Recently, I noticed a pretty interesting phenomenon: the concept stocks of data center servers have been hot since last year, and many investors are asking whether this trend can continue into 2026. I took a close look at the industry chain structure and found that the entire server ecosystem is actually divided into three key segments, each with different leading companies vying for position.

First, let's talk about core complete machine assembly. Foxconn, as the world's leading OEM for servers, has deep ties with NVIDIA, and they are the main supplier for cabinet systems like GB200. They can handle everything from components to full machine assembly, and this kind of vertical integration is quite rare in the industry. Quanta mainly serves Google, AWS, and Meta, with R&D capabilities recognized as the strongest in the industry. Wistron is fully focused on data center business, with 100% of its revenue in this area, working closely with large-scale cloud service providers, and has strong profitability in ASIC servers and liquid-cooled cabinet integration. There are also Celestica and Microchip. Celestica has differentiated advantages in 800G switches and Google TPU manufacturing, while Microchip is a pioneer in modular servers and early layout of liquid-cooled cabinets.

The infrastructure segment is even more interesting. Because AI chips consume a lot of power, electricity and cooling have become the biggest pain points for data centers. Vertiv is a leader in global thermal management and power management; their cooling distribution units are essential when AI cabinet power exceeds 100kW. Taiwanese companies Qihong and Shuanghong lead in cooling modules, supplying 3D VC and liquid cooling plates for GPUs. Eaton is a giant in power management, helping data centers cope with grid stress.

The value of core components is several times higher than that of traditional servers. Taiguang Electronics dominates high-speed copper-clad laminate markets, Jinxing Electronics has extremely high technical barriers for high-layer count PCBs, and Qinye provides high-strength cases capable of supporting heavy GPU modules. These seemingly small parts are actually becoming increasingly important in the cost structure of AI servers.

I believe these five stocks are worth paying close attention to. Celestica has performed remarkably in recent quarters, with revenue up 28% year-over-year and non-GAAP adjusted EPS up 52%. The average Wall Street target price is $374.50, with a potential upside of 22.44%. Vertiv, due to its leading liquid cooling technology and deep collaboration with NVIDIA, saw a 29% year-over-year increase in net sales in Q3 2025, with order backlog reaching $9.5 billion. The average analyst target price is $206.07, with a potential upside of 27.38%.

Here in Taiwan, Quanta, as a major global EMS manufacturer, focuses on high-performance server assembly. In Q3, revenue grew over 20% year-over-year, with quarterly net profit surpassing NT$15 billion, and full-year EPS reaching NT$17.37. Management has raised the full-year guidance for 2025, expecting AI-related business to continue as a core engine into 2026. Morgan Stanley maintains an outperform rating with a target price of NT$330.

Wistron is growing the fastest, with full-year consolidated revenue up 163.68% to a record high, and EPS reaching NT$275.06. Order visibility extends into 2027. Foxconn, as the largest ICT manufacturer globally, has an AI server market share exceeding 40%. Full-year 2025 revenue is estimated to surpass NT$8.1 trillion, with management expecting AI-related revenue to reach the trillion-dollar scale. The average Wall Street target price is NT$306.00, implying a 29.66% upside from early 2026 market prices.

However, investing in these data center server concept stocks involves certain risks. First, as investor focus shifts from revenue growth to profitability, shorter depreciation cycles and rising power costs may suppress some companies’ profits. Second, most AI server stocks have already experienced huge gains, with high P/E ratios for the leaders, and market concentration is very high. If signs of an AI bubble burst appear, it could trigger sharp corrections.

Several variables for 2026 still warrant close attention. Will cloud service providers’ investments in AI infrastructure meet expectations? The progress of non-x86 architectures and self-developed ASIC chips? The development of edge AI? The localization of AI chips in China? Data sovereignty policies in various countries? Changes in US regulations and tariffs? And whether the market can shift from speculation to actual commercial value creation? All these factors will directly impact the future performance of data center server concept stocks.

In summary, the fundamentals of this sector are still intact, but valuations and risks are also present. Investors should carefully evaluate entry timing and stay closely tuned to those factors that could change the game.