[By Guancha Network columnist Yanno]

Excluding the part involving abuse of power for personal gain, Trump's ideas for curing America were correct, including rebuilding manufacturing, restoring fiscal discipline, and shedding the burden of being the "world's police," which are all necessary. However, his hasty remedies turned what was right into wrong.

There is no winner in a tariff war, yes, but there is a biggest loser, which is the United States.

Generally speaking, the U.S. has an advantage in the tech war, but it is also believed that this advantage is rapidly disappearing. Among various high-tech fields, AI technology is the most prominent arena. Measured by time, many Chinese and foreign experts agree that China only lags behind by several months.

If measured by regulations, since this year, China has maximized its摆脱 being strangled. Even in the semiconductor field, on April 11, China Customs adjusted the rules for determining the origin of chips, meaning American-made semiconductors are now insulated from the world's largest chip consumer market. The ones truly being strangled are those American tech companies with hefty valuations.

Before Trump's administration imposed tariffs on semiconductor-related products, I would say: In the tech war, China will definitely win, and tariffs will only widen the winning margin, not the opposite.

The main reason is that American tech giants are facing various constraints, inevitably leading to increased costs, while their profitable markets are shrinking. Conversely, their Chinese counterparts have found effective ways to catch up.

This article briefly discusses the current state of the AI industry to glimpse the tech war.

Who is throttling whom between technology and the market?

Since the 2008 financial crisis, the stock returns of American tech industries have soared by 400%, compared to only 25% growth in other sectors. Interpreted positively, this phenomenon reflects the high demand and profitability of cloud computing capabilities; interpreted negatively, it means massive funds are concentrated in a few tech giants, making the market dependent on the fate of a few individuals. If any one of the seven giants (Mag 7) encounters problems, it could trigger a domino effect leading to a bubble burst. These two perspectives—positive and negative—are both opinions from Goldman Sachs, and even Goldman Sachs is relatively optimistic about the current AI bubble. In other words, most experts worry that Paul Krugman's concerns are not unfounded—the American tech industry is standing on the edge of a cliff. Krugman believes that AI investments are defensive in nature, with major companies merely fearing they will be surpassed by competitors if they do not invest, rather than based on objective demand estimates.

Therefore, Trump's continued tightening of export and investment restrictions on Chinese tech products is precisely at a time when American tech giants need profits to offset their massive AI investments, or, from another perspective, when they need the Chinese market the most. NVIDIA is a clear example; its stock price surged by 238.9% in 2023 and 171.2% in 2024, but since the beginning of this year, bad news has been continuous, and as I write, it has fallen by 23.2%.

The ban on H20 exports to China by the Trump administration has left this company, which had just announced a $50 billion investment in the U.S., embarrassed, much like Apple. Both companies thought spending money could buy them immunity, like TSMC.

Just before Trump threatened to impose tariffs on semiconductors—including end products containing semiconductors—it was reported that TSMC's 4-nanometer chips produced in its U.S. factory would increase in price by 30%. This is a heavy blow to the seven giants, but bad news keeps coming. Around the same time, Huawei launched the CloudMatrix 384 super-node cluster,对标英伟达GB200 NVL72, surpassing it in some performance metrics.

According to explanations from the well-known semiconductor media semianalysis, Huawei is breaking through the limits of AI system performance. In the current stage of AI development, enhancing overall system performance is more important than simply pursuing chip performance. This is also the story told by DeepSeek—mastering the technical path of "small against big" can overcome the technical shortcomings of individual components.

The CloudMatrix 384 uses 384 Ascend 910C chips (7-nanometer process), while the GB200 NVL72 uses 72 Blackwell GPUs (4-nanometer process) plus 32 Grace CPUs. This means that since the performance of the Ascend 910C is only 1/3 of the Blackwell GPU, 5 times the number of Ascend 910Cs are used to compensate for the performance gap. The more chips used, the higher the requirements for system optimization to achieve "small against big."

Although the CloudMatrix 384 is not fully domestically produced, it is close to the goal. For American AI giants, this progress cannot be stopped by technological bottlenecks alone. On the contrary, the pressing issue for American companies is market share because when investors doubt whether AI is over-invested, profitability becomes the only important thing.

Thus, China's formal rejection of "Made in America" is a major event because it means that American tech giants' investments in the U.S. will not only bear high costs but also face pressure from "limited market" sales. In simple terms, what they produce becomes more expensive, and where they can sell becomes fewer.

From the supply chain perspective, if you want to sell to China, the product and its parts must be produced outside the U.S., but Trump is bringing overseas part manufacturers to the U.S., including but not limited to chips, servers, phones, and other vast supply chain ecosystems.

The key point is that if these part suppliers are forced to raise prices due to high costs in U.S. factories, it will affect products outside the U.S. as well. For example, if TSMC raises the price of 4-nanometer chips, both U.S. and Taiwan factories will increase prices simultaneously. Buyers won't find cost-saving paths unless they turn to mainland China or other enterprises capable of producing parts without setting up factories in the U.S.

In other words, looking only at the AI industry, from technology to market, the U.S. is now strangling itself, and the stock market will eventually reflect this reality.

In 2022, the first batch of equipment arrival ceremony for TSMC's new factory in Arizona was held. Photo source: Bloomberg

"Taiwan Semiconductor" capacity insufficient for U.S. giants

America's advantage in chips is also its weakness because reliance on this advantage has driven China to develop a "system surpassing" technical path. Conversely, American companies must bear rising chip costs, which may spread to rising system costs.

Apple and NVIDIA have successively promised $50 billion in investments, but the output of TSMC's Arizona factory ("Taiwan Semiconductor") is barely enough to meet the orders of these two companies. Trump's tariff on semiconductor imports effectively closes the escape route for these large factories.

In 2023, TSMC's annual production in Taiwan was 16 million equivalent wafers (a standardized unit used to measure and compare the production volume of different-sized wafers). Currently, the first phase of TSMC's Arizona factory has achieved an annual production of 540,000 equivalent wafers. The second phase N3 factory is expected to start mass production in 2026, with an estimated output slightly higher than the first phase, around 360,000 12-inch wafers, equivalent to 810,000 equivalent wafers. Therefore, the total output of the first and second phases is approximately 1.35 million equivalent wafers annually.

In other words, by the end of next year, TSMC's American factory's production will be only 8.43% of Taiwan's factory (in 2023). Even if the third and fourth phases, with billions of dollars in additional investment, are realized before Trump leaves office, the rough estimate would still only account for 20% of Taiwan's factory production in 2023. I doubt this meets Trump's expectations, so it's possible that TSMC will be required to increase investment further.

Just Apple's current annual demand is already 540,000 equivalent wafers of "Taiwan Semiconductor," and with its planned $50 billion investment in AI servers, demand will naturally rise. Other demands from NVIDIA, AMD, Qualcomm, etc., are no less significant. By the time Trump leaves office, "Taiwan Semiconductor's" capacity will far from meet the needs of all American clients, even if Samsung and SK Hynix are included.

This reality means that regardless of how much American tech companies invest, they cannot meet American market demand because domestic chip production is insufficient. In other words, shouting about $50 billion, but to fully implement investments, it still depends on whether raw materials can supply production after several factories are built. This is the second困境.

The third困境is whether America's electricity supply is sufficient for these American giants to extensively build AI infrastructure.

Energy supply does not keep pace with Trump's ambitions

Siemens Energy's stock price soared by more than 300% last year, outpacing even NVIDIA. Why? One of the main reasons is the rapid development of AI data centers requiring substantial electricity supply.

From 2019 to 2023, the U.S. invested approximately $328.5 billion in AI infrastructure, while China invested about $132.7 billion. According to the International Energy Agency's estimates, by 2030, 80% of data center demand growth is expected to occur in the U.S., China, and the EU.

Additionally, according to Gartner's latest report predictions, the power consumption required for AI applications may surge by 160% in the next two years, doubling the 2023 consumption level by 2027. Globally, AI industry demand may exceed current power supply capabilities as early as next year, with 40% of AI data centers potentially facing operational restrictions due to power limitations.

For the U.S., the largest investments Trump can attract or "plunder" are "energy-guzzling monsters." For small and medium-sized businesses and ordinary people, this means higher electricity bills.

From the perspective of manufacturing, although the U.S. is the world's largest natural gas producer, overall energy costs in China are lower than in the U.S., especially with an advantage in grid upgrade and construction costs. Compared to this, upgrading America's existing grid requires spending approximately $720 billion. Therefore, China's current industrial environment and energy costs are more suitable for developing energy-intensive industries. Arizona, where TSMC is located, strongly supports semiconductor factories and AI industries, but its energy costs are four times higher than in Taiwan.

It should be noted that CloudMatrix 384, due to using a large number of chips and a less advanced chip process than NVIDIA's GB200 NVL72, naturally consumes more power. American experts believe this highlights that China is not short of electricity, whereas America's outdated grid equipment is the bottleneck restricting AI infrastructure development.

DeepSeek must be mentioned here, whose low-energy models can reduce the burden on AI data center scale and power demand to a certain extent. DeepSeek shocked the American industry and stock market mainly because of its cost-saving and energy-efficient approach.

In summary, energy issues may become the ceiling for American AI giants' development because they must pay higher energy costs than their Chinese counterparts.

Even gaining the "jewel in the crown of Taiwan" comes at the cost of the "Iron Throne."

H20 is a product specifically designed for the Chinese market by NVIDIA. Now that it cannot be supplied to China, I think it is good news because it means that investors will have a pessimistic view of the prospects of American tech giants, which will help Chinese related enterprises break away from the U.S. Additionally, the positive effects generated by investing in the U.S. are far less than those from investing in China, thus cutting off the road for major American companies.

Former U.S. Treasury Secretary Janet Yellen said that using tariffs to bring manufacturing back to the U.S. is a pipe dream, and the reason can be simplified to one word: cost. Trump surely knows this, but at least two reasons drive him down this dead-end path: one is his prideful personality, and the other is the small amount of spoils he is bound to reap, which deludes him into seeing death as an opportunity.

These spoils include the surrender document from Taiwan... perhaps the most complete surrender content.

Due to the political orientation of the "Taiwan independence" authorities and the degree of economic dependence of Taiwan on the U.S., the general trend of large-scale electronics industry moving to the U.S. should be determined. Moreover, the "Taiwan independence" authorities might strictly follow the "imperial edict" and tax electronic components, which is the largest trade project across the strait, giving Trump some propaganda material to claim victory.

However, once Taiwanese businesses join the American team, they will face the same predicament as American businesses: soaring costs, reduced market, and slowed research and development speed. When mainland China's technology catches up, they will be trapped on an island in the world.

If Trump insists on taxing semiconductor-related products, it means he hasn't yet recognized that the tech war is a protracted conflict. Fantasizing about quick victories through hasty measures will only result in faster and more thorough defeats. Even if he gains the "jewel in the crown of Taiwan," the cost will be the "Iron Throne."

Therefore, what needs to be guarded against is not Trump imposing tariffs, but if he abandons taxing semiconductor-related products—a very low probability—he needs to change tactics in dealing with the tech war, but the outcome of winning will not change.

This article is an exclusive contribution by Guancha Network, and the content purely represents the author's personal views, which do not necessarily reflect the platform's stance. Unauthorized reproduction is prohibited, or legal consequences will ensue. Follow Guancha Network WeChat public account guanchacn for daily interesting articles.

Original link: https://www.toutiao.com/article/7497030695383974439/

Disclaimer: This article solely represents the author's viewpoint. Feel free to express your opinion by clicking the "Like/Dislike" buttons below.