Google, riding high on buzz surrounding its latest AI model, is pushing hard into robotics as it hopes to leapfrog Big Tech rivals like OpenAI, Meta, and Amazon in the shift toward embodied artificial intelligence.
The tech giant shattered the perception that it’s falling behind in the race toward artificial general intelligence (AGI) with its new Gemini 3.0 model. According to The Verge, early evaluations show the multimodal AI model topping leaderboards with unusually strong reasoning and greater accuracy interpreting text, images, audio, and video.
DeepMind, Google’s cutting-edge AI research arm, is framing the Gemini 3.0 model as the core intelligence layer for all things in the digital and physical realms. Its architecture has been extended into a suite called Gemini Robotics, which is designed to run on machines of any shape or size so they can perceive their surroundings, reason through what they see, and then act without extra training. They’re pitching it as the missing link between digital and embodied intelligence.
Google is making its biggest play into robotics since its high-profile 2013 shopping spree, when it acquired Boston Dynamics, Schaft, Redwood Robotics, Meka Robotics, Industrial Perception, and several other leading firms in an ambitious, but ultimately short-lived, bid to define the future of automation. Only time will tell whether the newfound momentum will last but its latest investments are unprecedented in scale.
$600M Bet on Physical Intelligence’s Universal Robot Brain
According to Bloomberg, Alphabet led a recent $600 million Series B investment into Physical Intelligence, a San Francisco-based startup already valued at $5.6 billion without a product. CapitalG, Alphabet’s independent growth fund, was joined by other high-profile backers like Jeff Bezos, Lux Capital, Index Ventures, and T. Rowe Price.
Launched in 2024, Physical Intelligence is developing foundational models for robots across form factors to perform complex, contact-rich tasks without custom programming. Its software has early success automating chores long considered too unstructured or unpredictable for modern robots, including folding laundry, box assembly, kitchen tasks, and industrial manipulation.
News of the investment broke as Physical Intelligence shared what it called a breakthrough that empowers robots to learn on the job. The startup says its new π*0.6 vision-language-action (VLA) model uses a reinforcement learning method called RECAP that incorporates autonomous attempts, human correction, and reward feedback.
According to its technical report, the system more than doubled throughput and cut failure rates for some tasks. The team shared footage of a dual-armed robot using the π*0.6 model to make espresso for 13 hours straight. They say the next big goal is to move toward more autonomous behavior without human resets and interventions. They also want the AI to earn in real-time instead of in batches and to explore new ways of solving tasks more effectively on its own.
Former Boston Dynamics CTO Joins DeepMind
Google DeepMind has recruited Aaron Saunders, former chief technology officer of Boston Dynamics, as vice president of hardware engineering. In a LinkedIn post, Saunders said he’s tasked with getting “robots from research labs out into the large, beautiful, and endlessly complex world around us.”
“This is a pivotal moment for AI and robotics, and I’m honored to be joining what I believe is the best AI team on the planet,” he wrote.
Saunders spent more than 20 years at Boston Dynamics, starting as a robotics engineer in 2003. The Massachusetts robotics leader promoted him to VP of hardware engineering in December 2013, VP of engineering in February 2018, and CTO in June 2021. He left the Waltham, Mass., based company in July 2025.
During his tenure, Saunders oversaw the development of the company’s entire portfolio, including its world-famous Atlas humanoid and Spot quadruped robots. His worked spanned mechanical design, actuation systems, sensing, balance, and locomotion algorithms, and the integration of AI-driven perception and dynamic motion control. Under his leadership, BD transitioned several prototypes into commercial products, including Spot for inspection and security and Stretch for warehouse box-handling and logistics automation.
He was also there through the company’s turbulent ownership changes. He led hardware engineering during Google’s 2013 acquisition and the years that followed when BD struggled to fit inside Google’s long-term plans and when Alphabet gave up and sold it to SoftBank in 2017. The company is now owned by South Korean automaker Hyundai, which has committed to deploying tens of thousands of its robots, including Atlas and Spot, into its manufacturing operations.
Announcing his departure on LinkedIn over the summer, Saunders said he was taking a pause to spend time with his family and to figure out his next steps. He was apparently in talks to join DeepMind in the weeks that followed. He began hyping DeepMind’s foundational AI models for robotics in September.
“We’re moving from machines that just react to commands towards intelligent agents that can reason, plan, and generalize to solve complex problems in our world,” he wrote.
Apptronik’s Humanoids
Google’s relationship with Austin, TX-based robotics leader Apptronik has become one of the most strategically important ties in the humanoid robotics sector.
The tech giant played a major role in driving Apptronik’s funding surge in early 2025, participating in its oversubscribed Series A round, which was initially $350 million and later expanded to $403 million. DeepMind is actively collaborating with Apptronik to integrate and test its Gemini-powered robotics models on its Apollo humanoid platform.
Heading into 2026, Apptronik is reportedly looking to raise an additional $400 million for a pre-money valuation around $5 billion. Apptronik, which was spun out of the University of Texas at Austin’s Human Centered Robotics Lab in 2016, is piloting its humanoids for at Mercedes-Benz and GXO Logistics.
Rivals Bring the Heat
Right now, Google is getting showered with attention but hype in the tech world is notoriously short-lived. Big breakthroughs come and go every week, and the spotlight constantly shifts. One week it’s Google, then the next week it’s OpenAI, and then Meta or Amazon grab headlines. So, there is by no means a clear frontrunner heading into 2026.
OpenAI and Meta are both moving aggressively into hardware, each framing physical agents as the next phase of their roadmaps. OpenAI has spent much of 2025 reviving its robotics ambitions after several quiet years, hiring heavily in humanoid development and positioning its large-scale models as control systems for next-gen machines. Meta, meanwhile, has launched a new robotics-focused team within Reality Labs to build both the hardware and software for humanoids.
Amazon is also on a robotics tear. It boasts a fleet that has grown from 750,000 to more than one million units in 2025 alone, its largest single-year increase ever. Robots now handle a massive share of tasks including sorting, lifting, packaging, and moving goods inside fulfillment centers. The company is training humanoid robots for last-mile automation as it reportedly aims to replace more than 600,000 human roles with robots by 2033.
Subscribe here to keep up as the race evolves through 2026 and beyond.