Future of Robotics 2026
Nov 16, 2026

Future of Robotics 2026: the executive summary
Robots score 89.4% on household tasks in simulation, and only 12% in the real world. What closes that gap: better hardware, or better software?
Our Future of Robotics 2026 report maps where robotics capital and robotics revenue sat this year, using data as at August 2026. The short answer: hardware is improving fast, but the scarcest input is the data robots learn from. Here is what we found.
Why 2026 feels different
The case for robotics now is not just one breakthrough but four at once.
Robots have learned to reason. Vision-language-action models are now the default architecture. In 2023, hundreds of hours of demonstrations bought about 65% task success; today, dozens of hours reach 99.9% on defined tasks.
The hardware bill has collapsed. Humanoid build costs fell around 40% in a single year, and Unitree's R1 now retails at $5,900. Labour is the forcing function, with working-age populations shrinking in more than 50 countries. And capital has arrived: global robotics VC reached $55bn by August 2026 and is tracking to $89bn, the most active year on record.
What still holds robots back
The best humanoids run two to four hours between charges, short of a full industrial shift. The hands, not the eyes, are what hold robots back: a leading robot hand can move in 21 independent ways, against 27 for a human hand.
Logistics needs 99.9% accuracy, so 99% is a failing grade. Units can cost anywhere from $15,000 to $250,000, and payback frequently exceeds two years in early pilots. No humanoid safety standard exists yet.
The home: a real market, but not the one you think
Consumer robotics is a $17.6bn market this year, and it is mostly floor cleaning: robot vacuums alone made up 47.8% of 2025 revenue. A few Chinese companies already dominate it. Roborock, Ecovacs and Dreame took 62% of 2025 shipments, and iRobot, maker of the Roomba, filed for Chapter 11 in December 2025.
The home humanoid is a separate business with no installed base yet. 1X's NEO went on pre-order at $20,000, but much of its early task set is teleoperated.
Citi GPS expects households to become the largest humanoid use case by 2050, but that is a long-range scenario. On units, robotics is already a consumer market; on revenue, it is mostly business-to-business.
Where the money went
Investors have backed the brain and the body. Of the 61 US and European companies on Prosus Ventures' index, 57 sit in the model layer or the machine itself, four sit in data, and none sit in deployment. The brain here means foundation and world models, a layer consolidating to a handful, with Nvidia and the model labs setting the price.
Later-stage rounds now take 70–80% of robotics VC dollars. Figure AI carries a $39bn valuation on no published revenue.
Meanwhile, the deployment layer, where measured revenue already is, has barely been funded.
Where we see value
If capital has gone to the brain and the body, the gaps sit around them: the data, software and checks that make robots usable at scale.
Training data. LLMs learned from the internet; robots have no internet-scale dataset of physical interaction. The Prosus map counts four data providers against fourteen model companies.
Evaluation. Prosus names it as one of two places value accrues, yet it barely exists.
Software across robots. Sereact's picking software runs on any robot brand, and mixed fleets need orchestration that no single robot maker has an incentive to standardise.
Compliance tooling. Conformity and safety-case software carries software economics against a deadline: from 20 January 2027, EU Regulation 2023/1230 ends self-certification for machine-learning safety functions. No funded pure-play exists yet.
From machines to the layers around them
So far, robotics capital has focused on the machine: the models that drive it and the bodies that carry it.
We believe the next opportunity sits in the software and data around it.
There are already 4.66m industrial robots in operation worldwide, and that installed base generates service, compliance and resale revenue. From 20 January 2027, every new machine in the EU whose safety functions use machine learning needs third-party assessment, and an update that adds a new hazard reopens it.
A robot is sold once. Keeping it compliant becomes a subscription, not a project.
Our Future of Robotics 2026 report explores this in depth: where capital went, the measured market behind the forecasts, the thirteen segments we screened across the stack, and where Raw Ventures sees opportunities. It reflects our view as at August 2026 and is not investment advice.
Download the full Future of Robotics 2026 report to explore the data and our complete investment thesis.