Nvidia CEO Jensen Huang defends pricey AI chips after DeepSeek's success

Photo : Reuters

Nvidia CEO Jensen Huang has strongly defended the need for the company’s costly AI chips after some people questioned their value. This came after the success of Chinese AI company DeepSeek, which used cheaper chips to train its models. Speaking at Nvidia’s yearly GPU Technology Conference, Huang said many people did not understand how fast the demand for powerful AI hardware is growing.

Stay informed with the latest news. Follow DXB News Network on WhatsApp Channel  

“Almost the entire world got it wrong,” Huang said on stage. He called the event the “Super Bowl of AI” and explained why stronger AI chips are still important. He said that the fast growth of agentic AI, which can do complex tasks with little human help, now needs much more computing power.

“The amount of computing we need for agentic AI and reasoning is now 100 times more than what we thought last year,” Huang said. He explained that these new AI systems need more powerful hardware, making Nvidia’s expensive chips still necessary.

Recently, DeepSeek made headlines by using older, cheaper chips to train its language models. This made people wonder if Nvidia’s costly GPUs were still important. DeepSeek also showed a new trend in AI: companies are now focusing more on "inference" rather than just large-scale training. Inference is when AI models give real-time answers based on what they have already learned, which uses less computing power than the training phase.

Even with these doubts, Nvidia’s AI chips remain very important to the company. Last year, Nvidia earned $130.5 billion in sales, mostly from selling high-end GPUs. Some of these chips sell for tens of thousands of dollars each. However, despite Huang’s strong defense, Nvidia’s stock fell. The company’s shares dropped by 3.4% on Tuesday, showing that investors were not fully convinced.

To answer concerns about DeepSeek, Huang introduced Nvidia’s new open-source software called Dynamo. He said Dynamo could make AI models like DeepSeek’s R1 much faster. According to Ian Buck, Nvidia’s Vice President, Dynamo can boost performance by 30 times using the same number of GPUs and setup.

"Dynamo can deliver 30 times more speed on the same GPUs for models like DeepSeek," Buck explained, showing how Nvidia plans to stay ahead of its rivals.

The event showed that Nvidia is determined to remain the leader in AI hardware, even as competitors like DeepSeek explore cheaper AI options.