Aptiv said it ran advanced AI models on NVIDIA's Thor with air cooling alone
Expanding its work with NVIDIA, Aptiv framed the result as a thermal and cost claim rather than a capability one: supporting advanced AI models while relying exclusively on air cooling, which it called an industry first for Thor-based computing platforms.
Aptiv said on October 21 that it had expanded its collaboration with NVIDIA, and put the ability to support advanced AI models while relying exclusively on air cooling for thermal management at the center of the announcement. The company called that an industry first for Thor-based computing platforms, and said it results in significant systems savings for the carmaker, particularly on internal combustion vehicles.
The collaboration brings NVIDIA’s Hyperion architecture and its Halos safety system into Aptiv’s modular compute architecture, which the supplier said is able to support AI models developed in-house by a customer, by Aptiv, by NVIDIA or by a third party. Aptiv framed the commercial case around cost, saying affordability remains one of the most significant impediments to widespread consumer adoption of highly automated vehicles, and listed the intended benefits as cost-effective solutions from hands-free to fully autonomous driving, flexibility across sensors, software applications, AI models and vehicle architectures, and continuous improvement through software and AI model updates.
Sandeep Punater, Aptiv’s vice president for product and business development in autonomy, said the two companies were combining AI compute with the supplier’s industrialization capabilities in an open solution. Rishi Dhall, NVIDIA’s vice president for automotive, said the platform combined with Aptiv’s integration expertise can help reduce system cost and complexity.
The release names no carmaker, no program and no volume, and discloses no financial terms. Aptiv also described itself as supplying the first and only turnkey hands-free Level 2-plus driving platform from a technology supplier in production — its own characterization, without third-party corroboration. It said it expects to show the work running on Hyperion at CES 2027.
Source: Aptiv newsroom