Google's AI research arm is planting a new flag in Cambridge: a dedicated robotics hardware lab built to turn its Gemini AI models into working machines that walk, grip, and move through the physical world. The project was announced on LinkedIn by a company executive, and it puts Google DeepMind squarely in the middle of a Greater Boston robotics cluster it once walked away from.
Google DeepMind, Alphabet's artificial intelligence research and development wing, is constructing the lab to host a new team developing hardware based on the company's AI and robotics research, according to The Boston Globe. That announcement came from Aaron Saunders, who joined Google DeepMind from Boston Dynamics, where he served as chief technology officer. Saunders hinted the lab could open in 2027, though he did not specify its location or size, the Globe reported. The company plans to hire machinists, engineers, and researchers to staff the new team.
Saunders' arrival at DeepMind is no small hire. He spent 22 years at Boston Dynamics before joining Google DeepMind's hardware team in 2025, and he is Google DeepMind's vice president of embodied systems and robotics, according to AI Insider. At Boston Dynamics, he oversaw platforms including Atlas and Spot starting when he took the CTO role in 2021. Boston Dynamics, the Waltham-based robotics maker known for backflipping and dog-like robots, spun off from the MIT Leg Lab founded by former MIT professor Marc Raibert, per the Globe's reporting.
Why Google Wants a Robot of Its Own
DeepMind CEO Demis Hassabis described the company's robotics strategy in late 2025 as an “Android play” — building Gemini into a universal AI operating system that can run across many different robot bodies out of the box, according to the AI Insider report. The comparison frames DeepMind's ambition as an equivalent software layer for physical machines.
That ambition has already produced hardware demonstrations. In July, Google DeepMind unveiled Gemini Robotics 2, a vision-language-action AI model suite offering what the company calls “feet to fingertips” whole-body control for humanoids, shown running on Apptronik's Apollo 2 platform. More recently, in October, the company introduced a Video-to-Sim-to-Real training pipeline powered by Gemini 4 Argon, which generates MuJoCo physics simulations from video footage to train humanoid whole-body tasks in hours rather than weeks — a technique aimed at closing the notoriously difficult sim-to-real gap in embodied AI.
Per the Globe, DeepMind develops the Gemini AI models and the Nano Banana image generation software, and the company is now investing in physical AI software meant to run on autonomous robots. It's working to integrate Gemini's capabilities into various kinds of robots, including, per the Globe, a humanoid robot that walks on two legs, talks, and performs basic delicate tasks.
A Return to a Robotics Hub Alphabet Once Left
Google's relationship with legged robots is not new, even if it is now re-forming. Alphabet originally acquired Boston Dynamics in late 2013, before selling it to SoftBank in 2017; South Korean automaker Hyundai Motor Company later acquired majority control in 2021, according to reporting cited in the AI Insider piece. While Alphabet previously owned and then divested Boston Dynamics, DeepMind's hiring of Saunders and its stated vision to build an “Android for robots” suggests Google intends to compete for the dominant software layer running generalist physical AI, rather than owning the hardware itself.
Building the lab in Cambridge puts DeepMind close to the people and workshops that make that kind of work possible. The Globe reports Google already has an office in Kendall Square near MIT's campus and employs about 2,000 local workers. The company first established its Cambridge presence in 2003 and had expanded its real estate footprint to more than 1 million square feet across Kendall Square by 2022, anchored by a 15-year lease for 362,000 square feet at 325 Main Street, according to Boston Properties.
Kendall Square's AI Land Rush
DeepMind isn't the only AI player circling the neighborhood. Walden Robotics signed a 50,000-square-foot office lease near Kendall Square this month, and Anthropic and Blitzy have both signed office leases at One Kendall Square, per the Globe's reporting. Google did not immediately respond to the Globe's request for comment on the new lab.
The influx tracks with a broader real estate shift in Cambridge. Greater Boston's commercial life-sciences lab vacancy reached 28.7% in the second quarter of 2026, helping open the door for AI firms to move into space once reserved for biotech, according to Bisnow. Anthropic's own 24,000-square-foot lease at One Kendall Square, signed in September, fits that same pattern of AI companies absorbing space left vacant by a post-pandemic slowdown in life-sciences venture funding.
DeepMind's move also lands the company near Waltham, where Boston Dynamics is based and where it moved forward in June with a $100 million campus consolidating production of its Spot, Atlas, and Stretch robots, built to host up to 1,250 local jobs and a robotics training center, as Hoodline previously reported. The region's robotics ecosystem runs deeper still: Boston-based nonprofit hub MassRobotics hosts more than 100 resident startups across 35 industries, and its resident and alumni companies have raised over $2 billion in cumulative venture funding, according to MassRobotics.
Open Questions as the Lab Takes Shape
Much about DeepMind's Cambridge plans remains unsettled. Saunders did not specify where in Cambridge the lab will be or how large it will be, and it is not yet clear whether Google intends to build its own proprietary robot hardware or continue relying on third-party manufacturing partners such as Apptronik. Massachusetts lawmakers have also filed “An Act to ensure the responsible use of advanced robotic technologies,” legislation that faced state hearings through 2025 and aims to regulate robotic manufacturing and ban weaponized autonomous systems statewide, a framework that could eventually bear on how companies test physical AI hardware in the state.
