Theses›Robotics & physical AI

Robotics & physical AI

Machines that move and the models that drive them.

  • 13 theses
  • 12 bear
Early5673192Open4Closing12Shut13810Not dated11 Early5673192Open4Closing12Shut13810Not dated11
Fig. 1Where the bets sit: windows from Early to Shut, left to right; each plot's area is the market's size by 2035. A double outline is a bull call, a single one a bear call, a dashed one a thesis with no call written yet. 1 not yet sized are drawn small.
The marketsnumbered as the plan
  1. 1Edge AI inference accelerators
    • Bear
    • Early
    • $9.6B
    • Support 72
     Robots need big models on a battery and today's chips waste energy moving weights, and the one device approach that computes inside stacked memory has thin evidence so far; early.
  2. 2High-mix manufacturing automation
    • Bear
    • Early
    • $2.3B
    • Support 66
     High-mix manufacturing automation: the deployed cell a factory buys, where the blocker is engineering cost per deployment and not robot capability
  3. 3Timing and PNT
    • Bear
    • Early
    • $14B
    • Support 64
     As the physical world fills with autonomous systems that must act together, the scarce bottlenecking input is precise, GPS-denied, trustworthy time and position (chip-scale atomic/optical clocks, PNT resilience), not compute
  4. 4Teleoperation
    • Bear
    • Open
    • $8.6B
    • Support 63
     Between 2026 and 2030, teleoperation is the dominant go-to-market and data-acquisition pathway for physical-AI in contact-rich, high-value verticals (defence and surgery first, then logistics, then manufacturing).
  5. 5Robot data supply
    • Bear
    • Early
    • $85B
    • Support 57
     Robot data is a stack, not a source; picks-and-shovels win at each part
  6. 6Robot autonomy
    • Bear
    • Early
    • $47B
    • Support 57
     General-purpose robot autonomy (open-world dexterous manipulation, not geofenced driving) remains data-bottlenecked and unreliable through 2030.
  7. 7Tactile sensing silicon
    • Bear
    • Early
    • $21B
    • Support 57
     As robots scale to whole-body, high-resolution touch, the contact-data torrent stresses the host-CPU model and pushes readout + first-stage processing toward the sensor node.
  8. 8Robot foundation models
    • Bear
    • Shut
    • $21B
    • Support 57
     Robot foundation models (VLA) are strategically central but value-capture is contested; only a proprietary self-reinforcing data loop earns durable margin
  9. 9Hazardous-environment robotics
    • Bear
    • Early
    • $5.5B
    • Support 57
     Hazardous-environment field robotics: machines bought to enter places a person cannot survive or cannot fit
  10. 10Functional safety silicon for robotics
    • Bear
    • Shut
    • $2B
    • Support 57
     Every scaling robot needs a functional-safety system that forces a safe state when the AI behaves unpredictably or the operator link drops.
  11. 11Service robotics
    • Not dated
    • Support 57
     Service robotics — the VLA inflection opens the $450B cleaning/service-labour market (watch)
  12. 12Humanoid actuators
    • Bear
    • Closing
    • $6.6B
    • Support 47
     Humanoid picks-and-shovels — the component layer holds value only where China compression fails (roller screws, novel-process motors, the sovereignty channel)
  13. 13Surgical robotics
    • Bear
    • Shut
    • $38B
    • Support 44
     Surgical robotics is the most commercially mature, profitable robotics segment (~$8-14bn 2025, ~13-14% CAGR).