AN-016 Company lists · list
Top AI Robotics Companies: Who Is Building the Models and Who Is Building the Bodies
Top AI robotics companies in the HardwareMap index: who builds robot models, who builds the bodies, who builds the chips. Hub, stage and raise for each.
by the HardwareMap editors3 min read
From the HardwareMap index, REV 26.09, refreshed 3 Oct 2026
Contents
Top AI robotics companies in the HardwareMap index fall into four groups: robot makers that train their own models, robots built around one AI task, driving models, and the chips that run robot models. This page lists 14 as of REV 26.09. Inclusion rule: the company's record names an in-house AI model, AI software or AI silicon as its product. It sits under top robotics companies.
Ranking rule: group first, then disclosed capital raised inside each group, undisclosed last.
Top AI robotics companies building both the model and the body
These companies train their own control models and build the robot that runs them. That is the most capital-hungry position in robotics, and it is where the largest raises sit.
Figure AI (San Jose) builds humanoids driven by Helix, its own vision-language-action model, announced in February 2025. Figure says Helix takes camera images and natural-language commands and outputs continuous control for the whole upper body, 35 degrees of freedom down to individual fingers, at 200 Hz, running on two low-power GPUs inside the robot (Figure, Helix). That last detail matters: a model that needs a network connection to act is a model that stops when the Wi-Fi does. Its Figure 03 was designed for volume production at the company's BotQ factory. Neura Robotics (Metzingen, Germany) builds what it calls cognitive robots, arms and the 4NE1 humanoid that combine vision, touch and sound sensing; its investors include Qualcomm, Amazon and NVIDIA.
Dexterity AI (Redwood City) applies its models to package handling in logistics and reports over 100 million autonomous actions in production at large enterprises, the most concrete production figure in this group. Sanctuary AI (Vancouver) builds the Phoenix humanoid and the AI control software behind it.
AI built for one robot task
A narrower bet: use learned models to solve one perception or manipulation problem well, and sell the result.
- Standard Bots (Glen Cove, New York): industrial arms controlled with AI software, assembled in the US.
- Carbon Robotics (Seattle): computer vision that tells weeds from crops so lasers can kill the weeds.
- Chef Robotics (San Francisco): robots that portion and assemble meals, run by its ChefOS software.
- CynLr (Bengaluru): vision-guided grasping that lets robots pick objects without per-object training.
These companies have less capital than the general-purpose group but clearer customers. A weed is a weed on every farm, so the model only has to generalize across one job.
Driving models: robotics on wheels
Self-driving software is the most mature form of robot AI, and the companies building it now use the same vocabulary as humanoid makers. Wayve (London) builds end-to-end AI driving software for automakers and closed a $1.2B Series D in February 2026 with Mercedes-Benz, Nissan and Stellantis participating. Waabi (Toronto) builds the Waabi Driver for autonomous trucks and is expanding into robotaxis with Uber. Helm.ai (Redwood City) sells perception and generative AI software to automakers.
The silicon that runs robot models
A robot cannot carry a datacenter. Running a model on the machine itself needs processors that fit a power budget measured in watts.
Hailo (Tel Aviv) makes AI accelerators and vision processors for edge devices, including the generative-AI capable Hailo-10. SiMa.ai (San Jose) designs machine-learning chips for robots, drones and cars, and raised a $150 million Series C in September 2026. DEEPX (Seongnam, South Korea) makes the DX-M1, a neural processor that draws about 5 watts, and is developing the DX-M2 for on-device generative AI. The full chip field is in AI chip startups.
What to watch in AI robotics
Three numbers separate claims from results, and few companies publish all three: the count of robots running the model in paid work, the hours or actions those robots log, and the share of tasks finished without a human stepping in. Dexterity's 100 million actions is one of the few production figures on record in this list.
The open question is whether one general model can run many robots and many jobs, or whether task-specific models keep winning on cost. The general-purpose group has the capital; the task group has more of the deployments on record. Watch for published fleet counts, hours worked and named customers, not demo videos. For the humanoid makers specifically, see humanoid robots. For earlier-stage teams, see robotics startups.
Continue with Autonomous Robotics Companies.
Frequently asked questions
What is an AI robotics company?
A company whose robot is controlled by learned models rather than only hand-written programs. In practice that covers three groups: makers that train their own models and build their own robots, makers that apply AI to one task such as grasping or weeding, and chip companies that run those models on the robot. Some pure model labs build no hardware and fall outside the index.
Which AI robotics company has raised the most?
Among robot makers in the HardwareMap index, Figure AI has the largest disclosed raise, followed by Neura Robotics. Counting driving models as robotics, Wayve's disclosed total is larger still. Disclosed capital is a measure of investor conviction, not of deployed robots, and the two rankings rarely match.
What is a vision-language-action model?
A model that takes camera images and a plain-language instruction and outputs robot motions directly. It extends the large language model approach to physical control. Figure's Helix is one published example. The approach promises robots that learn new tasks from data instead of programming, but public evidence of it working at fleet scale in production is still thin.
Do AI robotics companies build their own chips?
Rarely. Most robot makers run their models on GPUs or off-the-shelf modules. A separate group of chip startups, including Hailo, SiMa.ai and DEEPX in the index, designs low-power processors aimed at running AI on robots, drones and vehicles. Our AI chip startups list covers the wider field, including datacenter chips.
Are self-driving companies AI robotics companies?
By function, yes: a self-driving car senses, plans and acts in the physical world, which is what a robot does. Wayve and Waabi both describe their work as embodied or physical AI. The index files them under mobility rather than robotics, and this page includes them because their models face the same problems robot models do.
HardwareMap tracks AI robotics companies by what they ship. Open any company above for its datasheet, investors and part number.
The index
All 219 startups, filterable by category, hub, stage and funding.