Nvidia CEO says AI models like DeepSeek's R1 need 100 times more compute

NVIDIA Since Deepseek

The company once again exceeded even the highest revenue expectations, but investors reacted with restraint. The biggest question facing Nvidia since DeepSeek launched its extremely effective open-source models last month has been whether the Chinese company's efficiently trained models will reduce the demand for AI computing.

Since DeepSeek's most impressive model, R1, was a reasoning model, Huang's answer was a clear "no."

"Logic models can consume up to 100 times more computation. It may require a lot more computation to do reasoning in the future," he said during a phone conversation.

Huang called DeepSeek "a great innovation."

"But more importantly, they have open-sourced a world-class AI reasoning model. Almost every AI developer is using R1 or chain-of-thought and reinforcement learning methods like R1 to enhance the performance of their models," Huang said on Wednesday.

Huang spoke to Nvidia's falling stock price in an interview last week, saying investors had misinterpreted the DeepSeek event.

Cloud service providers also previously told BI that demand for Nvidia's most powerful chips will keep growing.

Synovus analyst Dan Morgan said in a note to investors that DeepSeek's long-term effect has been to boost "inference-intensive models," referring to resources such as chips and power for inference -- the type of AI computing that refines models and generates reasoning and answers to questions.

"Most of our computing today is really based on inference, and Blackwell will take all of that to the next level," Huang said Wednesday, referring to the company's latest generation of chips.

The Competitors

While Nvidia still holds the biggest share of all AI computing markets, analysts are starting to see a world where that's less certain.

"Competition is starting to affect Nvidia's position, although it is not very significant at the moment," said Lucas Keh, an analyst at Third Bridge, after Nvidia's earnings report.

Nvidia's competitors have been targeting this chip for several years, as it is expected to become a large market in the long term. Indeed, the number of new mining chip companies and backers has increased in recent months. Chip startup Tentorrent announced nearly $700 million, while Etched announced new funding of $120 million last year.

Investors are also growing concerned that custom AI chips ordered by cloud companies such as Google and Amazon could take away Nvidia's lead, especially in the area of ​​inference.