Helm.ai extends foundation models beyond automotive

Helm.ai has signed US$70m in commercial contracts over 12 months with automakers, industrial automation firms and Tier 1 suppliers, as its foundation-model software spreads from cars into mining equipment and robotics. The company says it is on course for operating breakeven as its automotive programmes approach start of production.

The deals cover perception, full-stack driving, automated data labelling and generative simulation, with automotive work spanning level 2+ to level 4 development. No customers, revenue or breakeven date were disclosed, leaving the scale of individual contracts unclear.

Away from the road, Helm.ai’s perception software is on a production track for heavy equipment in open-pit mining. Founder and Chief Executive Vladislav Voroninski said driving is the firm’s first market at scale, with the same models already being applied to robotics.

He argued its models need far less data and compute to grasp how the physical world is structured, giving it a route to breakeven in a segment known for burning billions. Helm.ai credits its unsupervised Deep Teaching method, which separates understanding an environment from acting in it.

That pitch targets the costs weighing on rival autonomy developers: fleet-based data collection, large end-to-end models and premium in-vehicle chips. Helm.ai expects to make further announcements later this year.

Why this matters:

• Tesla’s cross-embodiment bet, licensed out. Like Tesla extending its driving stack to Optimus, Helm.ai is betting one model lineage can transfer from vehicles to robots and industrial machines, though it sells that capability to OEMs and suppliers instead of keeping it in-house.
• Training method diverges from Tesla. Helm.ai’s case against fleet-scale data collection and end-to-end models targets the approach Tesla has made its own, so its production programmes could test whether unsupervised training can match fleet data at lower cost.
• Licensed autonomy appeals to OEMs. US$70m in signed contracts suggests automakers without Tesla-sized fleets see licensing as a cheaper route to L2+ and beyond, though signed contract value may take years to convert into revenue.