Google DeepMind Leases a 35,000-Square-Foot Robotics Lab in Watertown, Putting Robot Bodies Next to Gemini Robotics

Google DeepMind is building its first dedicated robotics hardware lab in the Boston area, according to a recruiting post from Aaron Saunders, the former Boston Dynamics chief technology officer who joined Google DeepMind in late 2025. "We're creating a first-class lab to house this team and take robotics to the next level," Saunders wrote on LinkedIn, kicking off hiring for what he described as a team under development since his arrival.

Reporting by The Boston Globe filled in the physical details. The lab will occupy 35,000 square feet inside the Arsenal on the Charles complex at 200 Talcott Ave in Watertown, Massachusetts — a building owned by Alexandria Real Estate Equities. Google signed the lease earlier this summer, according to a city permit, and construction is slated for completion in 2027. Saunders referred to the site as being in the "Cambridge area," but it sits officially in Watertown, a short drive from MIT and from Google's Kendall Square office, which employs roughly 2,000 people.

The hiring list is the most telling detail. DeepMind is recruiting machinists alongside engineers and researchers — the composition of a team that builds and calibrates physical robots, not a research-only annex. The Globe reported the facility will support development and testing as DeepMind shifts focus from AI models toward robot hardware, including humanoid systems that can walk, talk and handle delicate tasks.

The move complicates the strategy DeepMind has pursued so far. Under what has been an asset-light model, Google built the intelligence — the Gemini Robotics model line — and licensed it to hardware partners including Boston Dynamics, Apptronik and Agile Robots, while Alphabet folded its robotics moonshot Intrinsic into Google earlier this year. Saunders spent more than two decades at Boston Dynamics before joining DeepMind in November 2025, a hire that was widely read at the time as the centerpiece of Demis Hassabis's "Android for robots" ambition.

Owning hardware changes the data loop. Models tuned on telemetry from robots you control improve faster than models trained on partner data you only partially see, and every frontier lab has been converging on the same conclusion: OpenAI has backed humanoid maker Figure, Tesla builds Optimus bodies in-house, and NVIDIA sells the underlying compute stack to everyone. Until now, DeepMind was the notable holdout shipping only the brains.

The lab also plants Google's flag in a rapidly crowding corner of Massachusetts. Watertown and nearby Cambridge now host robotics startup Tutor Intelligence, and Walden Robotics, Anthropic and Blitzy have all expanded their footprints in the area recently, competing for the same talent pool that Boston Dynamics has drawn on for decades.

Google has not disclosed the cost of the facility, an exact opening date, or whether the lab will serve the existing Boston Dynamics partnership or operate independently of it. But the message of a machine shop two miles from MIT staffed with machinists is hard to misread: the company that makes the world's most capable AI models wants to control what those models run on, down to the torque of the grippers.