Announcing our Investment in Generalist

Generalist is a frontier lab for robot intelligence that trains robot foundation models which generalize across thousands of form factors, environments, and tasks. Today the company announced $400 million in new funding, and we are thrilled to join their journey.

Historically, robots have been narrow specialists: hand-programmed for a single repetitive job and brittle to changes in environment or hardware. This fragility has confined the discipline to a set of fixed and tidy settings and tasks. Some of these use cases in critical industries have supported large businesses, yet the vast majority of useful tasks demand greater generality -- one we believe is finally imminent.

Robotics now faces the same fundamental promise and conceit behind the success of large language models: that a general-purpose model can consistently and quite dramatically outperform a task-specific system on its own benchmarks. In language, the truth of this belief produced one of the largest compounding advantages in software history. Each incremental increase in data and compute improved performance on all tasks simultaneously. If the same dynamics hold for robotics, the companies that build and scale the best models first will capture an outsized share of physical work, creating data and compute moats that competitors cannot replicate. We believe Generalist is earliest on that curve and best positioned to lead.

The crux of the robot intelligence problem lies in dexterity. Progress at the cognitive frontier was limited by the scarcity of intelligent experts: there are only so many mathematicians or doctors, for instance, who can push the edge of model intelligence forward across different domains. Physical dexterity poses the opposite problem. Dexterity is not a skill that sits with a trained few. While there are constraints around existing pre-training data, the challenge has never been about the rarity of the skill itself but rather finding the right technical approach to actually scale. We're convinced that Generalist has found it. Planning and navigation have already yielded to scale (e.g., with autonomous driving), improving with more time in the market and real-world examples. Dexterity has been different to date; it remains the hardest unsolved problem in robotics. It is also where a working solution converts most directly into commercial value. Just as coding became the sharpest early wedge for language models, dexterity is the wedge for robotics -- the hardest thing to get right, and the clearest path to everything that follows.

But the deepest reason for our conviction is rarer than any single technical bet: the team's blend of research excellence and commercial pragmatism. These two qualities rarely coexist. Most frontier research teams are brilliant but commercially adrift, and most pragmatic robotics companies lack the depth to build something fundamentally new. Generalist is both at once: a founding team among the most-published researchers in the field, who are relentlessly focused on what actually ships, sells, and compounds. Pete Florence, CEO, was previously a senior scientist on DeepMind's robotics team and a senior author on PaLM-E and RT-2, two of the most influential papers in embodied AI. He is joined by Chief Scientist Andy Zeng (ex-DeepMind, lead author on Code as Policies) and CTO Andrew Barry (previously a senior roboticist at Boston Dynamics), with an early team drawn from the top of DeepMind, OpenAI, and Boston Dynamics. Among the researchers we trust most across the leading AI labs, one team comes up again and again as the most promising in the space: Generalist. Pete in particular is as much a builder as a researcher -- magnetic and unusually commercial for someone of his research caliber. We wanted to back him from the first conversation.

We followed Generalist's progress closely across successive generations of its models, and the signal was unmistakable: rapid adaptation to new robots and tasks, strong sample efficiency, and remarkable early commercial traction.

If foundation models reshape physical work the way they reshaped software, Generalist will be one of the defining companies of that transition. We're proud to be on the journey with Pete and the team.