Nvidia’s CEO, Jensen Huang, has responded to recent doubts regarding the company’s leadership in the AI chip sector, specifically countering assertions that rivals such as China’s DeepSeek have reached similar AI capabilities with far less hardware. During a significant event for software developers, Huang stated that the prevailing assumptions about the computational power required for AI have been fundamentally incorrect.
“Nearly everyone has misjudged the situation,” Huang remarked at the Nvidia GPU Technology Conference (GTC), which he referred to as “the Super Bowl of AI,”. He noted, “The computational requirements stemming from agentic AI and reasoning are easily 100 times greater than we anticipated just a year ago,” alluding to AI agents that can perform routine tasks with minimal human oversight.
Huang underscored Nvidia’s readiness to adapt to these evolving requirements, emphasizing the critical roles of speed and scale in AI inference. He stated, “If the response time to a query is too lengthy, customers will not return. This is akin to web search.”
In addition, Nvidia unveiled a new series of chips, including the next-generation GPU, Blackwell Ultra, which is scheduled for release in the latter half of this year. This chip will feature enhanced memory capabilities compared to the existing Blackwell model, allowing for the support of larger AI frameworks.
Following Blackwell, the Vera Rubin chip system is anticipated to launch in the second half of 2026, promising even greater processing speeds. Nvidia is also committed to advancing its chip development with the introduction of the Feynman chips, expected to debut in 2028.
These innovative chips are engineered to optimize AI systems for both swift and high-volume responses. Huang contended that Nvidia’s chips possess a unique ability to deliver on both fronts, which is essential for ensuring user satisfaction and loyalty.
