Looking back at 2025, technology became the main theme in A-shares.

Within tech, the standout was AI. AI computing power names such as “Yi Zhong Tian” (Eoptolink, Zhongji Innolight, TFC Communication) and “Ji Lian Hai” (Cambricon, Foxconn Industrial Internet, Hygon Information) carried much of the market’s technology trade. AI applications and AI edge devices, by contrast, lagged AI computing power.

Data source: Wind

Miaotou believes the excess returns in AI computing power come down to certainty, and to expectations being repeatedly exceeded.

Capital and companies have been willing to bet on the certainty of current and future AI computing power infrastructure buildout. Examples include policy support for domestic substitution of GPUs, and rising capital spending by the four major North American cloud vendors as well as Chinese internet companies such as BAT.

But in 2025, large-model companies still showed little prospect of profitability, while sales of edge products such as AI phones and AI glasses remained weak. Those areas did not give the capital market the investment certainty it wanted.

Even so, global tech giants are in a cycle of high-intensity, high-growth capital spending, and the market broadly expects this investment boom in core technologies such as AI to continue over the next few years.

So what will be the main theme for AI investing in 2026?

Looking to 2026, Miaotou believes AI computing power still offers strong certainty and is likely to remain the main technology theme. As AI infrastructure is built out and upgraded, sub-sectors such as computing power (GPUs), transmission capacity (optical modules), liquid cooling, and PCBs could create investment opportunities.

Here, Miaotou has compiled the market size of various AI computing power segments. Based on compound annual growth rates from 2024 to 2029, the ranking is: China AI data center GPUs > liquid cooling for China AI data centers > optical chips > AI-related PCBs with more than 18 layers > optical modules.

Data source: Miaotou compilation

Next, we will go through them one by one.

GPU

We begin with China’s AI data center GPU market, which has the highest compound growth rate.

In terms of market structure, NVIDIA still holds a large share, while domestic GPU vendors are rising with policy support for domestic substitution.

Unlike NVIDIA in the U.S. stock market, the valuation logic for domestic GPU makers cannot be fully quantified. It is supported by the broader narrative of domestic substitution.

Still, GPUs are starting to enter the earnings delivery phase.

Newly listed GPU makers MetaX and Moore Threads expect to become profitable as early as 2026 and 2027, respectively. Cambricon turned profitable on a single-quarter basis in 2024 Q4 and was already profitable in the first three quarters of 2025.

Cambricon’s revenue ramp depends on validation and deployment by internet customers, which is why its revenue grew explosively in 2025.

In other words, capital spending by domestic internet companies on AI infrastructure is the main growth driver for domestic GPU makers such as Cambricon.

China’s major technology companies are also increasing capital investment. Alibaba, Tencent, and Baidu saw capital expenditures rise sharply by 132.46%, 48.24%, and 74.49% year on year, respectively, in 2025 Q1-Q3.

Against the backdrop of domestic GPU substitution and internet companies stepping up AI investment, some institutions forecast that Cambricon could achieve net profit attributable to shareholders of 4.872 billion yuan in 2026 and 7.991 billion yuan in 2027, up 118.71% and 63.99% year on year, respectively.

Even so, the capital market remains concerned about whether valuations for GPU makers such as Cambricon are too high.

On valuation, Cambricon’s market capitalization is 571.6 billion yuan. Based on the 2027 earnings forecast of 7.991 billion yuan, its price-to-earnings ratio would be 71.48. NVIDIA’s P/E ratio (TTM) is 45.94.

Miaotou therefore believes that although Cambricon has pulled forward two years of earnings expectations, its valuation still contains “froth” compared with NVIDIA.

To digest that “froth,” Cambricon and its peers need higher earnings expectations, and domestic internet companies need to increase the intensity of their capital spending on AI infrastructure.

That said, it is indeed possible that China’s internet giants will raise AI investment and capital expenditure plans.

On December 23, foreign media reported that ByteDance plans to invest 160 billion yuan in AI in 2026, with half of the budget earmarked for AI chip procurement. Based on an estimated 2025 profit of $50 billion, ByteDance’s 2026 AI investment would amount to nearly half of its full-year 2025 profit.

After that report circulated, shares of GPU makers such as Cambricon recovered somewhat.

Miaotou believes that in 2026, a valuation re-rating for GPU makers represented by Cambricon will have to be driven by expectations, namely upward revisions to capital spending by major domestic internet companies. Investors still need to watch whether Alibaba, Tencent, Baidu, and others raise capital expenditure plans. GPUs could still beat expectations in 2026.

Liquid Cooling

Liquid cooling is one of the highest-certainty, high-growth tracks in the AI computing power era, and the industry is now in the early stage of a breakout as penetration accelerates.

Liquid cooling replaces air cooling with a liquid medium and offers stronger heat dissipation. For example, the cooling efficiency of single-phase cold-plate liquid cooling is far higher than air cooling. As the heat flux density of AI chips such as NVIDIA’s GB300 exceeds 500W/cm², traditional air cooling can no longer meet high-power requirements, and liquid cooling is moving from an option to a necessity.

Institutions forecast that liquid cooling penetration in global data centers will rise from 14% in 2024 to 31% in 2026.

Liquid cooling service providers still operate around a major-client business model, similar to Apple’s supply chain.

NVIDIA began large-scale use of liquid cooling with GB200 NVL72, while Google’s TPU v7p and other products are also expected to deploy liquid cooling. As a result, the capital market is paying closer attention to liquid cooling service providers that have already entered the supply chains of NVIDIA, Google, and others.

However, NVIDIA and Google use different liquid cooling supplier models.

Liquid cooling systems for NVIDIA’s GB200/300 and other products are mainly led by long-term high-end suppliers such as Vertiv and Delta. Domestic companies typically enter as secondary component suppliers or full-chain solution providers; companies such as Envicool and BYD Electronics have already entered RVL/AVL recommendation lists.

This “recommended list plus ODM free decision-making” model means domestic secondary suppliers can see large orders but cannot firmly lock in long-term ones, leaving order stability relatively weak.

For secondary component suppliers, Google’s liquid cooling orders are much more stable than NVIDIA’s and offer greater certainty. Google directly engages with and designates liquid cooling component suppliers.

Because various vendors will push liquid cooling solutions in 2026, Google may develop new liquid cooling suppliers with lower shares in NVIDIA’s ecosystem to avoid supply and capacity conflicts.

In other words, Google could bring a surprise upside to domestic liquid cooling suppliers. For example, Envicool has already won a bid for a Google data center project, and Siquan New Materials’ ultra-thin VC vapor chamber has passed Google certification.

Against the backdrop of expected volume growth in liquid cooling in 2026, Miaotou believes domestic liquid cooling component suppliers that can enter Google’s supply chain in 2026 could see valuation re-rating opportunities.

Investors also need to watch whether expectations around supply-chain orders from Google are fulfilled. Envicool, for example, has attracted capital after entering Google’s supply chain. On valuation, based on its 2027 earnings forecast and valuation, Envicool’s P/E ratio has reached 74.56, above the sector average.

Data source: Wind

Miaotou believes other liquid cooling suppliers with lower valuations could gain greater valuation elasticity once they enter supply chains such as Google’s. Investors should continue to watch changes in the liquid cooling supply chains of Google and NVIDIA.

Optical Modules (Optical Chips)

If GPUs handle computation and represent computing power, optical modules handle transmission and represent “transport capacity.”

The performance of optical modules directly determines the efficiency and stability of data transmission. If data transmission cannot keep up, it is like a traffic jam on a highway: no matter how strong the computing power is, it cannot be fully used.

The investment logic for optical modules has two parts:

Product technology iteration, such as the transmission-speed improvement from upgrading 800G optical modules to 1.6T optical modules;

Growth in downstream demand, namely capital spending by cloud vendors such as Google, Amazon, Meta, Microsoft, and BAT on data center construction.

Among A-share optical module companies, “Yi Zhong Tian” (Eoptolink, Zhongji Innolight, and TFC Communication) are the most representative. Since 2025, they have risen 450%, 422%, and 231%, respectively, with share prices repeatedly hitting new highs.

That has also made investors worry about “froth.”

Based on Wind consensus expectations, Zhongji Innolight, TFC Communication, and Eoptolink are expected to achieve net profit attributable to shareholders of 25.297 billion yuan, 3.873 billion yuan, and 20.670 billion yuan in 2027, respectively, implying P/E ratios of 27.54, 42.08, and 21.26.

Miaotou previously argued in “‘Yi Zhong Tian’s’ ‘Froth’ Has Been Priced Out to 2027” that the “high valuation” supporting optical modules comes from expectation gaps, such as upward revisions to cloud vendors’ capital spending, Google TPU chips, and the increase in optical module quantities driven by network pricing (Scaleup).

At present, these “above-expectation” drivers are being realized or have already been realized.

In other words, optical module earnings expectations have already been priced out to 2027. Still, institutions such as Goldman Sachs remain bullish on optical modules; Goldman Sachs, for example, raised its target price for Zhongji Innolight to 762 yuan.

Miaotou believes that in 2026, a further valuation re-rating for optical module companies will still require a new “narrative” or positive catalyst.

In 2026, investors can focus on whether North American cloud vendors again raise capital spending, and on new opportunities from optical module speed upgrades to 1.6T as well as technology trends such as silicon photonics, OCS, and CPO. These could all become catalysts.

Optical chips are the core components of optical devices and optical modules, and their performance directly determines information transmission speed and network reliability.

The investment logic for optical chips is broadly similar to optical modules, with one additional driver: domestic substitution.

In terms of market structure, European and U.S. companies such as Broadcom, Lumentum, and Coherent control the optical chip market. According to ICC data, overseas vendors account for about 75% of the 25G optical chip market and about 95% of the market for optical chips above 25G.

Against the backdrop of surging shipments of high-speed 800G and 1.6T optical modules, demand for 100G-and-above optical chips is strong, and their growth rate will far exceed that of medium- and low-speed optical chips.

On its 2025 Q3 earnings call, Lumentum said the optical chip supply-demand imbalance had worsened. Even after adding capacity, the company was still in a position where it had to make allocation decisions almost every day. The current gap versus total customer demand has widened from 20% last quarter to 25%-30%. Looking to 2026, high-end optical chip prices are expected to rise given the supply-demand imbalance.

Miaotou believes that if any domestic optical chip company makes a breakthrough in high-end optical chips in 2026, its valuation elasticity will be higher than that of optical modules.

PCB

The phones, computers, routers, and air-conditioner remotes we use every day all contain one or more green boards, though they may also be other colors. Those are PCBs, or printed circuit boards.

Compared with traditional servers, AI servers have significantly higher requirements for PCB transmission speed, layer count, density, and other specifications, pushing the technology path toward high-end categories such as high-multilayer boards and HDI.

This has created an investment logic for PCBs in which both volume and price rise.

On volume, Prismark forecasts that from 2024 to 2029, AI-related PCBs with 18 layers or more will grow at a compound annual rate of 20.6%, far above the average growth rate of the PCB industry.

On price, the PCB value per AI server can reach as much as 5,000 yuan, more than three times that of a traditional server.

As AI servers iterate, upgrades in PCB/CCL materials and processes bring new value increments.

For example, mainstream general-purpose servers already use M6-grade copper-clad laminates, while AI servers and 400G/800G switches mainly use M7 and M8-grade CCLs. As the number of layers rises and materials are upgraded, the “price per square meter” of PCBs also increases.

NVIDIA expects Rubin GPUs to enter mass production in the second half of 2026.

NVIDIA’s Vera Rubin architecture will use an orthogonal backplane, a specially designed PCB, to replace traditional copper cables. The backplane is made by laminating three 26-layer PCBs or four 26-layer PCBs, paired with a high-end material mix including M9 resin substrates, advanced HVLP4 copper foil, and Q cloth, or quartz fiber cloth.

As a result, orthogonal backplanes are expected to scale up in 2026, bringing growth to PCBs and raw-material segments such as CCLs.

In addition, NVIDIA’s second-generation Rubin Ultra NVL576 platform is planned for release in the second half of 2027. Its ultra-high-layer PCB backplane is expected to be designed with M9+Q cloth or M9.5+Q cloth materials. If adopted, this board material would open another new growth space for NVIDIA AI server PCB demand.

Notably, high-end PCB capacity is expected to remain tight in 2026. According to estimates by China Merchants Securities, listed domestic companies have about 120 billion yuan of effective PCB capacity that can match AI demand, while demand is expected to reach around 150 billion yuan.

Miaotou therefore believes that because GPU architectures continue to iterate, the PCB industry has strong growth expectations in 2026. However, PCB companies also posted large gains in 2025.

Combining current earnings forecasts and valuations for some PCB companies, we can see that many PCB companies trade at P/E ratios near or even above 50 times. They are currently in a highly valued state and will need high earnings growth over the next two years to digest those valuations.

Data source: Wind

Miaotou believes that if relevant PCB companies can also achieve high growth in line with 2026 expectations, their valuations still have room to rise. NVIDIA Rubin GPU mass production in 2026 will be an important factor affecting earnings growth and valuation re-rating for related PCB companies, and investors should watch it closely.

After all that, how should investors rank the thematic opportunities across AI segments?

Miaotou believes AI segments can be ranked from solid to speculative based on investment certainty.

The liquid cooling industry has a high growth rate and may produce “dark horses” that enter Google’s or NVIDIA’s supply chains, so it can be rated “solid.” GPUs, optical modules, and PCBs have high growth certainty but also high valuations, requiring above-expectation drivers for valuation re-rating, so they can be rated “top tier.” Because of expectations around price increases and technological breakthroughs, optical chips have slightly higher elasticity than optical modules and can be rated “premium.”

AI edge devices and AI applications, however, do not have high certainty in 2026 and need new catalysts, such as the launch of a new blockbuster AI edge product or a new hit AI application. Once catalysts appear, AI edge devices and AI applications will have large expectation gaps and elasticity. For now, however, they can only be rated “NPC.”

In addition, expectations of a looser macro environment could bring a liquidity premium to technology sectors represented by AI computing power. The Federal Reserve is likely to cut rates twice in 2026, which could improve liquidity for A-shares, and high-growth sectors such as AI computing power may receive higher valuations from the market.

Miaotou believes that in 2026, companies across AI computing power segments such as liquid cooling, optical chips, GPUs, PCBs, and optical modules are likely to deliver varying degrees of earnings growth. If growth exceeds expectations, related companies could achieve a Davis double play under expectations of easier liquidity.

Overall, AI computing power offers high certainty and the potential to beat expectations. Compared with other tracks, it is more likely to generate excess returns and remains worth watching in 2026.