Google Sold the World's Most Famous Robot Maker. Now It Wants to Be the Operating System Inside Every Robot.

Google bought Boston Dynamics in 2013. It had no idea what to do with it. Four years later, it sold the company to SoftBank and walked away from robotics entirely.

Last week, it walked back in.

On February 25, Google announced it's absorbing Intrinsic — its robotics software unit — into the main company. Not a side bet. Not a moonshot. Core Google. The pitch: Intrinsic will be the Android of robotics.

That's not a metaphor the company is being coy about. It's the explicit strategy.

The Android playbook, explained with arms

Android didn't build phones. It built the software layer that let Samsung, Motorola, and Xiaomi build phones without reinventing the wheel. It went from 4% market share in 2009 to 75% by 2019. Google didn't need to manufacture a single handset to dominate mobile.

Intrinsic is running the same play. It builds the operating system that sits between AI models and physical robots — so manufacturers "can focus more on solving the problem, and not the plumbing," as the company puts it. Its partners include FANUC, Universal Robots, and KUKA. If you've never heard of them, that's fine. They make the robots that build your car, sort your packages, and weld your airplane parts.

The idea: you shouldn't need a PhD in robotics to program a robot. Just like you didn't need to understand Linux kernel architecture to build an Android app.

"We're trying to make it accessible for anyone," CEO Wendy Tan White told CNBC. "It doesn't matter what the hardware is and it doesn't matter what the AI model is."

Why this time is different

Google's first robotics chapter was a mess. Between 2013 and 2014, Alphabet bought eight robotics companies in six months under Andy Rubin's "Replicant" division. Boston Dynamics was the crown jewel — walking robots, viral YouTube videos, Pentagon contracts.

The problem: Google had no business model for any of it. When Rubin left in 2014, the whole thing drifted. A 2016 Atlas video showing a humanoid stacking boxes went viral for all the wrong reasons — people found it terrifying, not exciting. Internal teams couldn't agree on what the robots were for. Google sold Boston Dynamics and Schaft to SoftBank in 2017 for a reported $165 million. The lesson was clear: owning the coolest robots means nothing if you can't ship a product.

The AI explosion changed the equation.

In mid-2025, Google DeepMind launched Gemini Robotics and Gemini Robotics-ER — AI models that translate language commands into physical actions. Last month, Google partnered with Boston Dynamics (yes, the company it sold) to integrate Gemini into Atlas humanoid robots for manufacturing. It also hired Boston Dynamics' former CTO.

The circle closed. But this time, Google isn't buying hardware companies. It's building the brain.

The $370 billion question

McKinsey projects the general purpose robot market could hit $370 billion by 2040. That number explains why every major tech company is suddenly interested in robots that do things in the real world.

Tesla's Optimus humanoid is targeting factory tasks and eventually home use, with Musk claiming a sub-$20,000 price point. Amazon has deployed over 750,000 robots across its warehouses. Figure AI raised $675 million for humanoid robots that work alongside humans. Apptronik, a Texas startup, is building Apollo — a humanoid designed for manufacturing — with Google as a partner.

At MWC Barcelona this week, even Honor — a smartphone company — showed off a humanoid robot doing backflips. When phone makers start building humanoid robots, you know the market is heating up.

But here's what makes Google's play different from everyone else's: it doesn't need to win the hardware race.

Software eats robots

Every humanoid robot company faces the same bottleneck. Building a body that walks is hard. Building a brain that knows what to do with that body is harder. Programming robots today still requires specialized engineers writing task-specific code. That doesn't scale.

Intrinsic's bet: the winners in robotics won't be the companies that build the best robot. They'll be the companies that build the best platform for all robots.

This is exactly what happened with smartphones. Nokia and BlackBerry built great hardware. Android ate them alive because it gave every manufacturer access to the same software ecosystem — maps, apps, AI, cloud. The hardware became interchangeable. The platform became indispensable.

By sitting inside Google, Intrinsic now has direct access to DeepMind's AI research, Gemini models, and Google Cloud infrastructure. That's the kind of stack no standalone robotics company can match.

Brian Gerkey, Intrinsic's technology chief, put it bluntly: they're "standing on the shoulders" of DeepMind's models and adding specialized robotics data on top.

The competitors who disagree

Not everyone thinks the platform play wins.

Tesla's approach is vertical integration — build the robot, the AI, the manufacturing line, the training data pipeline. If Optimus works, Tesla doesn't need a third-party operating system. It is the operating system.

Amazon built its warehouse robotics internally for the same reason. When you control the entire stack, you move faster and keep the margins.

Then there's the open-source world. ROS (Robot Operating System), maintained by Open Robotics, has been the default standard in academic and industrial robotics for over a decade. FANUC, KUKA, and others already use ROS-based systems. Google's Intrinsic needs to be dramatically better, not just marginally different.

And China is building its own robotics ecosystem entirely. Shanghai just announced a humanoid robot training facility and autonomous driving test base, part of a national push to lead in what Beijing calls "embodied AI." Chinese manufacturers may not want to build on a Google platform any more than Huawei wants to depend on Android.

What this actually means

Forget the humanoid robot demos. The real story is quieter and more consequential.

Today, roughly 80% of manufacturing tasks that could be automated aren't — because programming robots is too expensive and too specialized. Small and medium manufacturers can't afford it. If Intrinsic makes robot programming as accessible as building a mobile app, the number of robots deployed globally could multiply by an order of magnitude.

That's not a robot revolution in the sci-fi sense. No Terminator. No robot butler. Just millions of factory floors, warehouses, and logistics centers running on Google's software layer.

Google failed at robotics when it tried to own the body. Now it wants to own the brain. If the Android playbook holds, it won't matter who builds the best humanoid. It'll matter who builds the best platform underneath it.

The last time Google made that bet, it captured three-quarters of the global market.