Today Bloomberg exploded with a bombshell: China is planning to invest about 2 trillion yuan (~295 billion US dollars) in the next five years to build a nationwide data center network. This number may not feel much when viewed alone. But by comparison, the total AI capital expenditure of Meta, Microsoft, Google, and Amazon for the entire year of 2026 is $725 billion. China's 2 trillion yuan will be spread over five years, with an average of about 59 billion US dollars per year, which is less than the annual level of American giants. As of March 2026, the daily average token usage in China has exceeded 14 trillion, an increase of over 1400 times compared to the 100 billion at the beginning of 2024. In the week of the openclaw outbreak in early February 2025, the weekly token call volume of Chinese AI models reached 4.12 trillion, surpassing the United States for the first time. China's AI is not about one company being strong, it's about the entire chain being rapidly formed, from electricity, to computing power, to models, to applications. The AI capital expenditure of American tech giants in 2025 will reach $370 billion, a year-on-year increase of 48%. The Stargate project has a four-year plan of 500 billion US dollars. The scale of AI related debt has reached~1.2 trillion US dollars. The quarterly return pressure on Wall Street has turned AI investment into an unstoppable gamble, and stopping it will lead to a decline in the balance sheet. China is taking a completely different path. The 2 trillion yuan plan reported by Bloomberg has several key points: 1. State owned telecommunications giants such as China Mobile and China Telecom bear the majority of the operation and interconnection of data centers 2. The blueprint stipulates that at least 80% of AI chips and other technological equipment must use products from local suppliers such as Huawei. Just like building highways and high-speed railways in the past, what is being built now is a computing power highway in the digital age. Huawei Ascend chips have announced an iterative roadmap for the next three years, from the 950 series in 2026 to the 970 series in 2028, with computing power continuing to climb. The release of the Ascend 950PR accelerator card and Atlas 350 AI server in Q1 2026 marks the beginning of the Ascend ecosystem's large-scale implementation phase. Under the pressure of chip export control, there are at least nine companies with domestic AI chip shipments exceeding 10000 kcal. The ultimate goal is to achieve a localization rate of 80%.