NVIDIA (Nvda) CEO Jensen Huang does not sound the alarm on the introduction of models of AI like the deep deep R1 depth of China. Instead, the chip executive said that the world needs much more computing power to manage so-called reasoning and agenic AI applications than previously expected.
“Last year, it was there that almost the whole world was wrong,” Huang said during his speech at the GTC 2025 in Nvidia.
“The requirement for calculation, the AI scale law is more resilient and, in fact, hyperaclealée. The quantity of calculation we need at this stage following agentic AI, following reasoning, is easily a hundred times more than we needed this time last year,” he explained.
The agency AI is a type of AI that can take measures in the name of a user. Reasoning or reflection on AI is a type of AI that imitates the way humans think by decomposing the problems step by step to find the best response to a user query.
Deepseek’s R1 came on stage at the end of January and sent Wall Street to a merger when the company said that the reasoning model corresponded to the capacities of the Openai model and that it formed its wider V3 Deepseek model for around 5 million dollars, compared to the tens of millions of silicons spent comparable models.
Find out more: How does Nvidia earn money?
Investors went to streaming for the doors, sending Nvidia’s market value Hunt nearly $ 600 billion On the fears that Cloud companies would no longer need to spend billions on Nvidia fleas.
More recently, NVIDIA was shaken by concerns about President Trump’s pricing threats and the potential for the United States to impose new export controls on china.
NVIDIA action is down 14% over a year, but still increasing 30% in the last 12 months.
While the prices and export controls are largely out of the hands of Nvidia, apart from lobbying for exceptions, the company is fully able to face these annoying deeksek concerns, and Huang did exactly this during his speech.
Throughout the window of about two hours, the director general explained how the reasoning models will benefit from fleas like the new ultra blackwell of the company and Vera Rubin Superchip. This, he explained, will only continue in the future, because physical AI, including humanoid robots and autonomous cars, it matures on the market.
The CEO has also widened its plans for Cuda software in Nvidia, which allows developers to take advantage of the company’s fleas for general treatment, its omniverste simulation platform and a litany of other services.
