Figure AI raises $675M from Bezos, Nvidia, and OpenAI

For the past year, Big Tech poured billions into making AI think and talk behind glass screens. Now, the industry's heaviest hitters are teaming up to give that intelligence a pair of hands and feet.

Humanoid robotics startup Figure AI announced it raised $675 million in a Series B round, catapulting its valuation to $2.6 billion. The investor roster reads like a who's who of tech royalty: Jeff Bezos, Nvidia, Microsoft, and the OpenAI Startup Fund, alongside the Amazon Industrial Innovation Fund, Intel Capital, Parkway Venture Capital, Align Ventures, and ARK Invest.

Per reports by Bloomberg, Bezos committed $100 million through Explore Investments, Microsoft put in $95 million, Nvidia and an Amazon affiliate invested $50 million each, and OpenAI added $5 million after deciding against an outright acquisition.

The Silicon Valley super-team gives robots a mind

The funding is big, but the operational alliance is bigger. Alongside the capital, Figure signed a commercial agreement with OpenAI to jointly develop next-generation AI models specifically tailored for humanoid robots. The goal is straightforward: pair OpenAI's frontier multimodal models with Figure's robotics hardware and software to give machines semantic understanding and natural language reasoning.

To back the effort, Figure will rely entirely on Microsoft Azure to handle the massive compute and cloud infrastructure required to train and deploy these models.

OpenAI famously dissolved its in-house robotics hardware arm back in 2021 to focus entirely on scaling large language models. This deal marks an official comeback. Peter Welinder, OpenAI VP of Product and Partnerships, noted that the company always planned to return to robotics, and Figure offered the right vehicle to see what frontier multimodal models can accomplish in the physical realm.

Figure was founded in 2022 by Brett Adcock, previously a co-founder of electric aircraft maker Archer Aviation. He quickly built a veteran team pulling from Boston Dynamics, Tesla, Google DeepMind, and Archer. Earlier this week, Figure released footage of its prototype, Figure 01, walking while tethered in a warehouse, picking up a plastic crate, and setting it neatly onto a conveyor belt.

Built for a human world

Rather than building single-task machines on tracks or heavy industrial rigs, Figure is chasing a strict 1:1 human form factor. The Figure 01 prototype is an all-aluminum, electrically actuated bipedal machine designed to stand 5-foot-6, weigh 130 pounds, and carry up to 44 pounds.

Contrast that with Boston Dynamics' Atlas, which leans heavily into bulky, function-over-form hydraulics, or Agility Robotics' Digit, which uses backward-bending legs optimized for warehouse shelves. Figure's bet is that because the modern world—from door frames to factory floors—was designed exclusively for human bodies, creating a robot that mirrors human proportions eliminates the astronomical cost of retrofitting existing facilities.

Commercial players are already betting on that premise. In January, Figure signed a deployment agreement with luxury carmaker BMW to trial humanoid robots on the assembly line at its South Carolina plant, targeting repetitive and dangerous tasks.

Why it matters

We think this deal represents a turning point where generative AI stops being a desktop assistant and starts becoming embodied labor.

For enterprise operators in manufacturing and logistics, the value proposition is huge: instead of programming hardcoded robotics for single tasks, you get a general-purpose worker that understands spoken English, can parse visual context, and works inside existing human workstations. Adcock told Axios he expects humanoid robots to be active in the real world within 24 months. That said, everyday developers and small businesses won't touch this anytime soon; Figure 01 is still tethered and undergoing factory trials with Fortune 500 partners.

For the broader tech landscape, Figure just set a new bar for how hardware startups survive against giants like Tesla and Amazon. Trying to build custom foundational models while simultaneously manufacturing precision hardware is a suicide mission. Figure's genius move was outsourcing frontier intelligence to OpenAI, compute to Microsoft, and silicon to Nvidia, effectively turning the fiercest cloud rivals into co-owners of its platform.

Still, massive engineering hurdles remain. Fitting high-torque electric actuators, compute, and sufficient battery life into a 130-pound frame is an immense thermodynamic challenge, and walking on a safety tether is a long way from dodging workers on a chaotic shop floor. Big Tech wrote these massive checks because nobody wants to miss the physical embodiment of AGI, but turning warehouse demos into reliable line workers will be Figure's real test over the next two years.