■Jiang Minghong
U.S. retail sales increased by 1.2% monthly in August, indicating solid consumption momentum; the job market gradually cooled down in September after a strong rebound in August, with only 29,000 new jobs added, the unemployment rate rose slightly to 4.2%, and wage growth was relatively moderate. In terms of prices, driven by rising gasoline prices, the Consumer Price Index (CPI) increased by 0.4% monthly in August, higher than the 0.1% increase in July. There are various signs that the current US economy is still resilient.
The high-end market is in shock, and the market is waiting to see the change of chip generations.
The current high-level fluctuations in the stock market are mainly due to the continued expansion of AI infrastructure capital expenditures by large cloud companies. In addition, inflation concerns have made the market worried that the Federal Reserve will continue to raise interest rates. The U.S. 10-year Treasury bond yield remains high, affecting the performance of AI stocks. In addition, the third quarter coincides with the replacement of AI chip generations and the increase in component prices. The market remains on the sidelines on supply chain adjustments during the transition period between new and old products.
Please read on...
U.S. retail sales increased by 1.2% monthly in August, indicating solid consumption momentum. (AFP)
But in the long term, the growth trend of AI has not changed, and it is becoming more and more "power hungry". The power consumption of a single mainstream AI chip is rising rapidly, from 700 watts in the H series to 1,000 watts in the Blackwell generation B200, and the next generation Rubin generation is expected to exceed 2,000 watts. The power consumption of a single cabinet has doubled and may exceed 1 million watts in the future, driving a significant increase in demand for liquid cooling. As the new platform gradually increases its volume, its specifications and single-container output value increase, it is expected to drive supply chain operations upward; the next-generation platform will also lay a long-term upgrade trajectory. In addition, with the rise of “agent AI” that can perform tasks autonomously, computing requirements continue to expand, and the AI bottleneck is extending from chips to electricity. In addition to the continued expansion of GPU data centers, the shipments of self-developed chips by large cloud companies (such as Google and AWS) are also expected to grow significantly in 2027. The International Energy Agency predicts that global data center power consumption will double by 2030, and Taiwan's supply chain is expected to continue to benefit.
As AI becomes more capable, security issues become more important. In April this year, a new AI model demonstrated its ability to find thousands of software vulnerabilities on its own, the longest of which had been hidden for 27 years, attracting great attention from financial institutions and governments. Ordinary people are faced with AI face-changing and voice-changing scams that are difficult to distinguish between genuine and fake; if companies allow AI to make decisions for their employees, once they are controlled by hackers, the losses will be even more difficult to estimate.
Supervision upgrades promote computing power protection, data center power supply becomes new test
Supervision and protection are upgraded simultaneously. The European Union has required that AI-synthesized videos must be prominently labeled, and Taiwan has also passed the Basic Law of Artificial Intelligence; research firm Gartner estimates that global AI security spending will increase significantly from approximately US$25.9 billion in 2025 to approximately US$86 billion in 2027. The defender must "fight AI with AI" to add new uses for computing power needs; while data centers operate around the clock, stable power supply and backup power are also part of security protection.
Reasonable supervision is like the brakes on a car. It does not make the car drive slower, but makes people dare to drive faster. Only by winning trust can AI enter all walks of life, and supervision will help the healthy development of the industry in the long run. No matter which AI manufacturer wins in the end, stable power and information security protection are infrastructure that cannot be bypassed.
Taiwan stocks focus on "three main lines" and it is advisable to keep cash for operations and wait for pullbacks
The industry can pay attention to three main lines: 1. Hardware: in advanced packaging and testing, IC carrier boards, high-end PCBs, liquid cooling and power management, optical communications, etc., Taiwanese manufacturers have obvious advantages. 2. Software: The key line of defense for the implementation of AI. There are few related targets in Taiwanese stocks. We can observe the world's leading manufacturers. 3. Non-AI: Those with growth potential and low price-to-earnings ratios, such as biotechnology and some consumer electronics.
Looking ahead to the market outlook, the current long-term interest rates remain at a high level. In addition, geopolitical risks affect oil prices and inflation expectations. The stock market is expected to maintain a high-level shock in the short term, and the trend of yield rates will be a key variable.
The long-term prospects of the AI industry are clear, but in the short term, corporate revenue and legal statements still need to be further verified. Operationally, it is advisable to retain some cash levels and wait for opportunities to pull back.
(The author is the fund manager of FT Taiwan Perpetual High Interest ETF (00961))
Grasp the economic pulse with one hand. Click here to subscribe to Free Finance Youtube channel
