IN Brief:
- Aptiv is integrating NVIDIA Hyperion and Halos into modular computing platforms for automated driving.
- The company says its system optimisation can support advanced NVIDIA Thor workloads using air cooling rather than liquid cooling.
- Aptiv plans to demonstrate its automated driving capabilities on NVIDIA Hyperion at CES 2027, but has not disclosed detailed thermal benchmark results.
Aptiv is expanding its collaboration with NVIDIA to integrate the Hyperion automated driving platform and Halos safety technology into modular automotive computing systems, including a proposed air-cooled configuration for NVIDIA Thor processing. The development targets vehicles requiring increasingly complex perception and automated driving functions without the additional equipment associated with liquid cooling.
The agreement, announced on 21 October, extends earlier work between the companies on production-supported NVIDIA Jetson computing platforms. The new programme concerns the integration of automated driving hardware, software and safety development technologies into complete automotive systems, with Aptiv planning to demonstrate its latest capabilities at CES 2027.
Aptiv is combining NVIDIA Hyperion with its own computing architecture, sensing technology, software integration and manufacturing capabilities. The approach is intended to allow vehicle manufacturers to select different combinations of processing hardware, sensors, middleware and automated driving functions according to the requirements of an individual vehicle programme.
The available functions range from advanced driver assistance to more extensive automation. Those applications impose different requirements for sensor coverage, computing performance and responses to abnormal operating conditions, even where they share parts of the same electronic architecture.
Modern automated driving systems receive information from cameras, radar and other sensors, then process those observations to identify objects, interpret the road environment and calculate appropriate vehicle behaviour. The electronics must complete these operations within the time available for the intended function, while continuing to manage communication with other vehicle systems.
Increasing the complexity of the AI models used for perception and planning can raise computing demand, particularly where several sensor streams must be processed simultaneously. Memory bandwidth, processor utilisation and the execution characteristics of the selected models influence whether the system can produce the required results within its operating limits.
The resulting thermal load is an important part of automotive electronic design. Electrical energy consumed by processors and supporting circuits is substantially converted into heat, which must be transferred away from sensitive components. Semiconductor junction temperatures must remain within specified limits during sustained operation and under the environmental conditions expected in the vehicle.
Aptiv says its system optimisation can support advanced AI models on NVIDIA Thor while relying exclusively on air cooling for thermal management. The company describes the arrangement as an industry first for Thor-based automotive computing and identifies reduced system cost and complexity as potential benefits.
An air-cooled assembly transfers heat through thermal interfaces, heatsinks and surrounding airflow. Its effectiveness depends on the heat generated by the electronics, the available surface area, airflow, mechanical arrangement and temperature difference between the equipment and its surroundings.
Liquid cooling provides an alternative in applications where the available airflow or packaging geometry cannot remove sufficient heat. Depending on the installation, a liquid system may require pumps, coolant connections, additional heat exchangers and associated control or monitoring equipment. Removing those components can simplify an automotive assembly, but air cooling still requires appropriate mechanical design and verification.
Vehicle conditions can make thermal performance difficult to generalise. A computing module installed behind an instrument panel, within an equipment enclosure or in another restricted location may receive different airflow and experience different ambient temperatures. Heat from surrounding electrical or mechanical equipment can also affect the operating environment.
Airflow must remain adequate when computing demand changes, including periods of sustained model execution. Thermal behaviour during a short demonstration may differ from operation at a high ambient temperature over an extended journey. Qualification therefore requires measurements covering representative workloads and installation conditions.
Aptiv has not disclosed the Thor configuration, sustained processor power, maximum ambient temperature, heatsink dimensions or measured thermal margin of the proposed platform. It has also not published a controlled comparison quantifying the savings against a liquid-cooled design. The air-cooling capability remains a supplier engineering claim rather than an independently characterised product specification.
The collaboration also incorporates NVIDIA Halos, a set of technologies and development processes intended to support safety engineering for automated vehicles. Aptiv is combining that framework with its own integration and validation capabilities, allowing the selected computing architecture to be evaluated as part of a complete vehicle system.
The inclusion of a safety development framework does not establish that a particular vehicle has completed all relevant safety assessments. The selected sensors, processors, software, communications and control functions must be examined together, including their behaviour when a component fails or produces unexpected information.
A modular computing architecture may allow different software models to operate on a common hardware foundation, but that flexibility introduces configuration and verification requirements. A model developed by a vehicle manufacturer may behave differently from one supplied by another organisation, particularly in processing demand, memory use and responses to uncertain sensor observations.
Aptiv says manufacturers will be able to combine their own AI models with Aptiv, NVIDIA and compatible third-party software. The practical integration must provide suitable interfaces between those elements and preserve the constraints established for the vehicle’s intended automation functions.
The programme is distinct from Aptiv’s June 2026 announcement concerning lifecycle support for NVIDIA Jetson platforms. That earlier work concentrated on production software foundations, security maintenance and support for embedded computing across several industries. The October development adds a defined automotive architecture combining Hyperion, Halos and a system-level cooling approach.
Aptiv plans to show its integrated automated driving capabilities on NVIDIA Hyperion at CES 2027. The demonstration will provide an opportunity to assess the resulting architecture, although measured thermal performance, complete system cost and application-specific validation remain necessary before the claimed integration advantages can be evaluated across production vehicle programmes.
