Physical AI (Robots, Drones, Autonomy)
Physical AI: we cover AI that perceives and acts in the physical world through a robot, drone or autonomous machine, and the software that runs and trains it: robot foundation models and robot learning, humanoids, mobile manipulators, intelligent robot cells, autonomous vehicles and drones, and the simulation and data platforms built specifically to train robots. The focus is industrial: machines that build, inspect, move material through or monitor factories, plants, warehouses, job sites, mines and infrastructure, plus the autonomy stacks that also serve defense and transport. Consumer and household robots are outside the thesis.
Core roster = companies whose primary Investment List is Physical AI. The 24 adjacent players carry it as a secondary list and are shown separately below — never mixed into these numbers.
The Read
Physical AI is a young, capital-heavy list. It holds 168 startups with $44.4B in known funding, or $37.4B without Anduril's $7.0B. The median company was founded in 2022, 90 were founded since 2022, and 121 have raised since 2025. Read the total with care: a few very large late-stage rounds carry much of it. Drones Autonomous Vehicles & Field Robotics, 9 companies, accounts for $15.0B.
The list is scoped by the thesis above, so it follows the factory rather than the robot. What we track is the software that makes the machines useful, read through the MOVES layers: robot foundation models and policies, perception, data and simulation, and the orchestration that turns a robot into a production cell. Hardware-first vendors are tracked where they ship meaningful software. 52 work inside the factory walls, 21 are general-purpose with factories as a main market, and 11 work on construction and mine sites. Robotic & Autonomous Manufacturing (35) and Robot Perception, Data & Simulation (26) are the largest groups by count. 101 of the 168 are based in the United States (60%) and 15 in Germany.
The caveat is evidence. Pilots run long in robotics, and buyers rarely name the vendor of a robot cell, so verified customer evidence trails the capital by a wide margin. Treat the funding as a measure of investor conviction rather than of deployed revenue.
Incumbents & Evergreens
Two different kinds of defense: platform incumbents with installed bases, and evergreens — long-established, revenue-led private players that never raised big rounds and never exited.
Platform incumbents (5 curated products)
Cognex Cognex Machine Vision
FY2024 revenue ~$820M; publicly traded NASDAQ:CGNX; vision software bundled with hardware
Keyence Keyence Sensors, Vision, Measurement
~¥920B (~$6.1B total FY2024); SW+HW tightly integrated; direct sales model; no hardware/SW split
NI (Emerson) NI LabVIEW / TestStand
~$1.7B pre-acquisition; acquired by Emerson 2023 for $8.2B; SW deeply tied to HW platform
Teledyne Teledyne FLIR Imaging Systems
Est. FLIR portion of Teledyne ~$5.5B; acquired 2021 for $8B; imaging SW+HW integrated
Evergreens (1)
1 long-established, revenue-led private player compounds in this list without ever raising big rounds or exiting. The named evergreen roster is a dashboard dataset.
Evergreens are counted here but excluded from the aggregate statistics above.
Top 3 Startups by ThreadMoat Score
Applied Intuition
Skydio
Machina Labs
Ranks 4–144 with full score components
The complete sortable cohort, per-company score justifications, and financial estimates live in the ThreadMoat dashboard.
Ranked by Weighted Startup Quality Score (composite of market opportunity, team execution, technology differentiation, funding efficiency, growth, industry impact, and moat) across the 144-company core roster; acquired and exited companies excluded from the ranking.
Funding Velocity
Latest disclosed round per company, grouped by the year it closed, across the 144 core companies — a funding-velocity signal, not a complete capital history.
Founding Cohorts
When the 144 core companies were founded, in five-year cohorts. Median founding year: 2021.
Round Mix
Latest known funding stage per company, bootstrapped through late-stage.
Industries Served
The industries the core roster sells into, ranked by company count (multi-select — a company serves several industries).
Verified Customers
295 verified customer relationships across 252 organizations — every relationship individually verified, joined by database record ID, never by name matching. Evidence exists for 65 of 144 core companies.
Buyer industry, buyer region and buyer size splits, and the customer-to-startup mapping, are dashboard data.
Absence of public customer proof is not absence of customers, but it is absence of evidence. The splits by buyer industry, region and size, and the named customer-to-startup mapping, are a Strategist dataset.
Who Funds This List
Institutional, venture and corporate investors behind the core roster, with per-investor portfolio counts resolved through the linked investor table.
Geography
Country mix of the core roster against the full ThreadMoat dataset baseline.
Adjacent Players
24 companies carry Physical AI as their secondary investment list. They are adjacent players — shown here, never counted in the statistics above.
24 adjacent players are named, with their primary list, in the dashboard.
Go deeper on Physical AI
The interactive versions of every section on this page — full rankings, named customer mappings, investor matrices — live in the ThreadMoat dashboard.
Frequently Asked Questions
What is Physical AI?
Physical AI: we cover AI that perceives and acts in the physical world through a robot, drone or autonomous machine, and the software that runs and trains it: robot foundation models and robot learning, humanoids, mobile manipulators, intelligent robot cells, autonomous vehicles and drones, and the simulation and data platforms built specifically to train robots. The focus is industrial: machines that build, inspect, move material through or monitor factories, plants, warehouses, job sites, mines and infrastructure, plus the autonomy stacks that also serve defense and transport. Consumer and household robots are outside the thesis. ThreadMoat tracks the software layers that make those machines useful (foundation models, perception and data, simulation, orchestration), not the hardware alone.
Why does Physical AI raise so much more than other industrial software?
Investors are pricing the chance that general-purpose robot intelligence becomes a platform, the way large language models did for software. The rounds are larger because these companies also fund hardware, data collection and long pilots. The capital is concentrated: without Anduril's $7.0B the list holds $37.4B, and 9 Drones Autonomous Vehicles & Field Robotics companies account for $15.0B.
How mature is customer adoption?
Early. Verified customer evidence trails the capital by a wide margin: most deployments are still pilots, and buyers rarely publish which robotics software they run. Expect the evidence to lag the funding for several more years.
Related Investment Lists
Methodology
- • Core roster: companies whose primary Investment List is Physical AI (Robots, Drones, Autonomy), active companies only. Adjacent (secondary-list) players never enter the aggregates.
- • Evergreen roster additionally includes long-established Independent Evergreen players pending re-assessment, and PE-backed ISVs (a non-exhaustive set); neither is a startup, and both are excluded from aggregate statistics.
- • Customer relationships: individually verified, joined by database record ID. Investor data: linked institutional records only; individuals excluded.
- • Percentages are never shown for denominators below 30; every chart footnotes its exact denominator.
- • Data snapshot: 2026-11-15. Statistics recompute automatically from the canonical dataset at every build.