AI CAGR of 32%! Deleveraging Wrongly Sells Off High-Performing Stocks—Next Mainstream Wave?

This week, news emerged that Google is developing a new server chip codenamed 'Frozen v2' to solve the severe shortage of computing power. By directly etching the unchanging mathematical logic of its Gemini model into the chip—bypassing the traditional CPU/GPU cycle of 'fetch, decode, execute'—processing performance is estimated to reach 6 to 10 times that of current TPUs. However, the drawback is clear: the chip cannot be used if the Gemini model undergoes major architectural changes or for non-Gemini models.

A simpler analogy:

General-purpose GPUs are like 3D printers. Given different 3D blueprints (different AI models), they can print anything. While highly flexible, each print takes considerable time.

In contrast, Frozen v2 is like a factory's dedicated casting mold. It completely abandons the ability to print other items; once raw material (data) is poured in, a finished product is pressed out in one second.

As a result, Google plans to use Frozen v2 internally and will not mass-produce it. However, this also demonstrates that AI computing power shortages are so acute that, while building AI data centers, companies are adopting various methods to reduce server and computing costs.

Beyond the four major CSPs—Google, Meta, Microsoft, and Amazon—all developing AI ASIC chips, two leading AI companies, OpenAI and Anthropic, are also developing their own ASICs to support future massive computational demands.

Standard Chartered forecasts that global AI capital expenditure will grow at a compound annual growth rate (CAGR) of 32% from 2025 to 2030, with Taiwan's semiconductor sector benefiting the most. Whether it's NVIDIA, AMD, or the aforementioned AI ASIC chips, TSMC will be the most critical foundry. Consequently, TSMC raised its capital expenditure guidance during its recent earnings call—a positive signal—yet the market remains hesitant due to global deleveraging.

On July 20, in 'TSMC Earnings Call Useless!? Global Deleveraging Triggers Panic Selling—Is This the Final Bottom?', it was noted: '7 out of the 10 best-performing days occurred within two weeks of the 10 worst-performing days, and 6 of those 7 days happened the day after the worst 10 days—so the darkest hour is always before dawn.'

On July 21, 'Stocks Want to Rest but Keep Falling, Investors Want to Buy but Keep Getting Trapped! When Will the Taiwan Market Bottom Out?' predicted: 'This week, leading stocks will bottom out first. TSMC (2330-TW) will find support around 2300 NT dollars.'

These predictions have now been validated. TSMC has shown signs of bottoming and is leading the Taiwan stock market upward. The TSMC supply chain is benefiting from increased capital expenditure, making it an industry investors should monitor. Large-cap, high-performing AI stocks have reached relatively low valuations, revealing buying opportunities. These are ideal early-entry targets for more conservative investors.

Yesterday's price rise occurred on low volume, hiding potential risks. For detailed analysis, visit the YouTube channel: https://www.youtube.com/@money66868

Besides the TSMC supply chain and AI leaders, which other stocks should investors consider? Join our LINE@ to find out.

TSMC Supply Chain: Creative (3443-TW), Chung Sand (1560-TW), Sunsemi (8028-TW), Hsiang Ming (8091-TW), Fan Hsuan (6196-TW), Yashine (6139-TW), KOC (1785-TW), Taisol (5434-TW)

AI Leader Stocks: ASE (3711-TW), Tripod Technology (2383-TW), Lianmao (6213-TW), Unimicron (3037-TW), Nanya PCB (8046-TW), Accton (2345-TW), Molex-KY (3665-TW), Winbest (6515-TW)

More industry trend forecasts and intraday updates are available exclusively on Ye's LINE@:

LINE@ ID: @money668 (remember to include the @)

https://user225916.pse.is/money668

Source: Moore Investment Consulting – Analyst Ye Junmin

The securities recommended and analyzed by our company have no improper financial interests. Past performance does not guarantee future profits. Investors should make independent judgments, conduct careful evaluations, and bear investment risks themselves.