Robots Priced as Labor: The Physical-AI Wave Is Real — and Its Easiest Entrance Is Already a Tarpit
YC is doubling its robotics bet batch over batch, but the winners will sell headcount in unclaimed trades — not training data to a few dozen labs.
By PlatoSeed Research · grounded in the live corpus
The wave right now
For ten years the joke was that software ate the world except the part of it with mass. That excuse is gone. Vision-language-action models made manipulation trainable instead of programmable, and the labor market did the rest: welders, warehouse pickers, farmhands, and caregivers are scarce at any price. The companies in this wave aren't selling robot demos — they're selling headcount. InLoop Robotics rents a bimanual warehouse worker for a flat monthly fee with no capex; Servo7 sells container unloading that learns on the job. The pitch has changed from "look what it can do" to "here's the loaded cost of the human it replaces." That reframing — robots priced as labor, not equipment — is the single most important shift in the category, and it's why this wave deserves your attention even if the last one burned you.
The landscape today
The data gold rush is the loudest — and the most crowded — corner. A Scale-AI-for-robotics layer has materialized almost overnight: Human Archive claims 100,000+ contributors feeding multimodal capture across homes, farms, and factories, while Asimov literally pays people to record their daily movements for humanoid training. The differentiated entrants go after data nobody else can get — Remy AI straps tactile sensors and head-mounted capture onto industrial workers to document dexterity, and One Robot sidesteps collection entirely with physics-accurate world models for synthetic data and evals. The thesis is sound (foundation models are genuinely data-bottlenecked), but count the sellers against the buyers: there are only so many funded robot labs to invoice.
Human ArchiveMultimodal data provider for robotics and world modeling
AsimovReal-world human movement data for humanoid robots
Remy AIDocumenting human dexterity
One RobotWorld models for robot evals and training.
Robots-as-a-Service is the business-model unlock for labor-starved verticals. The smartest hardware companies in this cohort have stopped asking customers to buy robots. InLoop Robotics and Eden Robotics both lead with rental economics and remote human recovery, which converts a capex decision into a staffing decision a plant manager can make alone. The vertical picks are telling: Advanced Metal Research is going at American welding — a trade where the skills shortage is brutal and the labor math is undeniable — and Synphony is doing the same calculus on farms. The moat claim in every case is the same: production deployments generate the training data that makes the next deployment cheaper.
InLoop RoboticsStaff your warehouse with robotic employees today
Servo7Robots that 10x manual industry work
Advanced Metal ResearchMachine intelligence for American Welding
Synphonyrobots & software for farms
Humanoids are drifting toward soft, social markets — and the form factor is not the moat. Twolabs puts semi-humanoids into nursing homes for mobility support and companionship; Piggy Robotics promises chores "for the price of an iPhone." Meanwhile the brains layer is being built separately — Verne Robotics teaches robots new skills in hours, and Intelligence Factory bets that general manipulation across any hardware beats task-specific training. My read: the humanoid premium is a fundraising aesthetic, and the durable claims in this group are about deployment context (a nursing home that trusts you) and learning speed, not the shape of the machine.
TwolabsHumanoid robots for caregiving.
Piggy RoboticsHumanoid robots that do your chores for the price of an iPhone!
Verne RoboticsAI models that teach robots new skills in hours
The quiet opportunity is the aftermarket forming around deployed fleets. Every robot that ships creates recurring problems someone must own. Avea Robotics sells low-latency teleoperation as the human-oversight layer across heterogeneous fleets; Aseon Labs builds robotic pitstops — charging, cleaning, inspection — for autonomous vehicle fleets; DeepAware AI will deliver robots, parts, and RL environments in 72 hours. This is the least glamorous and least crowded pattern in the whole context, and it compounds with everyone else's success rather than competing with it.
Avea RoboticsTeleoperate your robots from anywhere in the world
Aseon LabsRobotic pitstops for self-driving cars
DeepAware AIRobots, parts & RL environments - delivered in 72 hrs or pickup in SF.
The cohort signal
This is a deliberate, accelerating program bet. By batch: Spring 2025 seeded six robotics companies, Summer 2025 three, Fall 2025 six, Winter 2026 five — and Spring 2026 alone has eleven, the largest robotics cohort in the dataset. The composition is rotating too: the Winter 2026 class skews picks-and-shovels (data, simulation, teleop-for-data), while Spring 2026 swings hard toward deployed machines doing real work — welding cells, warehouse staff, farm robots, caregiving. One more tell: a striking share of the current cohort arrived here via a rename — Eden Robotics, Tensr, and Mbodi AI among them — meaning experienced founding teams are actively pivoting *into* physical AI. Capital and talent are both rotating toward the atoms.
Lessons from the last cycle
The 2016–2021 robotics class left a legible scorecard. The horizontal platform plays died: Voodoo Manufacturing's "AWS for manufacturing" went inactive, as did Robby Technologies' sidewalk delivery fleet and Shone's autonomous-cargo-ship retrofits — all underestimated how long full autonomy and capital-intensive fleet ownership take to pay back. What got acquired? Supervised autonomy in a single high-wage vertical: Teleo (remote-operated heavy equipment) and Roin Technologies (concrete construction robots) both exited. The pattern is almost embarrassingly clean — keep a human in the loop, pick one trade with desperate labor economics, sell productivity rather than autonomy. The current cohort's excavator and teleoperation bets are rerunning exactly that acquired playbook, this time with foundation models doing more of the work.
If you're building here
Opening 1: Own the fleet aftermarket. Servicing, downtime insurance, safety certification, and maintenance for deployed robot fleets is nearly empty — one company is doing pitstops for AVs, nobody owns it for warehouse and factory robots. If even a fraction of this cohort ships at scale, the installed base needs a Jiffy Lube. You'd have to believe RaaS deployments actually scale past pilots in the next 18 months.
Opening 2: Pick an unclaimed trade and run the Teleo playbook. Welding already has two credible entrants (Advanced Metal Research, Forge Robotics); lab automation has Zeon Systems; data centers have Boost Robotics. But the list of $40/hour trades with no robot company is still long — pick one, sell supervised autonomy as labor, and let production data become your moat. You'd have to believe the trade's work is repetitive enough for today's manipulation models.
The tarpits, by name: generic robotics training-data collection — this context alone holds seven-plus data/sim sellers chasing a customer base of a few dozen funded labs, and the incumbents already claim six-figure contributor networks; and the consumer chore humanoid, where you're betting capital-intensive hardware against a price point ("an iPhone") that no bill of materials currently supports. Both are crowded precisely because they're the easiest version of the idea to start. The wave is real; the easy entrances are already jammed. Go where the labor is expensive and the founders aren't.
Key companies in this memo
The headline bets — outcomes and all. (+14 more linked throughout the piece.)
Human ArchiveMultimodal data provider for robotics and world modeling
AsimovReal-world human movement data for humanoid robots
One RobotWorld models for robot evals and training.
InLoop RoboticsStaff your warehouse with robotic employees today
Servo7Robots that 10x manual industry work
Advanced Metal ResearchMachine intelligence for American Welding
Synphonyrobots & software for farms
TwolabsHumanoid robots for caregiving.
Avea RoboticsTeleoperate your robots from anywhere in the world
Aseon LabsRobotic pitstops for self-driving cars
Remy AIDocumenting human dexterity
Eden RoboticsAutonomous services for humanity.
Build on this thesis
Generate grounded startup ideas steered by this memo — anchored to the real companies above.