TMTPOST -- Nvidia Corporation CEO Jensen Huang believes DeepSeek is a great innovation that has ignited a new round worldwide frenzy on artificial intelligence (AI) following OpenAI’s chatbot ChatGPT more than two years ago, while the most important contribution that the Chinese startup brought to the world is its open-source reasoning AI model.

Credit：Nvidia

“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 Thursday.

DeepSeek stunned Wall Street and Silicon Valley late January as it claims performance of its reasoning model R1 comparable to leading offerings like OpenAI’s o1 at a fraction of the cost. It took just $5.58 million for DeepSeek to train its V3 large language model (LLM). The startup said it used 2,048 Nvidia H800 chips, a downgraded version of Nvidia’s H100 chips designed to comply with U.S. export restrictions. DeepSeek’s power implications for AI training punctures some of the captical expenditure (CapEx) euphoria, commented Jefferies. The brokerage firm believes there are potential negative implications for the builders, as pressure on AI players to justify ever increasing CapEx plans could ultimately lead to a lower trajectory for data center revenue and profit growth.

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.

Huang sees 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 CapEx to accelerated computing and AI Data centers will increasingly become AI factories, according to Huang.

When asked about whether DeepSeek and other innovations have changed his confidence in the strong demand that can sustain into next year, Huang noted AI startups need a lot of computing power.

“We have a fairly good line of sight of the amount of capital investment that data centers are building out towards,” Huang said.”We know that going forward, the vast majority of software is going to be based on machine learning. And so accelerated computing and generative AI, reasoning AI, are going to be the type of architecture you want in your data center.”

Moreover, Huang pointed out there are many innovative startups that are still coming online as new opportunities for developing the next breakthroughs in AI, whether it's agentic AIs, reasoning AIs, or physical AIs. “The number of startups are still quite vibrant, and each one of them need a fair amount of computing infrastructure.,” said Huang.

Nvidia announced on January 30 that the 671-billion-parameter DeepSeek-R1 model is available on as an NVIDIA NIM microservice preview. The DeepSeek-R1 NIM microservice can deliver up to 3,872 tokens per second on a single NVIDIA HGX H200 system, based on Hopper architecture, the predecessor of Blackwell. Nvidia on Monday introduced DeepSeek-R1 optimizations for Blackwell, claiming they can deliver 25 times more revenue at 20 times lower cost per token, compared with NVIDIA H100 just four weeks ago.

The recent financial results showed Nvidia sales further slowed down for its fourth quarter of the fiscal 2025 year ended January 26, 2025, but still beat Wall Street estimates. Revenue from Nvidia’s top business Data Center soared 98% year-over-year (YoY) to a record of $35.6 billion. The latest generation Blackwell-architecture AI chips recorded sales of $11 billion for their first quarter of delivery, exceeding the company’s anticipation. Blackwell sales were led by large cloud service providers which represented approximately 50% of our Data Center revenue, the Chief Financial Officer (CFO) Colette Kress commented.

Countries across the globe are building their AI ecosystems and demand for compute infrastructure is surging, and geographically speaking, sequential growth in Nvidia’s Data Center revenue was strongest in the US., driven by the initial ramp of Blackwell, according to Kress. She noted Data Center sales in China remained well below levels seen on the onset of U.S. export controls. Absent any change in regulations, Nvidia expected that shipments will remain roughly at the current percentage in China where witnessed intense competition. The company vowed to continue to comply with export controls while serving its customers in the country.