Analysis
Google DeepMind is positioning its robotics models as a licensable platform for multiple robot makers, drawing an explicit comparison to Android's role in mobile, while Tesla builds its Optimus humanoid as a fully vertically integrated product the way Apple builds iPhones, according to an interview with DeepMind's Koray Kavukcuoglu published by The Information.
New Leadership, Early Signal
The strategy comes as Kavukcuoglu, DeepMind's senior vice president and chief AI architect, took over day-to-day leadership of the lab from co-founder Demis Hassabis in August 2026, reporting directly to Sundar Pichai. A robotics platform strategy surfacing this early under his leadership suggests it's a deliberate priority rather than a holdover initiative from the Hassabis era.
“A robotics platform strategy surfacing this early under his leadership suggests it's a deliberate priority rather than a holdover initiative from the Hassabis era.”
The Platform Bet vs. the Integrated Bet
Google's approach licenses its Gemini Robotics models to third-party hardware makers, betting that most robot builders would rather buy a proven AI brain than build their own -- the same logic that let Android out-ship Apple's iOS by volume by partnering broadly instead of controlling hardware. Tesla's Optimus takes the opposite path: the model, the actuators, the manufacturing and the deployment are all built in-house, mirroring Apple's tightly integrated hardware-software stack. Other robotics players are placing distinct bets on the same spectrum -- General Intuition, which trains world models on billions of gameplay clips for spatial reasoning, and Physical Intelligence are both closer to the platform end, supplying models rather than selling finished robots, while Boston Dynamics (owned by Hyundai) and Figure AI sit further toward integrated hardware plays.
What the Headline Misses
Android won mobile because carriers and phone makers needed a credible, free alternative to Apple's closed ecosystem at a moment when smartphones were already a mass-market product. Robotics isn't there yet: there's no equivalent carrier-like distribution chokepoint forcing hardware makers to standardize on one AI platform, and most humanoid and industrial robot builders today are still vertically prototyping their own control stacks rather than shopping for a licensed brain. Google's comparison is a useful framing for its own ambitions, but it assumes a platform dynamic that the robotics market hasn't yet proven it needs.
Neither Google nor The Information's report discloses which third-party hardware makers, if any, have committed to licensing Gemini Robotics at commercial scale -- the strategy is articulated well before it has visible proof points in the market.
The Money Is Already Choosing Sides
Capital flowing into physical AI this year leans platform-heavy: General Intuition nearly tripled its valuation to $6.2 billion in September on a world-model thesis that supplies reasoning to many downstream robots rather than building its own hardware, and a wave of similar model-layer startups have out-raised pure hardware plays in recent months. Tesla, by contrast, is self-funding Optimus development inside a public company with its own balance sheet, insulating the bet from needing outside robotics-specific capital at all. That divergence -- venture money chasing the platform layer while Tesla vertically integrates on its own dime -- is itself evidence the market hasn't settled which structure wins, and won't until a platform model actually ships inside someone else's commercial robot at scale.
