Helm.ai says it has $70M in signed contracts for its physical AI models

Helm.ai, a Redwood City, California company that builds AI software for autonomous driving and industrial machines, has signed $70 million in commercial contracts for its foundation models over a 12-month period, it told The Robot Report on October 8, 2026. The customers are global carmakers, Tier 1 automotive suppliers and industrial automation companies, with work in mining and construction alongside the core automotive business.

Chief executive Vladislav Voroninski said the company is now on a path to break even, which he described as rare in the sector.

Is $70 million revenue?

No. It is the value of contracts signed over a year, as reported by the company. Automotive software deals usually run over several years, with payments tied to development milestones or to vehicles shipped, so recognised revenue in any one year will be lower and less certain than the headline. Helm.ai did not give a revenue figure, a customer list or the length of the contracts.

Even so, a signed-contract number is more than most physical AI companies disclose. As we noted in our look at where physical AI revenue is booked, the money in this sector has so far gone mostly into funding rounds rather than coming out of customers.

One perception stack, several kinds of machine

Helm.ai trains its models with what it calls “deep teaching”, an unsupervised approach that separates learning to perceive an environment from learning to act in it. The company argues that the same perception stack can serve a passenger car, an open-pit mine and other industrial settings, and it has also applied its AI to delivery drones. Voroninski said Helm.ai wants to work with robotics companies across industries and body types.

That positions Helm.ai as a supplier rather than a robot maker, in the same layer as the general robot-model companies we compare in our foundation-model guide. Its advantage is that automotive customers already pay for perception software at production scale. The test for the robotics push is whether perception trained mainly on roads and mines transfers to manipulation, where most robot-model competition now sits.