Trends›Physical AI deploys this decade

Physical AI deploys this decade

General-purpose physical AI and robotics crosses from demos to real scaled deployment this decade, because the data and reliability bottleneck gets solved well enough for scaled use by 2030.

44 confidence, computed 23 Oct 2026
  • Technical
  • Contested
AprJulOctJan 2026AprJulOct44nowforagainst JulJan 2026Jul44nowforagainst
Fig. 1The evidence is split about evenly by weight; the newest points against.
What rests on thismarkets it underwrites or threatens
Underwrites
Threatens
Evidencethe schedule the figure draws
The evidence, newest first
  • Jul 2026
    against · primary     The volume leader's own disclosure shows early humanoid demand is NOT labour substitution: of 5,215 humanoids sold in 2025, >70% went to research/education and ~17% to commercial/consumer bought largely 'for show' (retail promoters, tourist sites, events), leaving little doing productive… Unitree STAR Market IPO prospectus: 2025 units, ASPs and customer mix
  • Jul 2026
    against · primary     Only ~40% of humanoid pilots reach production deployment within 24 months; industrial customers expect 99.9%+ uptime against a wide lab-to-production gap, and Tesla had not begun Optimus production at Fremont as of mid-July 2026 — pilot-to-production conversion, not capability, is… Humanoid Robots in 2026: What Is Actually Deployed
  • Jul 2026
    for · primary     Agility's Digit has logged 65,000+ operating hours across nine customer facilities (GXO, Schaeffler, Toyota Canada, Mercado Libre), passed 100,000 totes moved under a RaaS contract at GXO, and is the only humanoid generating revenue from paying commercial customers — sustained… Humanoid deployments in 2026 favor Figure and Agility
  • Jul 2026
    for · primary     Waymo reached ~500,000 paid rides/week by May 2026 (from ~250,000 in April 2025), targets >1M/week by end-2026 (~$1B annualised), and expands from 6 cities toward 20+ including London and Tokyo — the deployed-autonomy proxy still compounding. Waymo goes driverless in Las Vegas, with Denver, San Diego, Tampa next
  • Jul 2026
    for · observable     Figure 02 logged 1,250 hours at BMW Spartanburg before Figure 03 entered production at roughly one unit per hour at BotQ, at an estimated ~$20,000 unit price or ~$1,000/month RaaS — manufacturing cost curve moving toward deployable economics. Figure 03 enters production at BotQ
  • Jun 2026
    against · observable     Sustained months of reliable factory operation remains undemonstrated by any humanoid company as of mid-2026. Humanoid Robotics Challenges [2026]
  • Jun 2026
    against · primary     Data, not architecture or compute, remains the binding constraint on VLA progress — pooling heterogeneous robot datasets induces NEGATIVE transfer, in-distribution-only data gives near-zero transfer to new environments, and a curated 5% coreset recovers 85-90% of full-dataset performance, so most… Vision-Language-Action in Robotics: A Survey of Datasets, Benchmarks, and Data Engines
  • Apr 2026
    against · primary     Two hard questions remain unanswered: how much more data, and what kinds, will be needed to close the gap between lab performance and the 99.9% reliability threshold that production deployment demands. Bessemer Predicts: Robotics and physical AI
  • Feb 2026
    for · observable     Waymo raised $16B (Feb 2026) to expand to London and Tokyo and targets 1M paid rides/week by end-2026 across 20+ new cities. Waymo raises $16B to scale internationally (TechCrunch)
  • Jan 2026
    against · primary     The main challenge in 2026 is no longer building robots but creating datasets that truly reflect real-world conditions. Why Data, Not Models, Is the Real Bottleneck in Robotics
  • Dec 2025
    for · primary     Waymo ended 2025 at 450,000+ paid rides/week (up from ~175,000 at the start of the year) and served 14M+ trips in 2025, tripling YoY — a hard autonomy problem scaling commercially. Waymo 2025 Year in Review — 450k+ paid rides/week
  • Oct 2025
    against · observable     Many 'autonomous' robot demos still rely on teleoperation (e.g. Humanoid robots not ready for primetime — teleop reliance (TechTarget)
  • Oct 2025
    against · primary     No humanoid company has shown sustained months of reliable autonomous factory operation; reliability sits well below the 95-99% uptime industrial robots achieve — the reliability bottleneck is not yet solved. Reality is ruining the humanoid hype (IEEE Spectrum)
  • Jul 2025
    for · primary     Amazon passed 1,000,000 deployed warehouse robots across 300+ facilities (assisting ~75% of deliveries) and launched DeepFleet — physical AI at genuine production scale. Amazon deploys its 1 millionth warehouse robot (TechCrunch)
  • Apr 2025
    for · primary     π0.5 performed long-horizon tasks (cleaning kitchens, tidying bedrooms) in homes it had never seen, with generalization improving as training scaled from 3 to 104 homes — the data-diversity bottleneck loosening. π0.5: a VLA with open-world generalization (Physical Intelligence)
  • Mar 2025
    for · observable     NVIDIA released GR00T N1, an open VLA foundation model with a data pyramid (web video, synthetic, real-robot) explicitly designed to bypass the cost of large real-robot datasets. NVIDIA Isaac GR00T N1 open humanoid foundation model
What would move itthe measures watched