Nvidia CEO dismisses DeepSeek AI cost-cutting fears

Nvidia CEO Jensen Huang dismissed concerns that low-cost AI models like Deepseek could sink the demand for its AI chips.

He argued that new AI model types that generate complex answers will only drive the need for more computing infrastructure.

Speaking at Nvidia’s GTC Conference in California, Huang even described concerns over Deepseek’s R1 model impact as misplaced, saying, “The understanding of R1 was completely wrong. Computation demand is much higher.” 

Nvidia expects its clients to continue buying their AI gear even if the economy falls into a recession

Chinese AI startup Deepseek’s launch raised fears among AI chip companies when it introduced a high-performing AI model at lower prices. Some even suggested that the new model could usher in a new era of fewer chips and servers.

However, Nvidia’s largest customers reaffirmed their spending plans since then, alleviating some of the company’s fears.

At the company’s GTC conference on Wednesday in San Jose, California, Jensen Huang even stated that new AI models could increase demand for computer infrastructure, dismissing fears that Deepseek could affect its AI chip business.

He also expects that companies will continue to buy their AI products, even if the US economy recedes, believing that even the proposed tariffs will have a small impact on their firm.

According to Bloomberg Intelligence, even the largest data centre operators are spending much faster than anticipated on AI computing resources.

Moreover, Nvidia’s shares are up by close to 2.5% in New York trading, though they slightly dipped by 14%.

Nvidia’s Huang argues that big firms need to invest in better chips rather than developing cheaper, low-performing ones.

Nvidia has been pushing for more corporations to invest in its AI products; however, it’s still unclear whether its customers will continue to spend largely on AI infrastructure.

Some of Nvidia’s customers have already begun developing their own components,  chips that might replace his AI accelerators in data centres. In partnership with Broadcom, Google has been working on application-specific integrated circuits (ASICs). 

However, Huang has claimed that ASICS are designed but not actually used in data centres and, therefore, not much of a competition to its own chips, commenting, “It’s a different calculus.”

He even added that big firms need to develop better chips to earn more revenues rather than cheaper chips to cut costs.

Huang also claimed that they sold more Blackwell chips than Hopper-based ones, arguing that spending for AI infrastructure is still rising.