Google is entering the robotics race with an approach that differs from Tesla and other companies: rather than rushing to build flagship hardware, it is prioritizing the development of models and software capabilities behind robots, and expanding its reach through partnerships.

According to a recent report by The Information, Alphabet (GOOGL) DeepMind CEO Koray Kavukcuoglu publicly discussed the robotics strategy in relative detail for the first time last week. He stated that Google's core advantage lies in models, not the robot chassis itself. Meanwhile, Google has launched Gemini Robotics — a version of the Gemini model optimized for physical control scenarios — and has already partnered with robotics companies including Boston Dynamics.

This model is quite similar to how Google promoted Android years ago: Google provides the core software, and scale is achieved through hardware from various manufacturers, rather than relying on proprietary devices from the start. This also clearly distinguishes Google from Tesla and OpenAI-backed Figure — both of which have made building flagship humanoid robots a key priority. For Google, robotics is more like a new application scenario for extending Gemini's capabilities into the real world.

Software First, Replicating Android's Expansion Path

Speaking at The Information's AI Agenda Live event, Kavukcuoglu said Google hopes to "work with other robotics companies and have them use this system, so that intelligence can be infused into all robots in a safe and sound way."

This strategy mirrors the path Google took when promoting Android. The report noted that between 2007 and 2008, Google co-founder Sergey Brin also responded with a similar approach when asked whether Google would build its own phone. At the time, Google was working with T-Mobile and HTC to launch Android devices, and only later gradually developed the Pixel hardware business.

Currently, Kavukcuoglu has not completely ruled out the possibility of Google building humanoid robots in the future, but emphasized that the focus at this stage remains on model capabilities and the partnership ecosystem.

Gemini Moves Into the Physical World, Model Competition Heats Up

Google's robotics push is also an extension of its AGI strategy. Kavukcuoglu views Gemini Robotics as a natural extension of Gemini's multimodal architecture into the field of physical control: the same underlying model capable of processing text, vision, and audio is being further applied to robot control, rather than developing specialized models for different tasks separately.

He sees Waymo's autonomous driving as an important validation case for this approach, calling autonomous driving a "very, very safe way" to demonstrate the large-scale deployment of physical AI. If the Gemini architecture can support Waymo in achieving commercial operations, then the feasibility of migrating it to other physical scenarios will gain further validation.

Meanwhile, Google announced this week that it would release Gemini 4 Argon to select enterprise customers first, focusing on cybersecurity defense, with plans to open it to a broader user base soon. As the new model enters the application stage, whether Google can further narrow the capability gap with Anthropic and OpenAI will also become an important window for the market to observe its AI competitiveness.