Nvidia’s CEO Unveils AI Advancements at GTC
Nvidia, the dominant force in the AI chip market, is actively defending its position. At Tuesday’s GTC keynote address, CEO Jensen Huang outlined the company’s ambitious plans. While the presentation was packed with new hardware and software announcements, the market’s reaction was mixed. Despite a positive roadmap, Nvidia’s stock experienced a slight downturn, reflecting the current volatility in tech stocks.
Jim Cramer noted that Huang presented a “terrific roadmap,” but the stock dipped 3.4% on the day. After a recent rally, the stock has decreased 5% in the last two days. Concerns regarding chip sales restrictions, tariffs, and competition contribute to this volatility. Huang’s announcements, made in front of what has been called “Woodstock for AI,” weren’t enough to alleviate these concerns immediately. As the AI trade matures, market expectations become more demanding.
The Competitive Landscape
The emergence of Chinese startup DeepSeek’s more efficient AI model in late January caused some market uneasiness as major tech companies, Nvidia’s clients, considered their spending plans. However, these clients have confirmed their 2025 capital expenditure outlooks, solidifying Nvidia’s position.
As Cramer stated, “In this market, nobody wants to wait for what’s around the corner.” Despite this, Nvidia’s place as the leading AI player remains secure, even as its share price remains below pre-DeepSeek levels.
Key Announcements at GTC
Huang’s presentation revealed crucial updates and unexpected offerings.
Major Hardware Developments
One of the key announcements involved the updated version of Blackwell, Nvidia’s cutting-edge AI chip platform. The new Blackwell Ultra, set to be released in the second half of 2025, will boast increased memory capacity. Memory is critical because of the massive amounts of data involved in AI computing. The full data center rack version – using 72 GPUs and 36 CPUs – will deliver 1.5 times more AI performance compared to the current rack-scale Blackwell product.
Following initial issues with Blackwell’s rollout, Huang emphasized the company’s smooth transition to Blackwell Ultra. He then shared details about Vera Rubin, the architecture slated to succeed Blackwell in the second half of 2026. This includes a next-generation Vera CPU, which is expected to be twice as performant compared to Grace. The Rubin GPU will also incorporate HBM4, the latest high-bandwidth memory. The full server rack version of Vera Rubin will deliver 3.3 times the compute performance of Blackwell Ultra. In the second half of 2027, Nvidia plans to release Vera Rubin Ultra.
Nvidia also announced a new networking product designed to enhance communication between GPUs in data centers. This product will use silicon photonics to improve data transfer speeds and connectivity.
Software Innovation
Perhaps the most surprising announcement was the introduction of Nvidia Dynamo, a software designed for AI computing’s “inference” stage. This software underscores Nvidia’s commitment to its software suite, providing efficiency gains for the emerging reasoning models. This announcement reflects the company’s increasing presence in inference computing.
Additionally, Huang revealed the name of the architecture that will follow Vera Rubin: Feynman, scheduled to debut in 2028. This continues Nvidia’s tradition of naming its chips after influential scientists. Huang emphasized the importance of providing a transparent roadmap, as new product planning has significant complexities. The company announced DGX Station, a new high-performance desktop computer designed for AI developers, data scientists, and researchers.
Partnerships and Collaborations
Nvidia highlighted collaborations at GTC, including ties with General Motors on advanced driving systems. They also announced partnerships related to robotics and drug discovery with Alphabet subsidiaries as well as CrowdStrike.
