Quantitative
Katana computes its metrics for every company.
- Trailing P/E 0–20 (based on the fiscal year-end share price)
- Three-year revenue CAGR ≥ 10%
A trailing P/E of 20 or less and three-year revenue growth of 10% or more narrowed 3,826 Japanese stocks to 595. AI then read each annual report for a business supplying parts or equipment built into robots. It found none.
In July 2026, NVIDIA presented its AI and robotics work in Japan and named FANUC, Yaskawa Electric and Kawasaki Heavy Industries as building physical AI on its platform: robots that perceive their surroundings, decide and act with AI.
Fujitsu announced at the same time that it would explore physical-AI business with the same three companies. Investment media now list FANUC and others as the leading plays.
So is there a company that is both cheap and actually growing? We checked with Katana, using financial data and annual reports.
“Physical-AI and humanoid plays that are cheap and growing: a trailing P/E of 20 or less, a three-year revenue CAGR of 10% or more, and a current business supplying parts or equipment built into robots.”
We took the criteria from a popular post on X about cheap, growing physical-AI stocks, entered them by hand in Katana, and ran the screen.
Katana computes its metrics for every company.
AI reads each annual report and judges support and counterevidence.
Controllers, sensors or motors for non-robot uses, and plans to enter the market, do not count. Companies that build complete robots fall outside this question.
read the filings and found none.
All 595 judgments came back as requiring review; none were supported. Among the 298 companies growing revenue 15% or more a year, none described a robot-parts business. Below are four top-ranked companies whose filings do mention a robot business.
A Mitsubishi Electric distributor whose dealership contract covers industrial robots.Filing ↗
Makes and sells auto parts, control systems and robot systems.Filing ↗
Develops, makes, sells and maintains industrial robots that carry wafers between chipmaking steps.Filing ↗
Developing small precision motors for humanoid robot joints and fingertips.Filing ↗
These results are from October 3, 2026. Qualitative judgments use the filings available at that time, while quantitative screening uses the latest financial data.
We looked at the companies called the leading physical-AI plays with the same data. Their revenue has barely grown in three years, yet their trailing P/Es run from about 30 to over 200.
| Company | Trailing P/E | Revenue 3y CAGR | Op. income 3y CAGR |
|---|---|---|---|
| FANUC | 29.7 | +0.2% | −1.3% |
| Yaskawa Electric | 40.6 | −0.8% | −11.5% |
| Nabtesco | 31.3 | −0.1% | +4.0% |
| Harmonic Drive Systems | 203.9 | −5.9% | −36.9% |
Theme stocks tend to price in expectations before earnings arrive, and this screen shows exactly that. Ask for both cheap and growing, and the favorites drop out. What remained were a robot distributor, makers of complete robots, and a company just starting development.
Trailing P/E uses the fiscal year-end share price and does not reflect today's price or future earnings.
Complex theses, your own selection rules. Katana narrows companies with its own metrics, and AI reads annual reports to judge qualitative conditions. The selection you used to do one company at a time, AI now runs across the Japanese market.
Open the screener Another example: Companies quietly in semiconductors. Another example: Companies like Nintendo, across all of Japan. Another example: Companies like UNIQLO, across all of Japan. Another example: Recurring revenue, across Japan.