Where the capital is actually flowing in physical AI.
3 years of venture investment, 22 sub-sectors, global coverage. Built by Canonical to surface real concentration, hype, and where the market is underweight relative to TAM.
Four humanoid companies have raised more in 18 months than six entire sub-sectors combined.
Figure, Apptronik, Agility, and 1X have together captured more venture and strategic capital in 2024–25 than the entire surgical robotics, agricultural robotics, construction robotics, consumer robotics, exoskeleton, and quadruped industries, combined. The chasm exists because humanoids are pricing in 2030 commercial deployment. The adjacent sectors are pricing in 2026 unit economics.
The most under-appreciated opportunity in physical AI is surgical robotics. Intuitive's monopoly cracked for the first time in 25 years with Medtronic Hugo and CMR Versius Plus FDA clearance, yet VC dollars are 1/20th of humanoids despite 10x the proven unit economics. The chasm goes both ways.
The fastest markups in physical AI are not in humanoids.
Bedrock Robotics, a construction robotics startup from ex-Waymo engineers, went from $80M cumulative to a $1.75B Series B in seven months. That's a 21.9× markup with revenue from production deployments still measured in the millions. Figure's 15× over 19 months is the headline, but Bedrock's velocity is the leading indicator: when a single category gets a Waymo-pedigree mega-round, it pulls the entire stack with it.
Round velocity above 2× with less than 12 months between rounds is the cleanest bubble signal in physical AI. Most of the top-left quadrant won't have proportional revenue scaling. The exception is Anduril, which is why it's the only defense unicorn with a revenue base that justifies the multiple.
US kingmakers underwrite the model. China underwrites the factory.
The US humanoid ecosystem is funded by cloud strategics with RaaS exit paths and 10× revenue multiples. The Chinese ecosystem is funded by tech giants, state guidance funds, and EV-OEM strategic capital, and exits to public markets. Two paths to the same physical world, two completely different unit economics.
Important caveat: China shipped ~90% of humanoid units in 2025 but captured ~30% of dollars. And ~75% of those units went to universities and research institutions. Unitree's own IPO prospectus discloses >50% of revenue from the scientific research and education sector. The volume narrative needs a quality flag.
Unit volume in China is not yet commercial deployment. The structural advantage is real but the customer base is narrow. Watch for the first Chinese humanoid manufacturer to disclose >50% commercial revenue. That's the inflection.
From two unicorns to twenty-two in three years.
In 2022, exactly two pure-play defense tech companies were valued above $1B: Anduril and Shield AI. By mid-2026, the count has crossed 22, and Anduril alone has tripled to $61B post Series H in twelve months, anchored by a $20B DoD enterprise agreement signed March 2026. Andreessen Horowitz has been the most active defense investor with 18 deals across the period.
The most under-examined story is the capital structure. Shield AI's March 2026 round is $1.5B equity at $12.7B plus $500M Blackstone fixed-return preferred. The first sign defense unicorns are institutionalizing balance sheets with structured credit. The category is maturing into a real procurement layer, not a bubble.
Click into any of the 22 sub-sectors for top companies, totals, and active investors.
Filter by stack layer, geography, or sort by capital deployed. Every sub-sector has a Canonical POV blurb and a data-caveat flag.
Tip: hover any chip for what it filters. Click any card for the full company list, investor lineup, geographic split, and Canonical's POV plus the caveats behind the headline number.
How this was built, and where the data is weakest.
Sources & approach
- SourcesHarmonic, Pitchbook Premium, Crunchbase, FT, Reuters, Bloomberg, SEC and HKEX/SSE filings, Sacra, primary company press releases, 36Kr, Caixin.
- WindowQ1 2023 through Q1 2026. Public market caps updated weekly; private rounds version-stamped at last close.
- Taxonomy22 sub-sectors across embodiment / domain / stack layers. Multi-tagging supported. Canonical's framework, refined with each release.
- RefreshQuarterly with running corrections. Subscribe for the next update.
Known limitations
- Total VC '25Reported at $27.6B (PitchBook) to $40.7B (CB Insights). We show the range. Methodology cuts diverge on what counts.
- China dataUnit-share figures (Omdia, Counterpoint) and revenue mix (HelloChinaTech) use different methodologies. We flag both.
- Rounds in talksSeveral 2026 rounds (PI $11B, Helsing $18B, Project Prometheus $10B) are reported but not confirmed closed. Tagged accordingly.
- Coverage gapsStealth companies and unannounced rounds are by definition absent. Spot one missing? Tell us.
If you're building or investing in the picks-and-shovels of physical AI, we want to hear from you.
Canonical leads early-stage investments in technical teams building at the frontier. SF-based, founder-obsessed, pre-product friendly.
Built by Anand Iyer at Canonical · v0.1 · Educational tool. Not investment advice. Data is point-in-time and not exhaustive.