Best AI Robotics & Humanoid Robot Companies 2026

Figure AI holds two distinctions no other company can claim: the highest valuation in humanoid robotics ($39 billion), and the first sustained commercial humanoid deployment in industrial history. Figure 02 robots spent 1,250 operational hours on BMW's Spartanburg, South Carolina production line — loading 90,000+ parts with greater than 99% placement accuracy per shift, meeting the plant's 84-second cycle time requirements, and contributing to more than 30,000 vehicles. BMW's response was to expand: Plant Leipzig, Germany begins a Figure 02 pilot in summer 2026, making it the first European commercial humanoid deployment.

Founded in 2022 by Brett Adcock — who previously built Vettery (sold to Adecco) and Archer Aviation — Figure raised over $1.9 billion from OpenAI, Microsoft, Bezos Expeditions, Intel Capital, and Samsung. The company's Figure 03 robot is in development for 2026 deployment. BotQ, its dedicated manufacturing facility, has capacity for 12,000 units per year scaling to 100,000 annually.

Founded in 1992 as an MIT spinout by Marc Raibert, Boston Dynamics has more commercial robot deployment experience than any other company in this guide — by a wide margin. Spot, the quadruped robot, has been commercially available since 2020 and is deployed at thousands of sites worldwide for industrial inspection at oil refineries, offshore platforms, construction sites, and nuclear plants. This deployment base generates real-world operation data at a scale competitors cannot match — a durable moat for training the next generation of AI robot foundation models.

The Atlas electric humanoid, unveiled in 2024, won CNET's Best of CES 2026 award. All 2026 Atlas units are committed to Hyundai's Robotics Metaplant Application Center (RMAC) and Google DeepMind — commercial sales open in 2027. A strategic partnership with Google DeepMind (formalised at CES 2026) will integrate Gemini Robotics foundation models as Atlas's AI brain. Hyundai acquired Boston Dynamics from SoftBank in 2021 for $1.1 billion; Korean securities firms now implicitly value the company at $20–28 billion, with optimistic IPO projections reaching $88–103 billion. A Nasdaq listing is expected in 2027.

Agility Robotics was founded in 2015 by Oregon State University roboticist Jonathan Hurst, whose research focus on legged locomotion biomechanics directly shaped Digit's unique design. Unlike humanoids built to mimic human appearance, Digit (5'9", up to 50 lb payload) is engineered for logistics environments — legs and joints optimised for traversing warehouse floors, squeezing between shelving racks, and performing repetitive pick-and-place tasks at commercial speed and cost.

The Amazon deployment is Agility's proof of concept at scale. After 18 months of testing at Amazon's Sumner, Washington fulfilment centre, Digit robots achieve a 98% task success rate on tote recycling — collecting and returning empty totes from picking stations. Operating cost runs $10–12 per hour versus $30 per hour for equivalent human labour: a 60–70% cost reduction before accounting for 24/7 operation capability. Agility raised $641 million total, including a $400 million Series C in 2025 at a $2.1 billion valuation backed by Amazon's ICAD fund and DCVC. RoboFab, Agility's Salem, Oregon manufacturing facility (opened 2024), is the first high-volume humanoid factory in the US, with 10,000 Digit units per year capacity.

Apptronik's defining competitive advantage is its AI partnership with Google DeepMind. Apollo, its humanoid robot (73 kg, 25 kg payload), runs Gemini 3 and Gemini Robotics AI — the same foundation models powering Google's own robotics research. This means Apollo can perform diverse real-world tasks without environment-specific retraining: show it a new assembly task and it generalises from prior training rather than requiring weeks of task-specific demonstrations. This is meaningful differentiation from approaches that need extensive per-task data collection.

A University of Texas spinout founded in 2016, Apptronik raised $935 million in February 2026 at a $5 billion valuation from Google, Mercedes-Benz, B Capital, PEAK6, AT&T Ventures, John Deere, and Qatar Investment Authority. Apollo is commercially deployed at Mercedes-Benz factories (delivering assembly kits to production workers), GXO Logistics, and Jabil. Hardware design is modular — individual joints can be replaced without full robot replacement, which materially reduces maintenance cost and downtime versus competitors.

1X Technologies occupies a unique position: the most affordable commercially available humanoid robot at $20,000 (or $499/month subscription). The Norwegian-American company, formerly Halodi Robotics, raised over $130 million including a $23.5 million Series A2 led by the OpenAI Startup Fund and a $100 million Series B led by EQT Ventures. Founded in 2014, 1X is one of the oldest companies in this group — a decade of robotics development gives it deep mechanical engineering expertise despite smaller fundraising than American peers.

NEO features force-limited actuators and a body inspired by human biomechanics, designed for safe interaction in proximity to humans without mandatory safety barriers. The Hayward, California factory targets 10,000 NEO units per year initially, scaling to 100,000 by end of 2027. A landmark deal with EQT's 300+ portfolio companies will deploy NEO across manufacturing, logistics, warehousing, facility operations, and healthcare between 2026–2030. The OpenAI connection positions NEO to benefit from frontier language and vision AI for embodied cognition — NEO is designed to follow natural language instructions without task-specific programming.

Physical Intelligence does not build robots. It builds the AI brain that runs in robots made by other manufacturers — making it the only company in this guide operating at the foundation model layer rather than the hardware layer. Founded in 2023 by former Google Brain researchers Sergey Levine, Chelsea Finn, and Karol Hausman, the company raised $1 billion+ at a $5.6 billion valuation in a Series B led by Alphabet's CapitalG alongside Lux Capital, Sequoia Capital, Bond, and Redpoint.

Its pi0 (pi-zero) model is a Vision-Language-Action (VLA) flow model: it takes visual input plus language instructions and outputs robot motor commands, trained on data from diverse robot types and tasks simultaneously. This multi-robot, multi-task training enables generalisation — pi0 can instruct robots to perform tasks they were never explicitly programmed for. Physical Intelligence open-sourced pi0 in February 2026. The subsequent pi0.5 model doubled manipulation throughput on benchmarks including espresso machine filter insertion, laundry folding, and cardboard box assembly. Plans include deploying models across robot fleets via SaaS licensing — hardware-agnostic, like cloud software.

Humanoid robots in 2026 are not general-purpose substitutes for human workers across all tasks. Failure modes include: tasks requiring fine manipulation below 1mm precision (robot hands lack human tactile sensitivity); environments with unpredictable obstacles or highly variable layouts; tasks requiring tool changes more than once per shift (current robots handle one tool type reliably); and any task requiring human judgment about exceptions. Vendors with honest pilots will show you their robots failing on edge cases — walk away from vendors who claim 100% task coverage.