EP701 | 🐡

Rapid Rotation Helped Taiwan Stocks Recover Their Highs Ahead of Schedule After the July Plunge

The most important lesson from the third quarter was that Taiwan stocks suffered their sharpest decline of the year in July, with positions retreating about 30% from their highs. Although cumulative returns from the start of the year to the peak were still substantial, the sharp short-term drawdown was psychologically difficult for an approach using roughly 1.1–1.3x leverage. When the large-cap equal-weight line failed to stop falling at the point where it should have provided support, the host gradually cut positions that were still profitable, giving up on stubbornly holding the original holdings.

The initial expectation was that the drawdown might not recover until the end of the year, but Taiwan stocks had already retested their previous high by early September. U.S. stocks recovered more slowly, mainly because a heavily weighted individual holding fell and then consolidated for longer, dragging down overall performance. The rotation in August and September was toward newer leading themes, including optical communications, small-cap ICs, copper-related concept stocks, satellite-related themes, and some stocks that had fallen into deep-value territory, with an expected forward P/E of about 10x next year. The ability to identify strong groups early and get on board quickly is certainly important, but the host also acknowledged that luck was involved and that market moves cannot always be seen in advance.

Optical Communications and Small-Cap ICs Rally as AI Orders Spill Over

Taiwan stocks have recently seen the unusual phenomenon of many small IC design companies signaling strength at the same time. One possible explanation is that demand for AI chips is so large that chip-design teams and wafer-fabrication capacity have become scarce resources. After large players prioritize AI-related orders, they may distribute some demand externally if they lack sufficient spare capacity, giving small and midsize design companies opportunities they previously could not obtain.

The host cited MediaTek2454.TW taking on a large volume of TPU orders as an example, noting that when a large company concentrates resources on AI products, other orders may flow to small and midsize Taiwanese IC design companies. This does not mean every small company will succeed; rather, when demand becomes excessively concentrated at the top of the industry chain, it may spill down through the supply chain layer by layer, allowing smaller companies to capture some orders as well. If these companies also have relatively high development efficiency, the market may reassess the ceiling for their revenue and earnings.

AI Could Accelerate Chip Design and Increase the Operating Leverage of Smaller Companies

Another observation concerns OpenAI's chip project. The program reportedly took about eight to nine months from finalizing the architecture to tape-out; the actual timeline may have been closer to a year, but the overall process was still extremely fast. If the claims about official AI collaboration are accurate, AI may save substantial costs in engineering input and total labor hours, allowing a design company, within the same two- or three-year period, to move from launching only one product to potentially advancing three at the same time.

This case could also serve as an example for companies such as Block$XYZ that have chip-design capabilities. The host cautioned that the speed may not have been entirely due to AI and could also reflect the company's own engineering capabilities, so it is too early to draw a firm conclusion. However, if AI can indeed lower the barriers to design, small and midsize players may be able to use smaller teams to take on projects that were previously beyond their reach, creating greater operating leverage. Design companies have broadly incorporated AI into their workflows, but because customer secrets cannot be uploaded freely to external services, they more commonly use open-source models deployed in their own data centers in practice.

Closed-Source Models May Enter Corporate Data Centers and Expand Alongside Open-Source Options

Open-source models are already quite practical for basic tasks such as translation and data organization, but closed-source models still have an advantage in research, cutting-edge technology, and work requiring extensive knowledge integration. The program speculated that closed-source model providers may eventually stop requiring customers to connect exclusively to a cloud black box. Instead, they could deploy a processed version on customers' local servers and maintain its capabilities through subsequent updates.