TMTPOST -- Nvidia Corporation CEO Jensen Huang tries to quell people’s fears of artificial intelligence (AI) boom amid sudden rise of Chinese startup DeepSeek.
Credit：Nvidia
For Huang, the concerns that DeepSeek’s AI model R1 could reduce demand for chips and powerful servers for such software were misplaced. “The understanding of R1 was completely wrong, Computation demand is much higher,” Huang spoke with investors and analysts at Nvidia’s annual GTC conference on Wednesday.
In an interview with CNBC the same day, Huang called DeepSeek-R1 “a fantastic thing” because R1 is the first open-source reasoning model. “What makes R1 incredible is that it reasons. That’s why the answer is so good and it breaks the problem down step by step,” said Huang.
Huang signaled the huge potential of computing demand reasoning models like R1 are bringing to the market, thus a big positive for Nividia, the world’s most valuable AI chipmaker benefiting the AI frenzy ignited by OpenAI’s ChatGPT. “This reasoning AI consumes 100 times more compute than a non-reasoning AI,” he said. “It was exactly the opposite, it was the exact opposite conclusion that everybody had. It actually consumes 100 times more computing.”
It was reported Google and other companies have been working with Broadcom Inc. to develop their application specific integrated circuits (ASICs), which could replace Nvidia’s AI accelerators in data centers and reduce their dependence on the company. When asked about customers’ efforts to develop their in-house chips, Huang told analysts many ASICs are not always actually to be deployed in data centers.
What those big customers need are better chips that can generate more revenue from their infrastructure, instead of cheaper ones to save costs, Huang argued. “All of those companies are run by great CEOs who are really good at math,” he said. “The effects are not just cost. It’s a different calculus.”
Huang expected these companies would even shift more of their investments to AI gear if the U.S. economy head into a recession, because that move bets on where their business growth relies on.
Huang said in the CNBC interview that U.S. President Donald Trump’s tariffs won’t pose any meaningful impact in the near term since Nvdia is working with contractors like Taiwan Semiconductor Manufacturing Co. (TSMC) to move production to the United States. “We’ve got a lot of AI to build ... AI is the foundation, the operating system of every industry going forward. ... We are enthusiastic about building in America,” Huang said. “Partners are working with us to bring manufacturing here. In the near term, the impact of tariffs won’t be meaningful.”
Huang in his keynote at the GTC on Tuesday lauded DeepSeek’s R1 as a "world-class reasoning AI model" and an "excellent innovation" due to its open-source nature, highlighting its potential to push the industry towards more resource-intensive reasoning models.
For Huang, AI is experiencing a shift to agentic and reasoning and that means “the amount of computation necessary to train those models, and to inference those models, has grown tremendously.”
“And for good reason, because there is an inflection point in AI, the amount of computation we have to do in AI is so much greater as a result of reasoning AI, and the training of reasoning AI systems and agentic systems,” stated Huang. He anticipated the world will need 100 times more computing power for advanced AI than it deemed necessary a year ago.
On an earnings call last month, Huang hailed DeepSeek as ”excellent innovation”. “DeepSeek-R1 has ignited global enthusiasm. It's an excellent innovation. But even more importantly, it has open sourced a world-class reasoning AI model. Nearly every AI developer is applying R1 or chain of thought and reinforcement learning techniques like R1 to scale their model's performance,” Huang told analysts on an earnings call on February 27.
While investors become more wary of the shock from DeepSeek’s new models built on lower-cost Nvidia chips. The AI chip giant maintained upbeat on the strong demand for the ongoing AI development.
On the conference, Huang saw future reasoning models consume much more compute, stating “we're at the beginning of reasoning AI and inference time scaling”, and Multimodal AI, enterprise AI, sovereign AI, and physical AI are right around the corner. Looking forward, data centers will dedicate most of capital expenditure (Capex) to accelerated computing and AI Data centers will increasingly become AI factories, according to Huang.
A Bloomberg Intelligence report projected earlier this week the biggest tech companies will ramp up their combined annual AI spending to more than $500 billion by early next decade, driven in part by a newer approach to AI from DeepSeek and OpenAI.
Reasoning models offer new opportunities to make money from software, and potentially transfer more of the cost from the development stage to after the model is rolled out, according to the report. That’s likely to drive greater investment in this approach, and boost spending on AI overall, the report said. The immense amount of attention on DeepSeek will likely push tech firms to “increase investments” in inference, making it the fastest-growing segment in the generative AI market, per the report.
更多精彩内容，关注钛媒体微信号 （ID：taimeiti），或者下载钛媒体 App
