As Nvidia CEO Huang Renxun embarks on his trip to China this week, the latest deployment of AI applications by leading global technology companies has attracted industry attention. Driven by the current AI frenzy, the prices of electronic components such as memory have skyrocketed, coupled with concerns about energy sustainability, and large-scale application of AI is entering a critical stage of tackling challenges. According to Singapore's Lianhe Zaobao, Huang Renxun's first stop in China is Nvidia's new office in Shanghai. According to multiple sources cited by the media, Huang Renxun went to Nvidia's new office in Shanghai to meet with employees and answer many questions they were concerned about, while reviewing and summarizing the company's major events in 2025. At the previous International Consumer Electronics Show (CES) in 2026, Huang Renxun highlighted Nvidia's comprehensive production of the "Vera Rubin" platform, which is a system level computing platform containing six chips. Through extreme collaborative design, it can significantly shorten model training time and reduce inference costs. He said that Nvidia is reshaping everything about AI, from chips to infrastructure, to models and applications, and 'our job is to create the entire stack'. The core logic of this platform is to solve the "KV cache" (key value cache) problem in AI inference. As AI shifts from the simple learning stage to large-scale inference applications, the amount of data is exploding, and existing GPU and server memory architectures are no longer able to meet the demand. Nvidia has built a massive cache pool by introducing new data processing units (DPUs) and massive SSDs (solid-state drives) in an attempt to break this physical limitation. But this new product will trigger a transfer of production capacity among memory and hard drive manufacturers, significantly compressing consumer level production capacity and causing a surge in spot market prices. Multiple AI related manufacturers are also competing for these components, causing market concerns. In addition to the rise in electronic component prices driven by the application of new AI technologies, energy issues have also attracted increasing attention from all walks of life. Huang Renxun described AI as a "five layer cake" at this year's Davos Forum, covering energy, chip and computing infrastructure, cloud data centers, AI models, and ultimately the application layer. Huang Renxun's metaphor vividly reveals Silicon Valley's concerns: if the energy layer as the "foundation" can only rely on fossil fuels with high price fluctuations and environmental costs, or clean energy restricted by policies, then the upper level chip computing and model applications will lose stability. In January, Google purchased approximately 1.2 gigawatts of carbon free energy from Clearway Energy Group, a clean energy developer and operator, to power its data centers across the United States; Nvidia is committed to using AI to optimize solar and wind power plants... American entrepreneur Elon Musk also stated that the fundamental factor constraining AI deployment is electricity. He stated that the production of artificial intelligence chips is growing exponentially, but the slow growth of power supply is hindering the efficiency of training and deploying AI models in AI data centers. He even announced that SpaceX plans to deploy solar powered AI satellites in the coming years. The process of introducing AI in traditional industrial fields is expected to accelerate this year. Cui Jingyi, Vice President and General Manager of AVEVA Software in China, stated that in 2026, the focus of AI will return to practical issues to a greater extent. The improvement of governance framework will be combined with the continuous pursuit of quantifiable business value. The application of AI in the industrial field will place greater emphasis on input-output, stable performance, and sustainable operation. We are very optimistic about the development of industrial AI in 2026, because industrial AI is not starting from scratch and has become highly mature. Taking our decades of accumulated AI predictive maintenance as an example, such solutions have always been the best practice for creating business value through AI. Now, we are transferring these experiences to more fields such as engineering and design, energy management, quality assurance, and supply chain collaboration. ”She also said that in the technological path, the industry will more commonly adopt a "combination strategy" to promote AI applications: on the one hand, exploring cutting-edge models such as GPT-5 and Gemini Ultra. On the other hand, actively adopting models targeted towards specific industrial tasks. With a thousand questions of common sense DeepSeek、 Taking open source models such as Wenxin and Wudao from China as examples, they emphasize efficiency and achieve strong performance with lower computing power intensity through intelligent architecture and adaptive technology. Supported by large platforms such as Alibaba and Baidu, these models make advanced AI more cost-effective and implementable in real industrial scenarios outside of the laboratory. (New Society)