Robotics

Market Context

Robotics is experiencing one of its most concentrated capital moments on record, with $20.47B deployed across 36 deals in the past 28 days — a velocity score of 0.39 and rising, signaling accelerating institutional conviction. The convergence of foundation models, Vision-Language-Action (VLA) architectures, and humanoid hardware is collapsing the boundary between software AI and physical AI, drawing in hyperscalers, sovereign wealth, and cross-sector strategics simultaneously. Research from Google DeepMind is actively shaping product architectures across the industry, with systematic studies of hierarchical VLA design choices now directly influencing how companies like Figure and Galaxea build their robot AI stacks.

Investment Activity

• Pudu Robotics raised ~$150M at a $1.5B valuation, establishing a new service-robotics unicorn benchmark.
• Genesis AI emerged from stealth with a $105M seed round co-led by Eclipse and Khosla Ventures, one of the largest seed rounds in robotics history.
• A major robotics company raised a $1.4B Series C led by Tether Holdings at a $7B valuation, with participation from Qualcomm, Amazon, Nvidia, and Bosch.
• Maneva closed a $27M Series A with participation from Gokul Rajaram, targeting computer vision for factory floors.
• Human Archive raised $8.2M from Wing Venture Capital, NVP Capital, and Y Combinator to supply real-world physical training data to robotics and AI labs.
• Shifters raised a $10.2M seed round for battlefield robotics.

Key Players

• Genesis AI — Emerged from stealth with a $105M seed to build a foundational model for robots, co-led by Eclipse and Khosla, positioning itself as the "LLM moment" for physical AI.
• Physical Intelligence — Developer of the open-weight π0 and π0.5 VLA generalist policies, whose models underpin external research including WEAVER (which improved π0.5 real-world success rates by 38%) and Flow Reversal Steering experiments.
• Figure — Humanoid robotics company whose Helix hierarchical VLA system is cited in Google DeepMind research as using fixed-timer termination, flagged as the weakest termination strategy for long-horizon tasks, signaling a near-term architectural upgrade cycle.
• Pudu Robotics — Service robotics company that raised ~$150M at a $1.5B valuation, demonstrating that commercial deployment-stage robotics companies can now access unicorn-tier capital.
• Human Archive — Pays gig workers in India to capture multimodal physical training data, addressing the critical real-world data bottleneck for robotics AI training at a $8.2M seed stage.

Market Signals

• Google DeepMind published the first systematic head-to-head comparison of hierarchical VLA design choices (arXiv 2606.10267), using its own Gemini Robotics On-Device (GROD) 1B and 3B parameter models as low-level policies — signaling that hyperscaler-grade research is now directly competing with and enabling robotics startups.
• The $1.4B Series C at a $7B valuation backed by Tether, Nvidia, Amazon, Qualcomm, and Bosch represents the largest single robotics round in the dataset and one of the largest in sector history, with a uniquely cross-sector investor syndicate spanning crypto, cloud, semiconductors, and industrials.
• Chinese robotics remains a force: AgiBot, Unitree Robotics, Agile Robots, and Daimon Robotics are all active, with Agile Robots joining Nvidia's Cosmos 3 platform coalition.
• Top investors by deal count — 8VC, Bessemer Venture Partners, Felicis Ventures, and Andreessen Horowitz — each executing 4–5 robotics deals in 28 days, suggesting portfolio-level conviction rather than one-off bets.
• Academic-to-commercial pipelines are accelerating: Carnegie Mellon University researchers are co-authoring papers with direct commercial applicability to Apptronik, Figure, and 1X Technologies, while UC Berkeley's RAIL Lab (Sergey Levine) is producing the foundational RL infrastructure underlying multiple deployed robotics systems.
• Physical AI data infrastructure is emerging as a distinct investment category, with Human Archive and Protege AI both raising to supply the training data that robotics foundation models require.