The humanoid robotics sector has absorbed more venture and strategic capital in the past 18 months than in the preceding decade combined. By Clarqo’s count, at least $9.8 billion has flowed into humanoid robot startups since January 2025 — a figure that reflects both genuine technological progress and a Wall Street narrative that has, in some cases, run well ahead of operational reality.
The Investment Landscape
The headline numbers are striking. Figure AI closed a $675 million Series C in late 2025 at a $3.9 billion valuation, with strategic backing from BMW and a disclosed deployment agreement covering 10,000 units across BMW’s North American manufacturing network. Physical Intelligence (Pi), the San Francisco-based robotics foundation model company, raised $400 million in January 2026 to continue development of its general-purpose robot policy framework — software that aims to give physical robots the same kind of cross-task generalization that large language models brought to text.
Norwegian startup 1X Technologies closed a $150 million round in February, while Apptronik — which counts NASA among its early collaborators — secured $350 million in March to fund commercial deployment of its Apollo humanoid in warehouse and logistics settings. Tesla’s Optimus program, technically internal, represents tens of billions in implied investment based on the company’s disclosed R&D allocation.
Goldman Sachs estimates the global humanoid robot market will reach $38 billion by 2035, with unit deployments reaching 1.4 million. JPMorgan’s more conservative projection lands at $22 billion, citing manufacturing yield and software reliability concerns.
What’s Actually Deployed
Behind the funding headlines, the deployment picture is considerably more modest. Figure AI has shipped approximately 200 units to BMW facilities as of April 2026, with full-scale deployment of the contracted 10,000 units still subject to qualification testing. Agility Robotics’ Digit — arguably the most commercially mature humanoid on the market — has been deployed by Amazon in three fulfillment centers, performing bin transfer and tote-moving tasks. Amazon has not disclosed unit counts, but industry analysts estimate fewer than 500 units are in active operation.
The gap between demo-ready and production-ready is being driven primarily by software, not hardware. “The manipulation problem is largely solved for structured environments,” said one robotics engineer at a major automotive OEM who declined to be named. “The problem is edge cases. A robot that works 99% of the time is not good enough for a production line where it’s handling a $40,000 component.”
Physical Intelligence’s approach — training generalist robot policies on massive datasets of human demonstration rather than hard-coded task logic — is widely viewed as the most promising path to bridging that gap. Its RT-2 successor models have demonstrated meaningful transfer learning across novel objects and tasks in lab conditions. Whether that holds in the noise and variability of real industrial environments is the open question the industry is paying billions to answer.
The China Dimension
Any account of humanoid robotics in 2026 that ignores China is incomplete. Unitree Robotics, UBTECH, and a constellation of better-funded startups backed by Tencent, Baidu, and state-linked venture funds are producing humanoid platforms at price points that Western competitors struggle to match. Unitree’s G1 platform retails at approximately $16,000 — a fraction of the $80,000–$150,000 price range associated with US-made equivalents.
The US Commerce Department is actively reviewing whether humanoid robotics should be added to the Entity List framework or subject to export controls, citing dual-use concerns around the underlying AI models that power robot perception and manipulation. A formal rulemaking is expected by Q3 2026.
The Long Game
The fundamental bull case for humanoid robots rests on demographics: aging workforces in the US, Europe, Japan, and South Korea are creating structural labor shortages in manufacturing, logistics, and elder care that no conventional immigration or automation strategy can fully address. A general-purpose humanoid capable of operating in environments designed for humans — factories, warehouses, hospitals — offers a theoretically unconstrained labor pool.
The bear case is simpler: every previous wave of robotics investment has underestimated deployment complexity and overestimated timeline. The investors writing nine-figure checks today are betting that foundation model-driven AI has changed that equation. The next 24 months of real-world deployment data will determine who is right.
For now, the money is in, the ambitions are large, and the robots are learning. The rest is execution.
Discussion
Sign in to join the discussion.
No comments yet. Be the first to share your thoughts.