OpenAI has secured a massive industrial foothold in the East Bay. The company signed a lease on a 202,000-square-foot waterfront site at Portside Commerce Center in Richmond, California, spanning 15.75 acres near the Richmond ferry terminal. The building previously sat vacant after Moxion Power, a battery startup that ceased operations in 2024, developed it but never moved in.
The property’s power specifications are the detail worth examining. It carries over 14,000 amps of electrical capacity. Standard industrial buildings run between 200 and 4,000 amps. That gap isn’t incidental. Power at that scale is a prerequisite for advanced manufacturing, large-scale compute, or both. The SF Chronicle’s headline framed it directly: “OpenAI lands massive Richmond site for robotics push.”
The robotics angle is reported, not confirmed. According to an unnamed source cited by the San Francisco Business Times, and consistent with county records, the site is believed to be intended as a robotics manufacturing facility. OpenAI has not publicly confirmed what it plans to do with the building. That distinction matters. The physical facts of the lease are solid. The intended use is informed inference, not a company announcement.
Why does this warrant attention from AI practitioners? Physical AI is the space where software capabilities meet the real world – robots that perceive, reason, and act in physical environments. OpenAI has made no secret of its interest in this area. A 200,000-square-foot warehouse with industrial power capacity on the Bay Area waterfront is a different kind of infrastructure investment than a cloud compute contract. It requires physical space, physical access, and physical operations. You can’t spin it up in AWS.
The building’s prior tenant history adds context. Moxion Power built the facility for energy storage manufacturing before folding in 2024. The infrastructure it left behind, the power systems, the floor load ratings, the site layout, was designed for hardware at industrial scale. Whoever developed the Portside space built it to handle weight, heat, and electricity that most commercial real estate never sees. OpenAI is now the tenant inheriting those specifications.
For developers and engineers tracking OpenAI’s trajectory, the Richmond lease fits a visible pattern of physical expansion. The company has been building out its presence beyond San Francisco over the past 18 months. Whether Richmond becomes a robotics testbed, a manufacturing facility, or something not yet disclosed, the lease itself signals intent: OpenAI is acquiring the infrastructure that physical AI demands, not just the software capabilities.
What to watch: OpenAI’s official announcements regarding the facility’s purpose. Any robotics-related hiring or job postings tied to the Richmond or East Bay area. Movement from other frontier labs into comparable industrial real estate, if this is a pattern, it won’t be confined to one company. The 14,000-amp capacity figure is also a useful benchmark; future AI infrastructure stories should be measured against it.
The synthesis here is straightforward. The lease is confirmed. The robotics purpose is a credible, well-sourced inference. The building’s power specs make the inference more than speculation. Until OpenAI confirms, practitioners should treat Richmond as a strong signal rather than a fact, but strong signals from frontier labs at this scale are worth tracking.