Helm.ai books $70M in physical-AI foundation model deals
Helm.ai reported $70 million in signed commercial contracts over a 12-month period for its physical-AI foundation models. The agreements cover automotive OEMs, Tier 1 suppliers, and industrial automation customers, with additional work in mining and construction. Helm.ai uses an unsupervised deep-teaching approach that separates learning about the physical world from acting within it.
Helm.ai, founded in 2016 and based in Redwood City, California, says its software supports autonomous driving programs from SAE Level 2 to Level 4, perception systems for heavy industry, and growing robotics efforts. Its reported $70 million in contracts over a year involve automotive manufacturers, suppliers, and industrial automation firms, plus mining and construction customers.
The company says its unsupervised training approach first models physical environments and then determines actions within them. This separation is meant to lower data requirements and help systems adapt to new settings while operating within real-world computing constraints. CEO Vladislav Voroninski said the business is progressing toward break-even.
As physical-AI systems mature, they could change how vehicles, industrial sites, mines, and construction sites operate. Workers in those sectors may see tasks shift toward monitoring, maintenance, or exception handling, while companies could gain efficiency and safety benefits. Broader adoption may also raise questions about training, accountability, and access to the data and compute needed to compete. The near-term effects likely depend on deployment scale, regulation, and how organizations integrate these tools.