Theses›Robotics & physical AI›Robot data supply

Robot data supply

Robot data is a stack, not a source; picks-and-shovels win at each part

  • Bear
  • Early
  • $85B by 2035

The callRobot data becomes the binding input for physical AI, but NVIDIA and Encord are taking the data engine; the readout chip is the only slot still open.

D1D2R1R2R3C1foragainst D1D2R1R2R3C1foragainst
Fig. 1The balance. Drivers and catalysts against restraints, each block sized by its weight; catalysts hang dashed because they have not happened yet, sized by their odds. It tips toward the restraints.
  1. D1Humanoids and other robots go into volume production, and every unit trained rather than programmed needs recorded trajectories, force and touch.
  2. D2There is no internet of robot experience to scrape: web video and text carry no force or touch, so data must be collected on purpose.
  3. R1NVIDIA launched Isaac Teleop with a 50-university flywheel and Encord is already priced at about $550M, so the data-engine slot is being captured.
  4. R2Sensors are commoditising (PaXini at ¥199, Meta open-sourcing Digit 360) and module makers signalled at CES 2026 they will fuse compute into sensors.
  5. R3Collection cost is falling fast and world models could substitute for much of the collected data, shrinking what is sold.
  6. C1Cross-embodiment transfer proves out commercially, making robot trajectory data a traded good that needs a neutral clearing layer.2030
Rests onthe beliefs underneath
Supported by · confidence
What would prove itdated tests, open and resolved
202720282029today1234predictionswatch 202720282029today1234predictionswatch
Fig. 2Each ring is a dated test: dashed while open, struck through when overdue, ticked or crossed once resolved.
  1. 1By end-2027 a multimodal robot-data-engine platform reaches a >$1bn outcome or is absorbed by a Scale-AI-class incumbent; the pure annotation/labelling-services layer stays low-margin/services.Dec 2027 · open
  2. 2A multimodal robot-data platform passes $1B or is absorbed by a Scale-AI-class incumbent.Dec 2027 · open
  3. 3Module vendors folding compute into their sensors, and a first primary sizing of the robotics tactile market.Dec 2027 · open
  4. 4The tactile sensor layer commoditises with no Western module company at venture scale, while value moves to the readout chip.Dec 2028 · open
Expressionslisted, private, technologies
  • METAMeta Platforms, Inc.
    • cheap
  • NVDANVIDIA
    • cheap
     Dominant share of AI training compute.
  • APX.AXAppen
    • not screened
     Sells annotated sensor data, robot demonstration trajectories and embodied interaction logs to AI and robotics developers.
  • EXLSiMerit (EXL)
    • not screened
     Provides robotics video data collection, dexterous-manipulation data and annotation to robot developers; now part of EXL, ticker is the parent.
Private · 11 private companies, 3 of them in the US.

Encord · SenseGlove · PaXini Tech · GelSight · mimic · Adaptronics · 1X Technologies · Neura Robotics · Scale AI · XDOF · Lightwheel

Technologies
Who has already bet on itprivate rounds, from public sources

$16B raised by 6 companies, $16B of it in the last three years. Wave under way: Large rounds in the last three years; the category is funded and getting crowded.

Backed by a top-tier venture firm: Scale AI, Encord, XDOF. Backed by a strategic investor: Scale AI.

Reading
Show the reading