Nvidia CEO reveals new ‘reasoning’ AI tech for self-driving cars

Nvidia CEO Jensen Huang has introduced a groundbreaking AI technology called Alpamayo, designed to enhance the cognitive abilities of self-driving cars. Unlike traditional autonomous systems that rely primarily on reacting to historical data patterns, Alpamayo enables vehicles to perform chain-of-thought reasoning, allowing them to interpret and respond to complex, unpredictable road scenarios such as sudden roadworks or erratic driver behavior. This advancement aims to bring a more human-like understanding and decision-making process to autonomous vehicles, improving safety and adaptability in real-world environments.

During his keynote at the Consumer Electronics Show (CES) in Las Vegas, Huang emphasized that this development marks a pivotal moment for physical AI, akin to the impact of ChatGPT in conversational AI. He highlighted that Alpamayo not only helps vehicles navigate rare and challenging situations but also allows them to articulate the reasoning behind their driving decisions. This transparency is expected to build trust and facilitate safer, scalable autonomy in robotaxis and other autonomous transport solutions.

Nvidia has already begun integrating this technology into a driverless version of the Mercedes-Benz CLA, developed in partnership with the German automaker. This vehicle, which will launch in the US soon followed by Europe and Asia, demonstrates Alpamayo’s capabilities by driving naturally through urban environments like San Francisco. Notably, the car can operate without human intervention, with passengers able to sit safely without controlling the steering wheel, as the system learns directly from human driving demonstrations and continuously explains its actions.

In addition to Alpamayo, Huang announced the production of Nvidia’s next-generation Vera Rubin chips, slated for release later this year. These chips promise a fivefold increase in computing power over previous models, significantly enhancing the performance of AI applications such as chatbots. The Vera Rubin platform comprises six distinct chips and will be featured in flagship servers equipped with 72 graphics processing units and 36 new central processors. These can be combined into large-scale "pods" containing over 1,000 Rubin chips, boosting the efficiency of AI token generation by a factor of ten.

The improved performance is attributed to a proprietary data format developed by Nvidia, which the company hopes will gain widespread industry adoption. While Nvidia continues to lead in AI model training, it faces mounting competition from traditional rivals like AMD and even its own customers, including Alphabet's Google, particularly in deploying AI-powered services to millions of end users. The advancements in both autonomous vehicle reasoning and chip technology underscore Nvidia’s strategic focus on maintaining its edge in the evolving AI landscape.

Overall, Nvidia’s Alpamayo technology represents a significant step forward in autonomous vehicle intelligence, enabling cars to think, reason, and communicate their decisions much like human drivers. Coupled with the powerful Vera Rubin chips, these innovations are set to accelerate the adoption of safe, scalable AI solutions across transportation and other sectors.