【AI+NVDA】NVIDIA GTC Conference Unveils March 16 - What Are the Key Focus Points of Jensen Huang's Keynote? Beyond Chips, Is It Also Catching the "Lobster Farming" Trend?

Nvidia’s GTC 2026 Annual Technology Conference will be held in San Jose, California, from Monday, March 16 to Thursday, March 19. CEO Jensen Huang will deliver a keynote speech at 2 a.m. Hong Kong time on Tuesday, March 17, covering updates and plans for AI products, new chips, AI factories, open models, intelligent agents, physical AI, and future AI industry development directions.

Watching the Feynman Architecture: New AI Inference Chips

In hardware, the market is looking forward to Nvidia’s full-stack roadmap update from Rubin to Feynman. It is reported that the next-generation architecture chip codenamed “Feynman” uses TSMC’s most advanced 1.6-nanometer process (A16). Communication between chips will be replaced by optical communication instead of traditional metal wiring, reducing AI data center communication energy consumption by over 70%.

Meanwhile, there are rumors that Nvidia will release a dedicated inference chip at GTC, designed to accelerate model inference rather than training. If true, this chip would represent Nvidia’s latest effort to not only dominate the current training market, which accounts for about 80% of the market share, but also to compete more fiercely in the inference market.

In an interview with foreign media in February, Jensen Huang said, “We have an unprecedented chip that will be a complete surprise.”

Planning to Launch Enterprise-Level Open-Source AI Agent Platform Similar to OpenClaw

On the software side, Huang emphasized that intelligent agents will become the next major driver for inference demand. With the recent popularity of OpenClaw, as everyone is “raising lobsters,” Nvidia is also riding the “lobster-raising” trend. It is rumored that Nvidia will launch NemoClaw, an enterprise-level open-source AI agent platform similar to OpenClaw. This platform will provide a structured way for enterprises to build and deploy AI agents, enabling Nvidia to emulate products offered by companies like OpenAI.

Notably, on the Nvidia GTC webpage, it states, “Enter GTC Park to personally create a dedicated AI assistant powered by OpenClaw that operates around the clock,” along with an image of Jensen Huang sewing a lobster plush toy in chibi style.

▲【AI+NVDA】Nvidia GTC Conference kicks off on 3/16. What are the key points of Jensen Huang’s keynote? Besides chips, is Nvidia also catching the “lobster-raising” trend?

Additionally, the market is paying attention to Nvidia’s network roadmap, including more details on the Co-Packaged Optical (CPO) technology, as well as information on the Quantum-X (infinite bandwidth) and Spectrum-X Ethernet co-packaged optical switches, and potential supply chain partners.

Jensen Huang’s speeches are always highly anticipated. As AI competition intensifies and products become more diverse and complex, the market may focus more on whether Nvidia can leverage GTC to transform itself from a “chip-selling” hardware company into a platform company that “defines AI infrastructure rules.”