Nvidia unveils new AI products that it says can help DeepSeek’s models run faster

Nvidia’s new offerings could significantly boost the performance of DeepSeek’s R1 model, the chip giant’s CEO Jensen Huang said on Tuesday, in a message aimed at highlighting the US firm’s irreplaceable position in the era of artificial intelligence (AI)

At the company’s annual GTC conference in San Jose, California, Huang explained how Nvidia’s latest products could dramatically boost reasoning capabilities for large models like DeepSeek’s R1.

The CEO described Dynamo, the company’s new open-source inference software, as the “operating system of an AI factory.” Dynamo, now available on Github, can deliver up to 30 times greater performance on existing graphics processing units (GPUs) and architectures for AI reasoning tasks, according to Nvidia.

The dominant role of Nvidia as the sole “arms supplier” in the global AI battle was cast into doubt in January, when Hangzhou-based DeepSeek unveiled its low-cost, high-performance models, which raised questions about the need to buy expensive AI chips.

Nvidia, which is not allowed to sell its advanced chips to China owing to US export restrictions, has been trying to mitigate concerns about “scaling law” – which states that AI systems improve when more resources are used to develop them – following the success of DeepSeek.

Nvidia’s shares fell 3.4 per cent to US$115.4 on Tuesday, lower than its price on January 27 when the stock lost 17 per cent in a single trading day.