NVIDIA outlines AI & chip strategies for future growth

NVIDIA has outlined its future strategies in chip development and artificial intelligence with comments from its CEO, Jensen Huang, regarding a new direction for industry and technology.

Huang has predicted that all companies will operate two types of factories in the future, one dedicated to manufacturing and the other to mathematics. This vision is supported by NVIDIA's development of future chips using AI-accelerated Electronic Design Automation (EDA) tools, recently optimised for the company's CUDA programming platform.

Expanding its collaborations, NVIDIA has revealed a comprehensive partnership with General Motors. The alliance aims to incorporate AI technologies to enhance GM's car design, improve overall efficiency, and facilitate autonomous vehicle capabilities.

Huang has also introduced the concept of "agentic AI," characterised by new reasoning models optimised for longer thought processes to deliver more accurate responses. Unlike previous models prioritising speed, agentic AI demonstrates an understanding of context and problem-solving capabilities that span multiple modalities, such as interpreting text or video content simultaneously. This approach could better address complex questions beyond the capabilities of earlier AI models like ChatGPT.

Huang believes agentic AI significantly broadens the applicability of AI, potentially leading to progress in developing humanoid robots and autonomous vehicles, where real-world challenges like gravity and friction are factors.

Commenting on the DeepSeek R1 model, Huang addressed misconceptions about its compute requirements. He explained that while DeepSeek's model delivers more accurate results compared to a standard model from Meta, it utilises 20 times more tokens and 150 times the compute resources. This underscores a shift towards increased AI compute requirements, not less, as reasoning models gain prominence.

NVIDIA has also laid out an ambitious roadmap extending to 2027, culminating in its upcoming AI superchip, Rubin Ultra, promising more than 400 times the performance of its predecessor, Hopper. This development occurs against a backdrop of current US export restrictions, which limit China's access to advanced semiconductors and corresponding production equipment. Huang noted that these restrictions create a ceiling on AI compute capacity in China, potentially causing China's AI progress to lag compared to other global advancements.

Further to its semiconductor design, NVIDIA highlighted innovations aimed at addressing the power requirements associated with AI. These include co-packaged optics that enhance power efficiency and the Dynamo virtualization software, which multiplies inferencing performance by optimising workload distribution across GPUs.

Richard Clode, Portfolio Manager on the Global Technology Leaders Team at Janus Henderson, noted, "We continue to believe the power challenges that are needed to advance and run AI will be solved by technology innovation. Therefore, more compelling investment opportunities can be found across the technology stack rather than in utilities and power infrastructure."