# NVIDIA GTC Highlights Preview: "Lobster" Playground Coming Soon, Capital Markets Eagerly Anticipating

The craze of lining up to build “Lobsters” is about to hit San Jose, California. The computing giant NVIDIA will hold its annual GTC conference next week. This year, in addition to launching new products, there will be a special live interactive segment — NVIDIA technical experts will help assemble “Lobsters” on-site.

According to NVIDIA’s official blog, during the GTC conference from Monday to Thursday next week, besides the two-hour keynote speech by Jensen Huang on Monday, GTC Park will host a “Lobster Building” activity throughout the event, allowing attendees to deploy a continuously online AI assistant on-site.

The tech giant also subtly indicated that they hope participants will bring devices powered by NVIDIA chips, including the DGX Spark personal supercomputer, which far exceeds the power (and price) of a Mac Mini. Hardware will be available for purchase on-site, and “Lobsters” can also be deployed directly in the cloud.

Official data shows that over 30,000 people from more than 190 countries and regions worldwide will attend in person next week. Considering many are professional developers, this year’s GTC may not only be a “Lobster Conference” but could even turn into a “Shrimp Fight” event.

What are the other key focus areas?

GTC (GPU Technology Conference) is NVIDIA’s flagship annual event, where the company typically announces new products and discusses industry development visions.

For the capital markets, Jensen Huang’s keynote speech at 11 a.m. local time (2 a.m. Beijing time on Tuesday) will be the highlight.

In terms of hardware, the roadmap for computing chips, potential inference chips, and proprietary optical communication products have become key concepts for investors.

Bank of America senior analyst Vivek Arya summarized in a research report this week that there are three main areas of focus:

1. The latest product roadmap extending to the Feynman architecture;

2. A series of new, collaboratively designed (customized) and modularly split products (such as CPX for inference pre-filling stages and LPU for low-latency decoding);

3. Self-developed optical interconnect technology for large-scale expansion systems (such as switches integrated with CPO).

Arya also anticipates that NVIDIA may discuss the 102.4T Spectrum-6 switch (compatible with Rubin platform) and the 115T Quantum-X switch (using co-packaged optical CPO technology). NVIDIA might even jointly release a customized x86 CPU with Intel to further expand its adoption in enterprise data centers and possibly extend into the consumer CPU market.

Arya also mentioned that he does not expect Jensen Huang to officially provide sales guidance for the next two years, but news about mass production ramp-up of the Rubin architecture could boost the stock price after a period of stagnation.

As for the “unprecedented chips” Jensen Huang previewed, current speculation mainly centers on the LPU jointly released by NVIDIA and inference chip company Groq.

Inference refers to the process where AI models use learned knowledge from training to make decisions and generate responses. Faster, lower-cost inference, distinct from training, is seen as one of the final bottlenecks for large-scale AI deployment. If this chip launches as planned, it would mean NVIDIA, dominant in training markets, will officially step into the competition with custom chips from Google, Microsoft, and others.

Senior equity strategist Kevin Cook from Zacks Investment Research also stated that investors are eager for Jensen Huang to clarify how NVIDIA will handle its relationship with Groq. It is reported that NVIDIA paid $20 billion at the end of last year to license Groq’s patents, and Groq founders Jonathan Ross, President Sunny Madra, and core team members have already joined NVIDIA. This collaboration has attracted widespread attention.

On the software side, there are reports this week that NVIDIA will release an open-source platform aimed at enterprise AI agents, called NemoClaw (which also has a “lobster” connotation). This platform will provide enterprises with a structured way to build and deploy AI agents capable of autonomously executing multi-step tasks.

“Jensen Huang Roundtable” Focuses on Open Models

As a recent GTC tradition, Jensen Huang will hold a roundtable discussion next week to showcase NVIDIA’s most important industry directions.

This year’s theme is “The Current State and Future of Open Models.” Industry newcomers such as AI programming company Cursor, AI search engine Perplexity, healthcare AI assistant OpenEvidence, and former OpenAI CTO Mira Murati, who founded the Thinking Machine Laboratory, will attend.

NVIDIA’s announcement states that over the past year, one of the most significant changes in AI has been the rapid progress of open frontier models. Open innovation is accelerating advancements across companies and industries, clearly demonstrating that AI will be ubiquitous. Jensen Huang will have frank discussions with these industry leaders about “the latest developments and future prospects of open frontier models.”