Jensen Huang delivers keynote at Nvidia GTC AI conference

Jensen Huang, Nvidia CEO, will outline the company’s hardware and software plans at its annual developer conference in San Jose, where he is expected to unveil a next-generation AI chip called Feynman.

The four-day event includes a keynote set for 11 a.m. Pacific Time.

Huang is also likely to discuss data centers, the CUDA programming platform, digital assistants known as AI agents, and robotics.

Industry watchers expect Nvidia to address a market shift from model training to inference as firms such as OpenAI, Anthropic and Meta move to serve hundreds of millions of end users.

Analysts say Nvidia faces growing competition in inference chips from firms such as Groq that specialize in low-latency real-time model serving.

Nvidia’s technology remains widely used across data centers and commercial AI deployments worldwide.

🔗 Source: Reuters

🧠 Food for thought

Implications, context, and why it matters.

Nvidia’s next chip may move past copper

The Feynman platform could arrive in 2027 or later. Reuters calls it a next-generation AI chip that Jensen Huang may discuss, while detailed manufacturing specs remain unconfirmed 1.
Some industry commentary suggests Feynman might swap copper interconnects for silicon photonics, which uses light to move data between chips. Reuters coverage does not confirm this 2.
Groq, an AI chip startup focused on low-latency, real-time model serving, gets framed as an inference rival in some coverage. Reuters says Nvidia licensed technology from Groq in December, while Groq describes the deal as non-exclusive 1.
Groq says founder Jonathan Ross, president Sunny Madra, plus other team members will join Nvidia to help scale the licensed technology. Groq says it will keep operating independently under new CEO Simon Edwards 3.

GTC themes include inference rivalry and sovereign AI, supply chain details remain unclear

Reuters says countries including Saudi Arabia are building custom AI systems for their populations using Nvidia chips. Reuters does not quantify global spending tied to these sovereign AI deployments 1.
Claims that this changes Nvidia’s business model or that governments require sovereign-grade security features remain interpretation, since Reuters does not back them up.
Reuters says analysts expect Nvidia to explain its $2 billion investments in Lumentum and Coherent, which make lasers used in optical connections between chips. Reuters does not say ASML High-NA extreme ultraviolet (EUV) lithography tools are required for Feynman, or that TSMC will make 1.6-nanometer chips for it 1.
Statements that Micron will supply High Bandwidth Memory (HBM)4 that the Feynman architecture needs for performance lack support in Reuters coverage 1.