Trump backstabbed Jensen Wong! Nvidia H20 "first talk good is useless" was banned by the United States to lose 5.5 billion magnesium, how to enlighten TSMC at the end of the bright card negotiations?

The United States raided to ban the sale of Nvidia H20 AI chips to China, hit its stock price hard and recognized huge losses, behind which we saw Trump's showdown negotiation response to Huida, what enlightenment did it bring to the Taiwan stock technology supply chain? (Synopsis: Trump Huang Jenxun "talked and laughed"!) The United States will suspend the ban on Huida H20 chip to China, deepseek to take off? (Background supplement: NVIDIA GTC 2025 Summary: Huang Jenxun foresees the "AI Agent Revolution" entering the home consumer level, Blackwell Ultra or spawning a large number of one-person start-ups) The tension between the United States and China extends to the field of artificial intelligence (AI) chips, and Huang Jenxun and other Huida technology supply chain may have potential risks, according to the Wall Street Journal, the Trump administration recently tightened chip export controls to China again, requirements Nvidia must obtain a license to sell its H20 AI chips to China and a handful of other countries, which is tantamount to treating Huida's recent negotiations as a chess game between China and the United States. Huida preemptively negotiated backstabbed According to previous reports, Huida founder Huang Jenxun recently went to Mar-a-Lago to negotiate with Trump, Huida is willing to spend 500 billion US dollars to build a server center in the United States to ensure the leading position of US AI, causing an uproar in the industry, but it is worth noting that in exchange for the last H20 export ban relaxation did not happen as expected, the Wall Street Journal said that this move made Huang Jenxun and Huida very stunned, which made Huida must recognize the loss of H20 and other chip production inventory of about $5.5 billion this quarter. It is understood that Nvidia's H20 chip was originally designed to comply with the export control regulations set by the United States and specially designed "downsized" products, H20 performance is about 75% lower than Nvidia's top AI chip H100 two years ago, is for Huida in order to sell to China and heavily customized version, but this version of the chip if not sold in the Chinese market, may be completely uncompetitive, so can only be recognized as a loss. Nvidia's data center revenue in the fiscal year ending January was $115.2 billion, and the division is expected to generate $182 billion in revenue this year, according to analyst estimates, although H20 sales are not the largest share, but the loss of this revenue will undoubtedly affect Nvidia's ability to consistently exceed Wall Street's high expectations — the company's norm for at least three years. Even chips with reduced performance have been banned from being sold to China, indicating that the US government's regulatory policies are constantly changing, and it also shows that it intends to completely cut off China's access to American AI technology. One of the reasons why the U.S.-China model competition has prompted the U.S. government to take a tougher approach may be related to the rise of Chinese AI startup "DeepSeek". DeepSeek's success in developing powerful AI models with relatively little computing power surprised Silicon Valley and Wall Street, and caused a high degree of vigilance in the US government. Officials realized that even with a downspec chip like the H20, DeepSeek is still very effective for the "inference" of AI models to the application of AI models to real-world scenarios. U.S. Commerce Secretary Howard Lutnick bluntly said at his nomination hearing in January this year that Nvidia chips have driven the development of the DeepSeek model, and the United States must stop this "funding adversary" behavior. If they want to compete with us, let them do it themselves, but they shouldn't use our tools against us anymore. More importantly, in addition to DeepSeek in China, QwQ of the Alibaba system has also become a hot new model choice in the AIGC community, according to community tests, QwQ 32B models in the same model size, many indicators far better than DeepSeek 32B, which shows that China has begun to have an optimization route for small-capacity models, and even allows companies to deploy locally at low cost, all in the case of fewer graphics cards in China. That could make the White House judge the AI trade war even more severe. Trump's "change of order" has become a nightmare for Taiwan stocks Even if you think you play by the rules and sacrifice a little for the team, the rules may still change tomorrow, and you will still be hit again. It's a brutal environment for those who want to invest billions in a 30-year project. Scott Lincicome, a trade policy expert at the Cato Institute, said this about the high level of uncertainty facing Trump's tech companies amid geopolitical risks. For uncertainty, the U.S. export ban on AI chips is not limited to Nvidia and AMD, the global semiconductor supply chain is also feeling the brunt, and even TSMC's equipment maker Dutch company ASML shares have fallen, the company reported weak quarterly orders, and said tariff issues are exacerbating market uncertainty. The Trump administration imposed tariffs of up to 245% on Chinese goods just a few days before the chip ban, and exempted processors, smartphones and other electronic products from temporary tax exemption, but also targeted AI chips, this series of repeated and tough actions, for the technology industry with a very large equipment capital, such as TSMC, is a very large risk of great uncertainty. Before the 90-day trade suspension, customers with a keen sense of smell had already anticipated the imminent control and rushed to place orders before the ban took effect, but will this short-term market change again and become another pressure for overinvestment and overproduction? What did the showdown negotiations get in exchange for? Returning to Huang Jenxun's Mar-a-Lago Bureau, what did Huang exchange for a $500 billion investment commitment? The most pessimistic answer may not have changed anything, and people can't help but think of the picture of Wei Zhejia, chairman of the front desk of TSMC, going to the White House to promise to invest hundreds of billions of dollars, personally going to Trump to negotiate, quickly and urgently may be in his arms. What's more, the linkage of the semiconductor industry chain is very high, and Taiwan's stock market, which is highly dependent on the semiconductor industry, may usher in huge fluctuations under uncertain risks, if Huida and other potential policies and export changes, the impact is not only TSMC wafer foundry, but also packaging and testing, heat dissipation modules, server foundries, etc. are closely related, Nvidia's stock price has plummeted, which has a comprehensive appeal to Taiwan stocks. Taiwan's technology factories have flourished under the global division of labor, but now the confrontation between the United States and China has forced the supply chain to face the pressure to choose sides, and the risk of relying on a single customer or market has increased significantly. Although the ban may accelerate China's development of the local semiconductor industry in the short term, which in turn will produce a so-called "order transfer" or "de-beautification" effect, benefiting some Taiwanese manufacturers, the process is full of uncertainty, and the technical threshold and market competition are still fierce, which may make China seek local suppliers Citigroup analysts expect that the Chinese cloud service provider that originally planned to use H20 processors to meet 50% of AI accelerator demand in 2025 There may now be a shift to AI chips from local vendors such as Huawei and Cambricon. For investors in the layout of Taiwan technology stocks, it is necessary to more carefully assess the degree of geopolitical exposure of individual stocks, diversify investment risks, and pay close attention to the subsequent development of US-China relations, and do not put all eggs in the same seemingly glamorous AI technology basket...