The company has also confirmed that it does not plan to fill Hoeschele’s position.
These departures come amid a broader internal reshuffle of OpenAI’s infrastructure and compute teams. Late last year, the company brought in former Intel executive Katti to head its compute and infrastructure division.
Stargate was originally one of OpenAI’s most ambitious infrastructure plans. It aimed to secure direct control over data centres and long-term compute supply.
However, the strategy lost momentum after the company struggled to arrange financing and finalise a proposed joint venture involving SoftBank and Oracle. As a result, OpenAI has increasingly moved towards partnerships with cloud and infrastructure companies, renting compute capacity rather than building and operating things in-house.
This shift reflects a wider problem in the industry, as AI companies struggle to get enough computing power to build and run their models while keeping costs under control.
OpenAI has previously said it aims to secure more than $600 billion worth of compute capacity over the next five years. This aggressive plan has already led to multi-billion-dollar agreements with major providers, including Oracle and Amazon Web Services.
Last year, the infrastructure team locked in deals for around 8 gigawatts of compute capacity over the coming years. This fell short of an earlier internal target of 10 gigawatts tied to the original $500 billion Stargate plan announced in January 2025.
Even so, the scale is big. At the end of last year, OpenAI reportedly had access to around 2 gigawatts of compute, roughly comparable to the output of two nuclear power plants, the report said.
