DeepSeek uses more power than you think, NVIDIA CEO says

The arrival of DeepSeek caused a stir in the artificial intelligence industry. However, the biggest hit did not go to rival AI services. The most affected was, ironically, NVIDIA. The company’s shares crashed sharply in just 24 hours. That said, NVIDIA’s CEO says people are misunderstanding DeepSeek’s true computing requirements.

The “DeepSeek effect” was mainly due to the company’s claims about its investment in training. The Chinese firm said it spent around $6 million on its R1 model. This is just a fraction of what other major AI companies have invested. NVIDIA, as the world’s leading provider of AI hardware, took the biggest hit. Investors began to doubt the true computing power required for AI development. Massive stock sell-offs reflected this, resulting in a $600 billion drop in the company’s valuation in just 24 hours.

NVIDIA CEO claims that DeepSeek R1 “consumes 100 times more compute than a non-reasoning AI“

That said, industry experts subsequently began to question DeepSeek’s actual spending on AI. They even estimated that the Chinese company had actually invested around $1.6 billion. The latest statements by Jensen Huang, CEO of NVIDIA, go in a similar direction.

In an interview with CNBC, Huang claimed that the DeepSeek R1 model actually “consumes 100 times more compute than a non-reasoning AI.” “It was exactly the opposite, it was the exact opposite conclusion that everybody had” he added. DeepSeek R1 is the company’s reasoning-focused model. It was designed to compete against GPT-o1. These types of models take longer to respond but offer a much more complete and accurate output. It is especially useful for research areas on complex topics.

Despite the supposedly high computing power requirements, NVIDIA’s CEO had good words for DeepSeek. He stated that DeepSeek R1 is “fantastic” as it is “the first open-sourced reasoning model.” Let’s remember that the AI ​​model’s code is freely available for download. Therefore, anyone with the necessary hardware resources could run it locally.

NVIDIA focused on more projects

Huang also teased some of the projects NVIDIA is working on. The company’s CEO stated that they have new AI infrastructure for robotics and business in their hands. He also said that interest in AI has shifted from the generative sector to reasoning models. Lastly, he projects that global investment in computing will reach a trillion dollars by the end of the decade. Of course, most of those resources will be directed toward AI development.

The post DeepSeek uses more power than you think, NVIDIA CEO says appeared first on Android Headlines.