Figure AI Has 740 Robots and 650 Humans — and the Robots Are Winning

A Robot Campus Takes Shape in San Jose

Figure AI has nearly quadrupled its physical footprint, moving from a 27,900-square-foot office in Sunnyvale to a 98,700-square-foot industrial flex building at 3960 North First Street in San Jose. CEO Brett Adcock calls it a "robot campus" on his LinkedIn page. The facility will house manufacturing, fleet operations, and engineering under one roof. The lease, confirmed in March 2025, places Figure inside the Assembly at North First tech campus, a six-building complex once occupied by semiconductor firm Lam Research and now shared with tenants like Logitech.

The move is not cosmetic. The Sunnyvale office suited a startup still in R&D mode. The San Jose facility is a manufacturing and operations hub, a distinction that matters. When a robotics company starts leasing industrial flex space instead of office suites, prototypes are giving way to production units and fleet management is becoming a daily operational concern.

The timing lines up with the company's financial trajectory. In February 2024, Figure closed its Series B round with backing from Microsoft, OpenAI Startup Fund, NVIDIA, Jeff Bezos, and Intel Capital. Adcock said at the time that the capital would go toward scaling AI training, robot manufacturing, expanding engineering headcount, and advancing commercial deployment. The San Jose lease is the physical manifestation of that allocation. A company does not quadruple its square footage to write research papers.

The new headquarters sits next to Logitech's hub within the Assembly at North First complex. Lease terms were not disclosed, but the campus was renovated in 2018 and offers the power, loading dock, and floor-plate flexibility that a company running both manufacturing and software teams needs.

Figure's team, drawn from Boston Dynamics, Tesla, Google DeepMind, and Archer Aviation, had 80 employees as of the February 2024 funding announcement. The company was founded just 21 months before that raise, meaning the San Jose move comes roughly three years into its existence. During that period, it also signed its first commercial agreement with BMW Manufacturing and entered a collaboration with OpenAI to develop next generation ai models for humanoid robots. That partnership aimed to enhance the robots' ability to process and reason from language, a capability Figure runs on Microsoft Azure's AI infrastructure.

The San Jose campus ties these threads together: the funding, the OpenAI collaboration, the BMW deal, and the stated goal of bringing humanoid robots into commercial operations as soon as possible. A company preparing to deploy robots at scale needs more than engineers. It needs floor space to build, test, and maintain a physical fleet, configured for industrial work, not open-plan collaboration. Figure's new headquarters is built for the phase that comes after the demo.

The Hidden Hiring Blitz

Figure AI's career page tells a story its press releases don't. The company is hiring security engineers, building a Global Security Operations Center, and paying up to $350,000 a year for application security talent, a salary band that rivals what OpenAI offers for its own security roles.

Zero G Talent's board shows four Figure roles added in the past week alone: a Software Engineer for Privacy & Data Governance, a Security Engineer for Vulnerability Management and Automation, a Security Engineer for Application Security, and a Gear Machinist. Two of the four are security and privacy positions. The pattern holds across the broader listings.

The Application Security Engineer role, posted on LinkedIn and still open as of this writing, asks for experience in penetration testing, vulnerability research, secure coding practices, hardware security, and securing embedded systems (including secure boot, secure identity, and OTA updates). The job sits on a dedicated Security & Privacy team, which means Figure isn't bolting security onto existing teams. It's building a division.

Then there's the Security Intelligence Analyst role for Figure's Global Security Operations Center, also in San Jose. The position focuses on threat monitoring, proactive identification of risks to executives and their families, and analysis of threats to company operations. A GSOC is standard at large companies with physical operations and executive travel. It's not standard at a robotics startup, unless that startup is deploying physical robots into commercial environments and treating the security surface accordingly.

The hiring isn't limited to security. Figure is also looking for a Power Electronics Engineer for Charging and an Electrical Engineer for Actuator Systems, both in San Jose. But the security and privacy roles stand out because they don't map to the traditional robotics hiring profile. They map to the deployment profile, the roles a company needs when it's putting physical robots into customer facilities.

Compare this to what Figure is not aggressively hiring for. Mechanical design roles exist (the Gear Machinist and Actuator Systems Engineer listings confirm that), but they don't dominate the board the way security and ML infrastructure do. The bottleneck isn't building the robot. It's securing the robot, the backend services it connects to, the data it collects, and the people who oversee its deployment.

Agility Robotics, a humanoid robotics company based in Fremont, is running a similar playbook. LinkedIn shows Agility posted a Staff Application Security Engineer role in Fremont two weeks ago, plus a Senior Application Security Engineer listing at the same location. OpenAI posted its own Security Engineer for Application Security in the Bay Area within the same timeframe. The entire sector is competing for the same narrow pool of security engineers who understand both software and physical systems.

That competition is the signal. When a robotics company's hardest-to-fill roles are in application security and vulnerability management, not kinematics or actuator design, it means the industry has moved past the prototype phase. The robots exist. The question now is whether they can be deployed without creating a new attack surface in every warehouse, factory, and home they enter. Figure's hiring board is a bet that the answer requires dedicated security teams, intelligence analysts, and privacy engineers before it requires better grippers.

Robots Outnumber Humans — Now What?

In June 2026, Adcock posted a chart on X with a caption that doubled as a milestone marker: "For the first time, robots now outnumber humans at Figure." By Q2 2026, Figure's robot population had climbed to roughly 740 units. Human headcount sat at around 650. The gap is widening, not closing.

The trajectory matters more than the crossover itself. Between 2022 and most of 2024, Figure's robot count barely registered against its human workforce. Then production scaled in early 2025. Robot numbers cleared 100 before year-end and went vertical over the following six months. Human hiring grew too, but at a fraction of the rate. The chart Adcock shared shows the robot line curving upward on what RoboHorizon described as an "exponential trajectory," while the human line flattens.

This isn't a publicity stunt built on creative accounting. Figure's robots are physical units rolling off a production line, many headed to real deployment sites. The company has a partnership with BMW to place Figure 01 humanoids at the automaker's Spartanburg, South Carolina plant, the same facility where BMW builds its X-series SUVs. That robots now outnumber humans inside Figure's own walls is a leading indicator: the machines being built are moving out faster than the company can hire people to oversee them.

A 740-to-650 ratio means Figure has crossed a threshold that most robotics companies won't reach for years, the point where the product outnumbers the producers. In a traditional manufacturing context, this would signal a "lights-out" factory run mostly by machines. Figure isn't there yet. Its 650 employees still design, engineer, maintain, and deploy the fleet. But the ratio tells you where the leverage is shifting. Each human worker at Figure is increasingly an operator and supervisor of multiple robots, not a peer on the same production line.

Adcock himself has been blunt about the trajectory. After Figure's "Man vs Machine" challenge earlier in the year, where a Figure 03 robot lost a package-sorting shift against a human intern, he posted on X: "This is the last time a human will ever win." Whether that prediction holds is almost beside the point. The ratio has already flipped. The question for the industry is no longer whether a company can build a robot workforce that outpaces its human one. Figure answered that. The question is what kind of human workforce you need to manage 740 robots, and whether 650 people is enough.

The Commercial Pipeline Behind the Hiring Surge

Figure AI's hiring spree didn't start in a vacuum. It started because the company has customers (real ones, with contracts) and a funding base that demands it deliver.

The most significant commercial anchor is the May 26, 2026 agreement with Catalyst Brands, the retail conglomerate that operates JC Penney and other labels. The deal puts Figure humanoids into Catalyst Brands' distribution and logistics network, starting at a distribution center in Reno, Nevada. This isn't a pilot or a memorandum of understanding. It's a commercial deployment agreement, the kind that requires Figure to actually ship working robots, integrate them into existing warehouse workflows, and keep them running. That commitment forces a company to hire security engineers, ML infrastructure people, and operations staff in bulk, because the bottleneck shifts from "can we build it" to "can we deploy it without breaking something or getting hacked."

The Catalyst Brands deal sits alongside Figure's longer-running BMW partnership. BMW began deploying Figure robots at its South Carolina plant in 2024, and by November 2025, Figure reported that its F.02 units had contributed to the production of 30,000 cars at BMW. That's a concrete output figure, not a demo or a press event, but cars that left a factory floor with robot labor in the loop.

Then there's the OpenAI relationship, which is more complicated than it looks. The two companies announced a partnership in February 2024 to pursue that same goal, and OpenAI was part of the Series B round. For a year, the Figure 02 used OpenAI models for natural language communication. But in February 2025, Figure walked away. Adcock told TechCrunch the integration didn't work, that embodied AI at scale requires vertical integration, and Figure couldn't outsource its AI any more than it could outsource its hardware. OpenAI, for its part, is also a major backer of 1X Technologies, a Norwegian humanoid robotics startup, and filed a trademark application for humanoid robots within days of Figure's announcement. The relationship is now a former partnership, but the capital and the initial technical collaboration helped fund the hiring surge that followed.

The money behind all of this is real. Investors include Microsoft, NVIDIA, Amazon, Intel, and OpenAI. That capital pays for the 98,000-square-foot San Jose headquarters and the 800-plus employees filling it.

The OpenAI partnership may be over, but its residue is visible in Figure's hiring: the company is building the full stack in-house, from Helix, its own vision-language-action model, to the security infrastructure around it. The Catalyst Brands deal is what turns that build into an operational obligation. Together, they explain why Figure is hiring faster than almost anyone in humanoid robotics, and why the roles that are filling fastest have less to do with building robots than with deploying them.

Where the Real Bottleneck Lives

Figure AI's careers page lists 12 open roles on its Helix AI team, covering perception, data infrastructure, training infrastructure, reinforcement learning, and generative AI. Serious ML work. But scroll past those and a different pattern emerges: three security and privacy roles, a safety systems architect, a product compliance engineer, and a software engineer for privacy and data governance. None of them are building actuators or writing control algorithms. All of them are building trust.

That split tells you where Figure actually thinks the deployment risk lives. The company's San Jose headquarters holds roles for application security engineers, vulnerability management automation, and privacy governance, positions that map to the attack surface a physical robot creates, not the robotics problems that dominated the last decade of humanoid research. A robot that moves through a warehouse is a networked computer with the ability to break things and hurt people. Figure is hiring accordingly.

The safety team is small but specific. A safety systems architect sets the framework for how the robot behaves when something goes wrong. A product compliance engineer handles the regulatory side (the standards bodies, the certifications, the paperwork that lets a machine operate legally near humans). These aren't roles you staff when you re still figuring out whether the hardware works. They're roles you staff when you re preparing to put robots in front of customers.

Then there's the simulation and data infrastructure layer. Figure lists multiple roles for AI training infrastructure, video pretraining, and backend infrastructure, the pipeline that feeds the Helix model. The company also needs data creators, data quality analysts, and a data strategy associate. This is the unglamorous plumbing: labeling, cleaning, and structuring the data that teaches a robot how to move. Without it, the ML models are academic. With it, they're deployable.

This pattern isn't unique to Figure. The broader robotics sector is converging on the same hiring profile as enterprise AI companies: security, data infrastructure, reliability, and developer-facing roles. The hardware is the differentiator, but the bottleneck is everything around it.

The takeaway is straightforward: Figure AI isn't hiring its way out of a hardware problem. It's hiring its way out of a trust problem. The robots exist. The models are training. What's missing is the infrastructure that makes a customer comfortable putting a two-legged machine in their facility, the security posture, the safety case, the compliance framework, the developer tools, the deployment playbook. Every security engineer Figure hires is a bet that the real bottleneck in physical AI isn't making the robot walk. It's making the robot safe enough, secure enough, and easy enough to deploy that a warehouse manager signs the contract.

The Broader Signal

A September 2025 technical report by Víctor Mayoral-Vilches of Alias Robotics, published on arXiv, performed the most thorough public security assessment of a production humanoid platform to date: the Unitree G1. The findings read like an enterprise security audit, not a robotics teardown. Mayoral-Vilches documented persistent telemetry connections transmitting audio, visual, spatial, and actuator data to external servers without user consent or notification. He reverse-engineered a dual-layer proprietary encryption system, identified static cryptographic keys that enable offline configuration decryption, and demonstrated how a compromised humanoid could escalate from passive data collection to active counter-offensive operations against its own manufacturer's cloud infrastructure. The robot, in other words, is a networked enterprise endpoint with legs, and it has the same attack surface as any other, plus a physical one.

The G1 runs ROS 2 Foxy with CycloneDDS 0.10.2, software that reached end of life in May 2023, missing three or more version releases. It maintains continuous MQTT connections to cloud servers. It uses a Rockchip RK3588 SoC with documented kernel vulnerabilities. Its master service orchestrates 22 child services through a hierarchical process tree that, if compromised, gives an attacker system-wide control. Every one of these is a standard enterprise security concern. The difference is that a compromised server rack doesn't pick up objects and walk down a hallway.

That physical dimension is what makes the security and infrastructure challenge qualitatively different from deploying a large language model or a traditional SaaS platform. A humanoid robot is a mobile sensor platform with actuators, network connectivity, and autonomous decision-making, a combination that creates what Mayoral-Vilches calls "physical-cyber convergence," where a security failure can translate directly into a safety incident. The arXiv paper documents how the G1's multi-modal sensor suite (six RealSense cameras, dual microphones, LiDAR, IMU, GPS) feeds 40-plus active data streams that could be aggregated and exfiltrated. The robot doesn't just process data. It perceives spaces, maps environments, and captures audio and video in any facility where it operates.

Bain & Company noted in its 2025 technology report that actuators, joints, and battery systems are largely solved problems at the pilot scale, and that intelligence and perception are approaching human-level performance. McKinsey made a similar observation in its analysis of humanoid robots crossing from concept to commercial reality, arguing that tech providers must focus on building four essential bridges to reach scale. The bridges are not mechanical. They are trust, safety, integration, and developer adoption, the same categories that dominate enterprise AI deployment, except now the system can walk through your warehouse door and plug into your network.

The IEEE Spectrum humanoid robot overview tracks more than 20 companies developing commercial humanoid platforms, from Tesla's Optimus to Apptronik's Apollo to Agility Robotics' Digit. Every one of them is building a system that must eventually operate in human spaces, on human networks, under human safety constraints. The companies that solve the infrastructure problem, covering security, MLOps, fleet management, and developer tooling, will be the ones that deploy at scale. The ones that treat it as an afterthought will spend their pilots explaining data breaches instead of demonstrating productivity gains.

OpenAI added 39 roles in the past week, many in research engineering and business operations that support the integration stack. Zero G Talent tracks 1,046 open robotics roles across 263 companies. The talent market is voting with its applications: the next phase of humanoid robotics is an infrastructure game. The companies that recognize this now, that hire for security and operations with the same urgency they hire for actuator design, will be the ones whose robots actually make it past the pilot phase. The rest will keep demoing.

Working in robotics? Zero G Talent tracks the openings: browse robotics jobs, openings at OpenAI and Figure AI, and the people building the field.