On the morning of the 18th local time, Nvidia CEO Huang Renxun gave a 2-hour keynote speech at the Nvidia GPU Technology Conference (GTC) held in San Jose, California, revealing the company's latest progress and future goals in fields such as artificial intelligence and robotics. In this live speech, Huang Renxun unveiled Nvidia's chip roadmap. The next generation GPU chip Blackwell Ultra will be launched in the second half of this year, claiming that its AI processing performance will be improved by 1.5 times, with better AI inference performance and energy efficiency. He also stated that Nvidia will release a chip system called Vera Rubin in the second half of 2026, which will replace Blackwell and have faster speeds.
Analysts have stated that the performance improvement of Nvidia's Blackwell chip in inference far exceeds that in training. The Financial Times believes that Nvidia's newly released artificial intelligence chip marks the company's bet on the arrival of "inference" AI systems such as DeepSeek, which will stimulate greater computing demand.
According to Nikkei Asia Review, the successful development of a high-performance AI model using fewer AI semiconductors by Chinese startup DeepSeek in January led to a sharp drop in Nvidia's stock price. The market's concerns about the sluggish growth of demand for AI semiconductors have heated up.
In this year's speech, Huang Renxun defended Nvidia's leading position in selling expensive AI chips to customers. Almost the whole world got it wrong, "Huang Renxun said, stating that the demand for AI computing is actually growing at a rapid pace. According to CNBC website, Huang Renxun provided some data showing the growth of AI computing demand among the four major cloud service providers in the United States. In 2024, Nvidia shipped 1.3 million Hopper architecture GPUs to these four companies. Since 2025, the company has shipped 3.6 million Blackwell architecture GPUs to these customers. Huang Renxun said, "In just one year, you can see amazing growth in AI infrastructure
In addition to chips, the company is also targeting AI in multiple fields such as robotics, 6G, and even automobiles.
During the speech, Huang Renxun took the stage with the small robot Blue, which carries Nvidia's world's first open-source and fully customizable basic model GR00T N1, reportedly empowering general-purpose humanoid robots to achieve reasoning and various skills. Nvidia is also collaborating with Google DeepMind and Disney to develop a robot platform called Newton. In addition, in the automotive field, Huang Renxun revealed that Nvidia is collaborating with General Motors to use AI in the new generation of cars, factories, and robots.
Nvidia also announced that it will collaborate with leading companies in the telecommunications industry to jointly develop 6G wireless technology based on artificial intelligence. The company said that the next generation wireless network must be fundamentally combined with artificial intelligence to seamlessly connect hundreds of billions of mobile phones, sensors, cameras, robots and autonomous vehicle. AI Native wireless networks will provide enhanced services to billions of users and set new standards in spectrum efficiency.
Although Nvidia is ambitious in the development of artificial intelligence wireless access networks, there is still a long way to go in its efforts, "said Bi Qi, Chief Scientist of China Telecom Group and a member of Bell Labs in the United States, in an interview with Global Times on the 19th." The correlation between the input of RAN in wireless networks is far lower than that of language and images. What kind of AI model is more suitable without meeting the strong correlation conditions is currently an unanswered scientific research direction. Of course, such exploration is meaningful and requires attention to whether industry giants such as Qualcomm, Ericsson, and Huawei will deeply participate.
Some analysts believe that investors are looking forward to new revenue growth points, but areas such as robotics and quantum computing mentioned by Huang Renxun are unlikely to bring substantial returns in the short term. An AI expert told Global Times reporters that Nvidia has demonstrated a full stack monopoly capability from training to inference, hardware to software. A few hours before Huang Renxun's speech, Su Zifeng, the chairman of another chip giant, AMD, appeared in Beijing to introduce its layout in the AI field and showcase more products that are closer to consumers. It is worth noting that AMD is actively embracing the Chinese market. According to public reports, Su Zifeng revealed that AMD embraced and supported DeepSeek in the first place, and AMD engineers are still focusing on the continuous optimization of DeepSeek. Su Zifeng also emphasized that AMD will work with partners and communities to promote the development of China's AI PC ecosystem.
Liu Shaoshan, director of the embodied intelligence center at Shenzhen Institute of Artificial Intelligence and Robotics, paid attention to the GTC conference and Huang Renxun's speech. He told Global Times reporters that the Chinese market occupies a prominent position in the narrative of this GTC conference, DeepSeek The large model has become a core case for Nvidia to demonstrate breakthroughs in inference performance, with its 671B parameter model achieving a 36 fold throughput jump on the Blackwell platform. We need to understand to what extent the current state-of-the-art computing infrastructure will evolve, while considering our own next steps of development. But it should be emphasized that learning and benchmarking do not mean blindly following. China needs to find its own path in computing power. DeepSeek's innovation is already a precedent
(Editer：admin)
