3,000 -> 50,000 humanoid robots in two years: AI moves into the real world

AI is learning to act in the real world. The breakthrough behind chatbots — large “foundation models” — is now going into machines that move, so robots can understand and adapt to their surroundings. Nvidia's Jensen Huang calls it the “big bang of physical AI” and is wiring up the ecosystem fast — from robot brain models (IsaacGR00T) to the recent deals across South Korea (Hyundai, Samsung, LG) for robots and AI factories.

Self-driving cars — the first physical AI at scale. Robotaxis are the most mature form: Google's Waymo gives 500,000+ paid rides a week, and in 2026 the tech went global — Uber is rolling out robotaxis in London, Munich and Madrid. Many run on Nvidia's ‘Drive Hyperion’ platform and its ‘Alpamayo’ driving models, with BYD, Geely, Nissan and Foxconn signed on.

Humanoids — leaving the lab for the factory floor. 2026 is the year they start real work: this June, Figure hit roughly one robot an hour, Boston Dynamics began shipping its electric “Atlas” to Hyundai, and Tesla's Optimus Gen 3 is slated to start production this year. Nearly every big car maker now backs a robot maker (Hyundai Boston Dynamics, BMW – Figure, Toyota – Agility).

Beyond cars and humanoids — the quiet workhorses. The least-hyped uses are the most proven: surgical robots (Intuitive Surgical's da Vinci) already assist many operations; AI-run warehouses are spreading at Amazon; and Samsung has pledged all-AI factories by 2030. Physical AI is also reaching farming, defence and logistics.

Ionic View

The data center opportunity, vast as it is, is fundamentally about AI that lives in servers. Physical AI is about AI that lives in the real world, through vessels such as robots, autonomous vehicles, industrial machines, and logistics systems. We see this as the next frontier of the AI revolution. The key for investors: it's a stack, not a single robot – value spreads across the chips and “brains”, the sensors, gears, motors, power and safety software, plus applications already earning real money today. With many established, profitable firms not yet re-rated for physical AI, this could prove to be an early-innings, multi-year shift whose impact may be felt across the value chain.