'Excellent AI advancement': NVIDIA responds to rising popularity of China's DeepSeek

 As the Chinese AI model DeepSeek continues to dominate headlines, tech giant NVIDIA has weighed in on its success. NVIDIA, the leading supplier of AI chips and a key player in the global AI industry, praised DeepSeek’s innovation and explained its broader significance for the future of artificial intelligence. NVIDIA has also been one of the biggest losers after DeepSeek's V3 model was released. 

In comments to Investing.com, NVIDIA acknowledged DeepSeek’s success and said how it showcases the potential of innovative AI techniques.

“DeepSeek is an excellent AI advancement and a perfect example of Test Time Scaling,” NVIDIA said. “DeepSeek’s work illustrates how new models can be created using that technique, leveraging widely-available models and compute that is fully export control compliant.”

NVIDIA also pointed out that while DeepSeek’s cost-efficient approach is making waves, it still relies heavily on NVIDIA GPUs and high-performance networking for inference—the process of running AI models to produce outputs after they’re trained.

What is Test Time Scaling?

NVIDIA’s mention of “Test Time Scaling” refers to an AI technique that optimises the performance of models during their use, rather than focusing only on their training or post-training improvements. According to NVIDIA, DeepSeek’s success represents a new phase in AI development, adding a third “scaling law” to existing methods:

1. Pre-Training Scaling: Building models using vast datasets and powerful hardware.
2. Post-Training Scaling: Fine-tuning models after training to improve performance.
3. Test Time Scaling: Optimising models while they are actively in use to achieve better results with fewer resources.

DeepSeek appears to have leveraged this new scaling approach to create an AI assistant that is both powerful and cost-effective.

Why DeepSeek’s rise matters

DeepSeek has captured global attention by becoming the #1 free app on Apple’s App Store in the U.S., surpassing ChatGPT in downloads. Its AI assistant offers advanced capabilities at a fraction of the cost of competitors, thanks to its efficient use of resources.

DeepSeek’s creators claim the model was trained using NVIDIA’s H800 chips, which are specifically designed to comply with US export controls. The cost of training was reportedly under $6 million, significantly lower than what major AI players like OpenAI typically spend.

This success is forcing the tech industry to rethink the role of cost efficiency in AI development and raising questions about the effectiveness of U.S. export restrictions.

While some have speculated that DeepSeek’s rise could threaten NVIDIA’s dominance in the AI chip market, NVIDIA appears confident that its technology remains essential. DeepSeek’s AI assistant still relies on NVIDIA GPUs for inference, which means that NVIDIA’s hardware is integral to running these models efficiently.

However, the rise of cost-effective AI models like DeepSeek could shift the industry’s focus from expensive, large-scale models to smaller, resource-efficient alternatives. This might reduce the demand for high-end AI chips over time, though analysts suggest it could also lead to an increase in overall AI adoption, benefiting NVIDIA in the long run.