The OpenAI announcement that surfaced through Santa Clara County property records on February 25, 2026, didn’t exactly fit the description of a “secretive AGI fortress” that has been used in some discussion. The real filing revealed something more traditional, but no less important. This week, the business that created ChatGPT leased a 439,000-square-foot campus at 350 and 380 Ellis St. in Mountain View.
According to sources, the lease is for ten years. Norton LifeLock used to be located on the campus, which consists of three office buildings, but it left in 2021. The real estate deal is simple. There are no strategic ramifications.
| OpenAI Mountain View Campus and AGI Push — Key Information | Details |
|---|---|
| Company | OpenAI |
| Existing Bay Area Base | Mission Bay, San Francisco |
| New Campus Location | 350 and 380 Ellis Street, Mountain View |
| Total Square Footage | About 449,000 |
| Number of Buildings | Three to five (per varying reports) |
| Lease Term | 10 years |
| Lease Signed | Week of February 25, 2026 |
| Property’s Previous Tenant | Norton LifeLock (until 2021) |
| Property Owner | KKR Real Estate Finance Trust |
| OpenAI Valuation | More than $800 billion |
| Capacity Estimate | Up to 2,200 tech workers |
| Stargate Infrastructure Plan | About $500 billion over four years |
| Stargate Power Goal | 17 gigawatts |
| OpenAI Internal Roadmap | “AI research intern” capability by September 2026 |
| Reference Reporting | The Mercury News, SF Standard |
For the first time, the Mountain View facility significantly expands OpenAI’s Bay Area presence beyond its current Mission Bay hub in San Francisco. According to sources, OpenAI has been considering South Bay office locations since last summer and has spent months choosing between Sunnyvale and Mountain View.
OpenAI’s decision to locate in Mountain View puts it just a few blocks from Google, Waymo, and a number of other AI-focused businesses that have influenced the South Bay’s tech scene for twenty years. Beyond only convenience, the location is important. San Francisco doesn’t quite match the Ellis Street campus’s talent pool, academic connections with Stanford, and well-established AI research community, all of which are located within a reasonable commute.
The portrayal of the campus as “secretive” or “exclusively dedicated to AGI” is the aspect of the public discourse that requires greater examination than it has received. The lease has been easily reported to the public. Documents pertaining to real estate are not particularly confidential.
The Mountain View campus has not been officially identified by OpenAI as a dedicated AGI research site, and no significant information about the building permits, security arrangements, or team allocations that would set it apart from a typical tech office expansion has been made public. The “AGI fortress” concept that some observers have used perhaps reveals more about the general concern about OpenAI’s scalability than it does about particular real-world realities.
The actual weight of the lease is derived from the infrastructure environment beneath the extension. With the expansion of the Stargate super-data center project into a multi-stage, multi-location, $400 billion grand project, OpenAI announced an improved collaboration with Oracle and SoftBank. A 17 gigawatt power capacity is needed for the Stargate project, which would be sufficient to power 13 million American homes at once.
The Mountain View campus is a part of a much larger physical buildout that includes partnerships with Oracle, AMD, Broadcom, Microsoft, and Nvidia, data centers in Texas, and the kind of energy-and-compute infrastructure that sets OpenAI’s current scaling strategy apart from all other AI companies. The larger system to which the office space is attached is more important than the office space itself.
How to interpret the expansion has been influenced by the internal roadmap that Chief Scientist Jakub Pachocki and Sam Altman outlined in late October 2025. Before September 2026, OpenAI intends to make AI capable of working as an intern research assistant. A true AI researcher who can finish independent scientific study projects will appear by March 2028. Although the timescale is still aggressive, it is phrased more cautiously than previous “AGI by 2025” projections.
Altman has admitted that OpenAI’s description of its own advancements has changed. “In the past, we believed that AGI will materialize at some miraculous point in the future. However, we now see that it is more akin to a process, and you are already traveling this path. In this context, the Mountain View campus is not a single significant AGI milestone, but rather one of the tangible examples of that process.
Observing OpenAI’s growth throughout 2026 gives the impression that the firm has reached a stage when the obstacle to its goals is actually physical rather than just intellectual. Altman has stated that the United States’ current electricity infrastructure is already overloaded, that orders for gas turbines are not expected to be fulfilled until 2028, that it takes at least ten years to develop a nuclear power station, and that political opposition frequently impedes renewable energy initiatives.
The relatively simple aspect of the buildout is the Mountain View office lease. The most difficult aspects are the data centers, the energy capacity, and the personnel that the campus will eventually need to sustain. The next few years will determine whether OpenAI produces AGI by 2030, by 2028, or in a later timeframe that no one can forecast with certainty. One aspect of the solution that has been physically resolved is the 449,000 square feet at 350 and 380 Ellis Street. Even cautious Silicon Valley observers are finding it difficult to keep up with the speed at which the remainder of the picture is still being drawn.