According to CB Insights, humanoid robot companies are on pace for a record $10 billion in 2026 funding, a figure that would mark a turning point for a sector that spent decades confined to research labs and YouTube demonstrations. BMW has robots from Figure AI working inside its South Carolina factory. Agility Robotics merged with a SPAC at a $2.5 billion valuation. And foundation model breakthroughs from NVIDIA and Google DeepMind have given developers tools that compress years of custom programming into weeks of policy fine-tuning.
The shift feels sudden, though it has been building quietly. Figure AI raised over $1 billion in a Series C round in September 2025 at a $39 billion post-money valuation, a fifteen-fold jump from its Series B just eighteen months earlier. That kind of appreciation typically signals investors betting on platform economics rather than hardware manufacturing margins. Whether the bet pays off depends on deployment numbers that, for now, remain modest.
BMW's announcement that its Spartanburg plant deployment "proves that a humanoid robot can function not only under controlled laboratory conditions but also in an existing automotive manufacturing environment" carried weight precisely because such proofs have been scarce. The German automaker began testing Figure's robots in 2025 and by mid-2026 had advanced to deploying the Figure 03 model on complex sequencing tasks in its Hall 52 logistics operations. Plans to bring humanoids into German production facilities followed the U.S. pilot, a reversal of the usual innovation geography.
From Warehouse Aisles to Assembly Lines
Agility Robotics positioned its Digit robot as "the first humanoid robot employed and commercially operational in warehouse and industrial facilities," according to Churchill Capital's Michael Klein, who led the SPAC merger announced in June. The company moved Digit into live operations at GXO Logistics under a multi-year robotics-as-a-service agreement signed back in June 2024. Amazon began testing Digit in October 2023 and continued evaluations into 2026. Agility now lists nine customer sites spanning GXO, Toyota Motor Manufacturing Canada, and Mercado Libre, according to SEC filings tied to the SPAC transaction.
Boston Dynamics, which spent decades refining Atlas as a research platform, unveiled a product version of the electric humanoid in January 2026 for targeted industrial use. Hyundai said Atlas will deploy at its U.S. manufacturing facility by 2028 for sequencing tasks, marking the transition from viral demonstration videos to production-line work.
Apptronik, an Austin-based startup that spun out of NASA collaboration on the Valkyrie robot, announced its first commercial deployment with Mercedes-Benz manufacturing. The company raised $350 million in February 2025 and later claimed over $935 million for its Series A in February 2026 based on company reporting without independent corroboration. Apptronik has been less forthcoming about deployment timelines and customer specifics than some of its competitors.
Capital Flows and Valuation Questions
Figure AI's September 2025 Series C drew participation from Parkway VC, NVIDIA, Intel Capital, LG Tech Ventures, Salesforce, T-Mobile Ventures, and Qualcomm Ventures. The round followed a $675 million Series B in February 2024 at a $2.6 billion valuation, with Microsoft, OpenAI Startup Fund, NVIDIA, and Bezos Expeditions backing that earlier stage. Brett Adcock, Figure's founder, described the Series C as "critical to unlocking the next stage of growth for humanoid robots, scaling out our AI platform Helix and BotQ manufacturing."
European robotics funding hit €1.8 billion across 35 rounds in the first half of 2026, up 190 percent year-over-year, according to Tech.eu data published in August. Oslo-based 1X raised a $100 million Series B led by EQT Ventures in January 2024 and opened pre-orders for its NEO home robot in October 2025 at roughly $20,000. U.S. deliveries started in 2026. The company later signed a deal to deploy up to 10,000 units across EQT portfolio companies in manufacturing and logistics through 2030, a commitment that blurs the line between strategic investment and customer contract.
China's UBTech, listed in Hong Kong, posted 2025 revenue of RMB 2.001 billion, up 53.3 percent, with humanoid robot revenue from its Walker S series reaching RMB 820.6 million. That represents a 2,203.7 percent jump, though off a small base. The company announced mass production of Walker S2 in November 2025 and a partnership with Hitachi in July 2026, targeting 5,000-unit-per-year capacity. Whether demand materializes at that scale remains to be seen.
Agility's SPAC filing disclosed strategic investors including NVIDIA, Amazon, SoftBank Vision Fund 2, Schaeffler, Foxconn, DCVC, and Playground. Peggy Johnson, Agility's CEO, told reporters at the July 16 Fremont facility announcement that "Fremont is where we build the mind of Digit, while Salem is where we build the body," referring to the company's Oregon manufacturing plant. The geographic split reflects a software-first development approach that venture investors find more appealing than traditional robotics buildouts.
Foundation Models Change the Development Equation
NVIDIA launched Isaac GR00T N1 in March 2025, an open humanoid robot foundation model that allows developers to customize vision-language-action policies without building training infrastructure from scratch. The company followed with N1.5 in May 2025 and introduced GR00T-Dreams, a synthetic motion data pipeline. In June 2026, NVIDIA unveiled a reference humanoid robot built on the Unitree H2 Plus body, Sharpa Wave five-finger hands, Jetson Thor compute, and Isaac GR00T software. Academic researchers gained access later in 2026.
Google DeepMind released Gemini Robotics in March 2025, with enhanced ER 1.5 and ER 1.6 variants arriving in September 2025 and April 2026. The models support perception, planning, and code generation, with on-device variants and API access via Google AI Studio. Apptronik signed a strategic partnership with Google DeepMind to integrate Gemini Robotics on its Apollo platform, betting that Google's language model capabilities translate to robotic manipulation tasks.
Figure developed its own Helix VLA, announced in February 2025, and partnered with Brookfield in September to aggregate egocentric human video across residential and commercial properties. The company described the arrangement as "the world's largest humanoid pretraining dataset." The deal provides Figure with data-center infrastructure and real-world training environments for Helix, addressing a bottleneck that has slowed competitors.
Jensen Huang, NVIDIA's CEO, framed the shift at the March 2025 GTC keynote in characteristically sweeping terms: "Every company will have two factories: one for what they build and another for AI." He positioned GR00T, the company's Omniverse simulation platform, and Newton physics models as a stack that compresses development cycles and reduces the cost of teaching robots new tasks. Whether that vision holds depends on how quickly developers can move from simulation to physical deployment, a gap that has proven stubborn in past robotics cycles.
Chinese Competition and Pricing Pressure
Figure, Agility, Boston Dynamics, Apptronik, 1X, and Sanctuary AI represent the North American cohort targeting industrial and logistics applications. UBTech, Unitree, and Fourier Intelligence anchor China's push, supported by explicit government policy. China's Ministry of Industry and Information Technology launched a 2026 "real-scene training" initiative aimed at unlocking "ten-thousand-unit-scale" deployment capacity by year-end, following 2023 national guidance that set 2025 and 2027 commercialization targets.
Unitree sells research humanoids from roughly $4,900 for the R1 model to $90,000 for the H1, with the H2 Plus priced around $29,900. That pricing undercuts Western platforms substantially and creates pressure on vendors targeting academic and small-scale industrial buyers. Fourier's GR-3, unveiled at CES 2026, positions eldercare and social interaction as an adjacent market to factory automation, a bet that consumer applications might arrive faster than industrial contracts suggest.
Tesla's Optimus remains focused on internal factory deployment. Elon Musk told shareholders that "a majority of Tesla's long-term value will be Optimus," but the company has not disclosed external customer pilots or production volumes. The opacity makes it difficult to assess whether Tesla's vertical integration advantages outweigh the commercial traction other vendors are beginning to demonstrate.
The International Federation of Robotics reported 542,000 industrial robot installations globally in 2024, the second-highest year on record. China accounted for 54 percent of those installations, some 295,000 units. Humanoids represent a fraction of that base today, but foundation models and simplified integration promise faster adoption cycles than traditional programmed automation required. The question is whether "promise" converts to purchase orders at scale.
Regulatory Complexity and Standards Gaps
The EU's AI Act begins staged enforcement in August 2026, with high-risk AI embedded in regulated machinery subject to full compliance by August 2028. The bloc's new Machinery Regulation takes effect in January 2027. Manufacturers integrating humanoids into European facilities face overlapping conformity assessments across both frameworks: documentation, risk management, and traceability requirements that add cost and delay. Legal teams at BMW and Mercedes-Benz are working through those layers now.
The U.S. has no humanoid-specific standards. OSHA relies on general machine-guarding principles, and integrators lean on ISO 10218 for industrial robot safety and ISO/TS 15066 for collaborative robot guidance. NIST announced a Humanoid Robot Baseline Performance Benchmark initiative in April 2026 to establish standardized tests, but the program remains developmental. That vacuum gives early adopters flexibility but creates uncertainty for risk managers evaluating total cost of ownership.
China's MIIT real-scene training action and 2026 standardization committee work explicitly target mass deployment, with national ID systems and lifecycle management frameworks in development. The policy tailwind benefits local vendors but creates compliance divergence for multinationals operating across regions. A robot certified in Shenzhen may require separate documentation for Spartanburg and Stuttgart.
What Scale Looks Like
Goldman Sachs raised its 2035 humanoid total addressable market estimate to $38 billion in January 2024, with a blue-sky scenario reaching $154 billion. Morgan Stanley projects a $357 billion U.S. wage impact by 2040 and a potential $5 trillion global market by 2050, though the firm expects slow adoption until the mid-2030s followed by rapid acceleration. Those projections carry the usual caveats about long-range technology forecasts, which tend to overestimate near-term impact and underestimate long-term disruption.
Bernt Ø. Børnich, 1X's CEO, said the company's "next milestone will be scaling our data collection strategy for Embodied AI and offering NEO to consumers." That dual-market bet appears across the sector: industrial robotics-as-a-service revenue to fund development, direct-to-consumer home robots to unlock volume. Agility opened its Fremont facility in July to accelerate "Physical AI" development, a phrase that sounds like marketing but reflects genuine architectural choices about where computation happens and how models improve.
Figure's Brookfield partnership funds both data aggregation and training infrastructure buildout. The deal structure reveals how venture capitalists are evaluating the sector now. The classic hardware question persists: can software deliver margin structure that looks more like SaaS than contract manufacturing? Figure's $39 billion valuation prices in a platform outcome. Agility's SPAC at $2.5 billion and UBTech's public-market traction offer reference points, but deployment scale remains limited. GXO's multi-year agreement with Agility and BMW's progression from Figure 02 to Figure 03 provide early proof that repeat customer expansion is possible, which matters more than demo videos.
The next twelve months will test whether 2026's $10 billion funding pace converts into deployments measured in thousands of units rather than dozens. Foundation models have solved part of the software problem, or at least made it tractable. The hardware questions remain open: battery life, manipulation dexterity, cost per unit. The business model questions are equally unresolved: robotics-as-a-service economics, total cost of ownership versus human labor, speed of integration into existing workflows.
Capital is riding on the answers, which is perhaps the most human part of this story. The technology has advanced faster than most observers expected. Whether customers show up at the scale investors are betting on is a different question entirely.
