25 Fast-Growing Physical AI Companies in 2026
Ranked: the 25 fastest-growing physical AI companies in 2026, led by Sanctuary AI with humanoids, robot brains, and autonomy.
25 Fast-Growing Physical AI Companies in 2026
AI has left the chatbox. From humanoids on factory lines to autonomous logistics at scale, these are the 25 fastest-growing companies in physical AI in 2026 — ranked by technology, traction, and momentum.
Updated August 2026 · 25 companies · 8 categories
Sanctuary builds what may be the most human-like hands in robotics — hydraulic, 20+ degree-of-freedom hands with proprietary tactile sensors — paired with an AI control system it now deploys across third-party industrial robots, not just its eighth-generation Phoenix humanoid. In 2025–26 it used NVIDIA Isaac Lab reinforcement learning to achieve zero-shot in-hand manipulation and validated 99.5%+ task success on wire-assembly work at a Tier 1 automotive supplier. Ranked #3 globally for humanoid-related patents by Morgan Stanley, it is Canada's flagship physical AI company — and the one to watch in 2026.
Known for: Phoenix humanoid + hardware-agnostic Physical AI control system
sanctuary.ai →The most richly valued humanoid pure-play on Earth, Figure ripped from a $2.6B valuation to $39B in 18 months on the strength of its in-house Helix vision-language-action model and the consumer-grade Figure 03. It runs its own BotQ manufacturing line targeting six-figure annual output and has moved robots from BMW's Spartanburg plant into logistics and early home testing.
Known for: Figure 03 humanoid + Helix VLA model
figure.ai →Founded by Sergey Levine, Karol Hausman, and a murderers' row of Berkeley, Stanford, and DeepMind roboticists, "Pi" is building one foundation model to control any robot — open-sourcing π0 and demonstrating robots that clean kitchens they've never seen. Backers include Jeff Bezos, OpenAI, and Thrive. It is the purest software bet that robotics' GPT moment comes from a model, not a body.
Known for: π0 / π0.5 generalist VLA models
physicalintelligence.company →Anduril turned defense into a software-defined, mass-produced autonomy business, with its Lattice OS coordinating drones, interceptors, and unmanned aircraft. Its 5-million-sq-ft Arsenal-1 "hyperscale" factory in Ohio began ramping in 2026, and it took over the US Army's IVAS mixed-reality program. Doubling to a $61B valuation in a year — with ~$100B talks reported — it is the defining company of the defense-tech boom.
Known for: Lattice OS, Fury autonomous fighter, Roadrunner
anduril.com →Spun out of Carnegie Mellon by Deepak Pathak and Abhinav Gupta, Skild trains a single model that transfers across quadrupeds, humanoids, and manipulators — hardware-agnostic by design. Its January 2026 Series C, with backers including SoftBank, NVIDIA, and Samsung, made it the best-funded robotics foundation model company in the world. The anti-Figure: sell the brain, let others build the bodies.
Known for: Skild Brain — an omni-bodied robot brain
skild.ai →1X made the boldest consumer bet in robotics: NEO, a soft-bodied, tendon-driven humanoid for the home priced at $20,000 (or $499/month), opened US preorders in October 2025 and entered full-scale production in April 2026, targeting capacity for 100,000 units by 2027. If humanoids reach living rooms this decade, 1X likely gets there first.
Known for: NEO — the first mass-market home humanoid
1x.tech →Unitree commoditized legged robots, selling a capable G1 humanoid from ~$16,000 and an R1 from under $6,000 — its machines dominate research labs and viral demos worldwide. Already profitable on hardware volume, it launched a $619M STAR Market IPO in July 2026, the humanoid era's first blockbuster listing. China's most visible answer to Western humanoid hype: ship cheap, iterate fast.
Known for: G1 / R1 humanoids, Go2 quadruped
unitree.com →Born from UT Austin's Human Centered Robotics Lab and NASA Valkyrie work, Apptronik pairs its Apollo humanoid with Google DeepMind's Gemini Robotics models — a rare hardware-first company with a frontier-lab brain. Its record-setting Series A (over $935M) funds commercial deployments with Mercedes-Benz, GXO, and manufacturing scale-up via Jabil.
Known for: Apollo humanoid + Google DeepMind AI
apptronik.com →While rivals demo, Digit clocks in: Agility's bipedal robot handles totes for GXO and Schaeffler under robots-as-a-service contracts — the industry's first real humanoid revenue. Its Oregon RoboFab factory is built for 10,000+ units a year, and its 2026 move to go public at ~$2.5B makes it the first US humanoid maker to hit public markets.
Known for: Digit — first humanoid doing paid warehouse work
agilityrobotics.com →Wayve's camera-first, learned end-to-end driver needs no HD maps, letting it generalize to new cities in days — an embodied-AI approach that won over NVIDIA, Microsoft, SoftBank, Uber, and Mercedes as investors. Nissan is set to ship Wayve-powered ProPILOT driver assistance, with robotaxi pilots alongside Uber in London. Europe's strongest claim to a driving foundation model.
Known for: AV2.0 — end-to-end, mapless self-driving
wayve.ai →Applied Intuition is the picks-and-shovels giant of vehicle autonomy, supplying development tools, simulation, and its off-the-shelf Vehicle OS to carmakers, truckers, miners, and increasingly the Pentagon — including autonomy for US Army vehicles. Its valuation jumped nearly fivefold from 2024 as every "moving machine" maker outsources its software stack. Boring, essential, everywhere.
Known for: Vehicle OS + simulation for 18 of top 20 automakers
appliedintuition.com →Founded by former Huawei “genius youth” Peng Zhihui, AgiBot became the volume leader of the humanoid era — rolling its 10,000th robot off the line and making a splashy US debut with its full portfolio at CES 2026. It pairs mass manufacturing with serious research output, including its GO-1 embodied foundation model and one of the largest open humanoid manipulation datasets. No startup anywhere ships more humanoids today.
Known for: Yuanzheng A2 / Lingxi X2 humanoids + GO-1 foundation model
agibot.com →Founded by embodied-AI researcher He Wang, Galbot skipped legs for a pragmatic wheeled dual-arm humanoid and put it to work operating dozens of unmanned convenience stores in Beijing on its GroceryVLA model. With CATL as anchor investor, a ~$360M March 2026 round, and a Hong Kong IPO in the pipeline, it is among the first Chinese embodied-AI startups to reach real commercial replication.
Known for: Galbot G1 running unmanned retail stores
galbot.com →UBTech moved humanoids from pilot to purchase order: Walker S2 entered mass production in late 2025 with an order book exceeding RMB 800M (~$110M) from automakers like BYD, Foxconn lines, and logistics firms. Its hot-swappable battery design enables near-24/7 operation — a genuinely practical edge. As the West debates humanoid ROI, UBTech is already invoicing for it.
Known for: Walker S2 — mass-produced factory humanoid
ubtrobot.com →Europe's biggest embodied-AI bet, NEURA builds sensor-rich "cognitive" cobots already selling into factories alongside its 4NE1 humanoid, tied together by Neuraverse — an app-store-like platform for robot skills. Its June 2026 Series C of up to $1.4B, one of the largest robotics rounds ever, positions it as the continent's counterweight to US and Chinese humanoid leaders.
Known for: 4NE1 humanoid, MAiRA cobot, Neuraverse
neura-robotics.com →Founded by ex-NASA JPL autonomy lead Ali Agha, Field AI builds embodied foundation models that let quadrupeds, humanoids, and vehicles operate in GPS-denied, unmapped places — construction sites, energy plants, defense settings — where most robot brains fail. Its models quantify their own uncertainty, a safety-first twist on the foundation-model race. The $405M haul at a $2B valuation in August 2025, backed by Bezos Expeditions, Khosla, Temasek, and NVIDIA’s venture arm, made it one of the fastest-scaling physical AI startups in America.
Known for: Field Foundation Models — risk-aware universal robot brains
fieldai.com →Zipline crossed 2 million autonomous commercial deliveries — more than anyone on Earth — evolving from blood drops in Rwanda to suburban doorstep delivery for Walmart, which celebrated its 1 millionth drone delivery in May 2026 largely on Zipline's wings. Quietly, it runs the largest autonomous logistics network in existence.
Known for: Platform 2 "Zips" precision aerial delivery
flyzipline.com →Nuro pivoted from building its own delivery pods to licensing its Level 4 Nuro Driver — and landed the deal of the cycle: powering Uber's premium robotaxi program on Lucid Gravity EVs, with plans for 20,000+ vehicles and testing underway in San Francisco since spring 2026. Uber's deepened ~$500M commitment makes Nuro the credible third player in US robotaxis.
Known for: Nuro Driver powering the Lucid–Uber robotaxi
nuro.ai →Fei-Fei Li's bet that AI's next frontier is spatial: World Labs' Marble generates navigable, persistent 3D worlds from text or images — the substrate for training and simulating embodied agents at scale. Its ~$1B early-2026 raise underscores investor conviction that world models are to physical AI what LLMs were to language. Possibly every robot company's future training ground.
Known for: Marble — generative, explorable 3D world models
worldlabs.ai →Spun from CMU and Mistral alumni, Genesis AI attacks robotics’ data problem with ultra-fast physics simulation — its open-source Genesis engine generates synthetic training data orders of magnitude faster than real-world collection. By May 2026 it had gone full-stack, demoing a foundation model controlling human-like robotic hands on real hardware. One of the largest seed rounds in robotics history, now reportedly commanding a ~$3B price tag.
Known for: Open-source Genesis physics engine + dexterous manipulation model
genesis.ai →Fourier took an unusual road to humanoids — through rehabilitation robotics, where its exoskeletons already serve thousands of hospitals — giving it rare clinical credibility in embodied AI. Its GR-3 humanoid, launched in 2025 as a soft-touch “care companion” with emotional interaction features, targets healthcare and eldercare rather than the factory floor. In a crowd of warehouse humanoids, Fourier owns the caregiving niche.
Known for: GR-3 “care-bot” humanoid + GRx robot series
fftai.com →Serve's cheerful Level 4 sidewalk robots grew from an Uber/Postmates skunkworks into a public-company fleet of roughly 2,000 third-generation bots delivering across LA, Miami, Dallas, Atlanta, and Chicago. Partnerships with Uber Eats, DoorDash, and Shake Shack give it demand density rivals lack — the most visible everyday face of physical AI on American streets.
Known for: AI sidewalk delivery for Uber Eats & DoorDash
serverobotics.com →Founded by former Amazon Robotics VP Brad Porter, Cobot rejects the humanoid form factor on purpose: its cart-pulling Proxie robot does useful materials-handling work today in hospitals, manufacturing, and logistics hubs. The second-generation Proxie, announced in June 2026, brought “production-tested physical AI” from pilots into scaled real operations. It is the pragmatist’s physical AI company — boring form, real deployments.
Known for: Proxie — a practical non-humanoid mobile manipulator
co.bot →Gecko's magnetic wall-climbing robots scan power plants, refineries, dams, and US Navy ships, feeding sensor data into its Cantilever platform to predict failures before they happen. Unicorn status arrived in June 2025 as defense demand surged. Physical AI applied to the unglamorous trillion-dollar problem of keeping civilization's hardware from rusting apart.
Known for: Wall-climbing TOKA robots + Cantilever AI
geckorobotics.com →Dexterity builds force-sensitive, AI-driven arms that do the jobs humans hate most — loading and unloading trailers, palletizing mixed freight — for customers including FedEx, UPS, and Sumitomo. Its forklift-mounted Mech packs dozens of AI models running in real time to stack boxes with human-level touch. A reminder that the biggest near-term physical AI market may be the loading dock, not the living room.
Known for: Mech — two-armed "industrial superhumanoid"
dexterity.ai →