Recently, I've been looking into the server concept stocks in this sector and found that the performance of these companies this year is indeed worth paying attention to, especially those with differentiated advantages in liquid cooling technology and AI infrastructure.

Speaking of the server industry, the supply chain can actually be divided into three segments. First is core complete machine assembly, with companies like Foxconn, Quanta, and Wistron responsible for integrating CPUs, GPUs, and cooling systems into complete servers or racks. Their relationship with NVIDIA is extremely close. Then there's infrastructure and environmental control, which I think is the most easily overlooked segment—because AI chips consume too much power, electricity and cooling have become the biggest pain points for data centers. Lastly, there are key components, such as high-frequency copper foil substrates, high-layer-count PCBs, and server chassis, which have several times the value in the AI server era compared to traditional servers.

If I had to pick key targets from the server concept stocks, I would focus on these. Quanta’s R&D capability is truly strong, mainly serving Google, AWS, and Meta, leading clearly in high-performance AI server system integration. Wistron is a pure play focusing on data center business, with 100% of its revenue from data centers, especially skilled in ASIC server and liquid cooling rack integration. Foxconn, as the global leader in server OEMs, has an AI server global market share of over 40%, with a partnership with NVIDIA that’s unmatched. On the US stock side, Vertiv is an absolute leader in liquid cooling; when AI rack power consumption exceeds 100kW, their cooling distribution units are basically essential. Celestica has a differentiated advantage in 800G switches and Google TPU manufacturing, making it a key player in AI network transmission segments.

Looking at last year's data, these companies have indeed been growing rapidly. Quanta’s full-year revenue approached NT$1.9 trillion, with over 20% annual growth; Wistron’s revenue was even more impressive, reaching NT$950.6 billion, up 163.68% year-over-year, with EPS reaching NT$275.06. Vertiv’s net sales in Q3 last year were $2.68B, up 29% YoY, with order backlog reaching $9.5 billion. Celestica’s revenue in the first three quarters of last year was $3.19 billion, up 28% annually, with non-GAAP adjusted EPS increasing 52%. Foxconn’s full-year revenue broke through NT$8.1 trillion, with net profit up 35% year-over-year.

Wall Street’s outlook on these server concept stocks remains relatively optimistic. Celestica’s 12-month average target price is $374.50, with a potential upside of 22.44%. Vertiv’s average target price is $206.07, with a potential upside of 27.38%. Quanta’s institutional target price remains at NT$330, with a potential upside of 29.66%. Foxconn’s average target price at the start of this year was NT$306.00, leaving plenty of room for growth.

But I have to be honest—there are some risks in this sector worth noting now. First, these server concept stocks have already accumulated huge gains, with P/E ratios generally high, and market concentration is very high. Second, as investors shift focus from revenue growth to profitability and ROI, shorter depreciation cycles and rising power costs could suppress some companies’ reported profits. Additionally, the rise of non-x86 architectures and self-developed ASIC chips may change the market share structure of traditional servers. Lastly, if signs of an AI bubble burst appear or investors shift from growth to profit validation, these stocks could experience sharp corrections.

Therefore, before investing in server concept stocks, it’s essential to keep an eye on several key factors: the actual investment scale of cloud service providers in AI infrastructure, the impact of data sovereignty policies in various countries on the supply chain, US regulatory changes and tariff pressures, and the progress of de-speculation in the market. These factors could significantly influence the future performance of these targets.