Where America’s Most Valuable Private Companies Are Building

When a private company raises billions of dollars in a single round, the headlines focus on valuation, products, and the founders. The real estate decisions that follow tell a more lasting story. A funding round can be spent in a few years. A 12-year lease, a 500-acre factory site or a gigawatt-scale data center campus shapes a company, and the communities around it, for decades.

Ten of the most valuable venture-backed companies in the U.S. are Anthropic, OpenAI, Databricks, Waymo, Anduril, Cognition, Ramp, Prometheus, Ripple and Figure AI. Together they offer a rare look at how the next generation of corporate giants is making location decisions. Several are now among the largest office tenants in their markets. Others are building factories, fleet depots and data centers at a pace that rivals the largest public companies. One of them, Anthropic, filed to go public on Oct. 26, and its IPO filing offers an unusually detailed look at how one of these companies is planning for growth.

Many of these moves are classic site selection decisions driven by labor, power, speed to market, and economic incentives. Anduril chose a 500-acre site near Rickenbacker International Airport outside Columbus, Ohio, for its Arsenal-1 manufacturing campus, and the state's JobsOhio office awarded the company an estimated $310 million grant tied to 4,008 new jobs and at least $910.5 million in capital investment. In Abilene, Texas, the first campus of OpenAI's Stargate data center program received an 85% discount on its city property taxes for two decades of operation. Anthropic launched a $50 billion data center program starting with sites in Texas and New York. Even a 150-person startup like Prometheus crossed San Francisco Bay to Oakland because it needed industrial space with enough electrical capacity.

This analysis looks at where each company is headquartered, what kind of real estate it uses, and how its workplace policy affects how much space it needs. The result is three distinct footprint profiles, plus a fourth group that has stayed remarkably light despite enormous valuations.

At a Glance: Headquarters, Property Type and Workplace Model

The Urban Office Cluster: Anthropic, OpenAI, Databricks and Ramp

The AI labs and software companies on this list have become the most important source of large-block office demand in San Francisco, and they are doing it by clustering. Rather than spreading out across a region, they are stacking buildings within a few blocks of each other to keep research, engineering, and leadership teams close together.

Anthropic has assembled more than 1 million square feet in a three-block stretch of Howard Street in downtown San Francisco, a corridor landlords now market as AI Alley. The anchor is a full-building lease for 300 Howard St., a 25-story, 466,000-square-foot tower, plus an adjacent 18,000-square-foot historic building. The lease runs 144 months and includes 14 months of free rent. The company also converted a 240,000-square-foot sublease at 500 Howard St., its original headquarters, into a long-term direct lease, and added about 100,000 square feet at 400 Howard St. and about 70,000 square feet at 405 Howard St. It is also reportedly weighing a 100,000-square-foot industrial building in the Mission District.

OpenAI took a similar approach in Mission Bay, where its footprint now exceeds 1 million square feet. Much of that space was assembled through subleases of buildings that Uber and Dropbox no longer needed, along with a full-building lease of a six-story former Gap building. In 2026, OpenAI added a five-building, 450,000-square-foot campus in Mountain View on a 10-year lease that also gives the company the first opportunity to buy the property. The move puts an office within reach of South Bay talent who did not want to commute to San Francisco three days a week.

Databricks is the most traditional multi-market office user of the group. It is moving its headquarters from 160 Spear St. to about 150,000 square feet across nine floors at One Sansome St., roughly doubling its San Francisco space. It now occupies more than 455,000 square feet across two new towers at the Cityline development in Sunnyvale and more than 270,000 square feet in the Seattle-Bellevue area.

Ramp is the bicoastal outlier. The New York-based finance automation company started with a 66,000-square-foot sublease from Microsoft at 28-40 W. 23rd St. in the Flatiron District in 2023, grew to 132,000 square feet there in 2024, and signed a deal in early 2026 to more than double its New York headquarters again. Weeks earlier, it leased about 82,000 square feet at 235 Second St. in San Francisco, roughly quadrupling its West Coast presence.

Two patterns stand out. First, subleases have been the on-ramp. Several of these companies grew into space that larger tech firms gave up after the pandemic, then converted that space into long-term direct leases once their growth was proven. Second, the long-term commitments are arriving with generous concessions, a sign that landlords still see these tenants as essential to filling downtown towers.

The Compute Footprint: Data Centers as Real Estate

For the AI labs, the largest real estate commitments are not offices at all. They are data center campuses measured in acres and megawatts.

OpenAI's Stargate program began on roughly 1,000 acres on the southeast edge of Abilene, Texas, and has since expanded to six more announced sites: two additional Texas locations plus sites in New Mexico, Wisconsin, Michigan and Ohio. OpenAI generally does not own these facilities. Partners such as Oracle and SoftBank develop, own or lease the buildings and hardware, and OpenAI consumes the capacity. Not every site has been welcomed. Lordstown, Ohio, has since banned future data centers, and the Michigan site has faced local opposition.

Anthropic began with a similar partner-led model. Its $50 billion U.S. data center program is being built with Fluidstack, starting in Texas and New York, and Amazon has opened an $11 billion, 1,200-acre data center campus in Indiana dedicated to Anthropic's workloads. The strategy is now shifting. Under an initiative called Anthropic Direct, the company is leasing dedicated data centers that partners secure, power, and operate to its specifications, including a 20-year lease on a facility that has not yet been built. It has disclosed about $265.6 billion in fixed or minimum compute commitments with terms of up to 20 years, and it plans to add data center capacity outside the U.S.

These campuses are where economic incentives, utility capacity, and community relations matter most. A 20-year property tax abatement can be worth hundreds of millions of dollars on a campus of this size, but as the Ohio and Michigan examples show, incentives do not guarantee local support.

The Physical-World Footprint: Anduril, Waymo, Figure AI and Prometheus

The companies building hardware have real estate needs that look more like a manufacturer or logistics company than a software startup.

Anduril has the largest footprint on the list. Arsenal-1, its hyperscale manufacturing campus in Pickaway County, Ohio, is planned to reach 5 million square feet, close to the size of the Pentagon, with nearly $1 billion of company investment and an eventual workforce of about 4,000 people at an expected average salary of $132,000. The first two buildings total about 895,000 and 924,000 square feet. Anduril plans to lease the property rather than own it. The campus joins existing Anduril factories in Mississippi, Rhode Island and Georgia, its Costa Mesa, California, headquarters and a major new campus planned in Long Beach for 5,500 workers.

Waymo's footprint follows its fleet. The company installs its self-driving technology at a 239,000-square-foot vehicle integration plant in Mesa, Arizona, which can produce tens of thousands of vehicles a year at full capacity. Every market it serves also needs local depots for charging, cleaning, and maintaining vehicles. In some cities, Waymo is securing these industrial sites before robotaxi service is even legal, including a 32,000-square-foot facility in Northeast Washington, D.C.

Figure AI keeps everything under one roof. Its 98,700-square-foot campus at 3960 N. First St. in San Jose, California, combines its headquarters, engineering labs, and BotQ, a factory designed to build 12,000 humanoid robots a year. The company moved there in 2025 from about 28,000 square feet in Sunnyvale and added a 64,000-square-foot lab and manufacturing building across the street in 2026.

Prometheus, the AI startup co-led by Jeff Bezos, shows how power has become a site selection criterion even for young companies. After leasing about 30,000 square feet of office space at 101 Mission St. in San Francisco, it signed a 100,000-square-foot lease at the former American Steel complex in West Oakland. The company reportedly needed industrial space with 8,000 amps of electrical service, which San Francisco's limited industrial inventory could not provide. It has also been in talks for about 38,000 square feet in London's King's Cross.

The Light Footprint: Cognition and Ripple

Two companies on the list prove that valuation and square footage do not always move together.

Cognition, the San Francisco maker of the Devin AI coding agent, does not publish its street address and has announced no major leases. Its 2025 acquisition of Windsurf added Bay Area talent, but its headcount remains small relative to its valuation.

Ripple, the blockchain payments company headquartered at 315 Montgomery St. in San Francisco, runs a deliberately distributed model. It operates about 15 offices worldwide, including New York, London, Singapore, Dubai and Toronto, and has not made the kind of large lease commitments seen elsewhere on this list.

Workplace Policy Is Shaping the Square Footage

Attendance policy is one of the most important and least discussed drivers of these footprints. Based on company job postings, published reporting, and employee accounts, the 10 companies fall along a wide spectrum.

At the strictest end, Anduril and Figure AI require employees on-site, which is largely a function of hardware and, for Anduril, classified defense work. Cognition goes further than any company on the list. After the Windsurf acquisition, it offered the employees it inherited a choice between a buyout and a six-day, 80-hour in-office workweek. Ramp has moved steadily in the same direction, from using hybrid flexibility to recruit bankers, to a three-day minimum for new hires, to recent San Francisco postings that require five days a week in the office.

In the middle, OpenAI requires employees in the office Monday through Wednesday, and Databricks expects three days a week with badge attendance reported to managers. Databricks even described its One Sansome renovation internally as a project to support its return-to-office plan, a rare case of a company explicitly tying a lease decision to an attendance policy.

At the lighter end, Anthropic's job postings set an expectation of being in an office about 25% of the time. Waymo's corporate staff follow a light hybrid model, though its depot and plant workforce is on-site by necessity. Ripple describes itself as flexible-first, letting employees choose home, office or a combination.

The most useful lesson is that attendance policy alone is a poor predictor of footprint size. Anthropic has one of the lightest attendance requirements and one of the largest office footprints in San Francisco, because its headcount has grown so quickly. Its workforce grew from about 940 employees at the end of 2024 to about 5,300 by September 2026. Cognition has the most intense in-office culture and one of the smallest footprints because its team is small. Square footage is the product of three factors: headcount growth, attendance intensity, and the type of work being done. Companies that plan space around only one of those factors usually get it wrong.

What the Data Doesn't Show

Private companies disclose far less about their real estate than public ones, and this analysis relies on lease announcements, public records, company job postings, and published reporting. Anthropic's IPO filing is a notable exception, offering far more detail on its compute commitments and headcount than private companies typically share. Some square footage totals are estimates, several deals have not been confirmed by the companies, and workplace policies are often applied differently by team or role. Ramp's current New York total, Waymo's full depot network, and the office sizes of Cognition and Ripple are not publicly confirmed. The figures here should be read as directional rather than exact, and in a market moving this quickly, several will likely change within months.

Implications for Corporate Real Estate Leaders

Most companies are not raising billions of dollars a year, but the decisions these 10 are making carry lessons for any organization planning growth.

• Settle the workplace policy before signing the lease. Databricks and Ramp show how closely space needs track attendance expectations. A company that commits to space for a three-day policy and then moves to five days, or the reverse, will be either crowded or paying for empty desks.
• Treat power as a primary site selection criterion. Prometheus, Anduril, Figure AI, and every data center campus on this list chose locations based in part on electrical capacity. For manufacturing, labs, and advanced R&D, utility availability can now rule out a market faster than rent or labor costs.
• Negotiate incentives before announcing a location. Anduril's $310 million grant and the Abilene tax abatement were secured as part of the site selection process. Once a company publicly commits to a site, most of its leverage disappears.
• Use flexible space as a bridge, not a destination. Subleases helped Anthropic, OpenAI, and Ramp grow quickly without overcommitting, then converted to long-term direct leases once their growth was proven.
• Weigh clustering against cost. The AI labs are paying for the benefits of proximity to talent and to each other. That trade-off makes sense for research-driven companies, but many functions can perform just as well in lower-cost markets.
• Expect community scrutiny on large projects. Local opposition in Ohio and Michigan shows that incentives and approvals are only part of the equation. Public engagement belongs in the project plan from day one.

How Site Selection Group Can Help

Whether a company is planning its next headquarters, a manufacturing campus, a data center, or a new regional office, the same questions apply: where the talent is, whether the power is available, what incentives can be secured, and how much space the workplace strategy truly requires. Site Selection Group helps companies answer those questions with location analytics, labor market analysis, economic incentive negotiation, and real estate execution across all 50 states.

Unlike traditional brokerages that also represent landlords and developers, Site Selection Group exclusively represents tenants. That independence means our recommendations are driven only by what is best for our clients, not by any building, market, or landlord relationship.

To discuss your next location decision, reach out to us.