Computing power is productivity: An infrastructure competition concerning the next ten years

During the trading session on May 12, 2026, the share price of Zhongji InnoLight Co., Ltd. (中际旭创) broke through 1,000 yuan, with its market cap reaching 1.1 trillion yuan. It became the first "1,000-yuan stock" in Shandong Province and the second such stock in the history of the ChiNext.

Meanwhile, Montage Technology Co., Ltd. (澜起科技) has seen its stock price double since the beginning of the year, and Cambricon Technologies Corporation Limited (寒武纪) has reclaimed the "Stock King" throne on the A-share market. The market has cast the biggest vote of confidence in history for AI computing infrastructure with real money.

However, the starting point of all this was not the enthusiasm of the capital market, but a strategic resolve at the national level.

From "Digital Infrastructure" to "National Strategic Infrastructure"

In 2026, computing power achieved a historic leap within the policy discourse system.

For the first time, the "15th Five-Year Plan" outline dedicates a separate section to computing power, listing it alongside communication networks and data networks as the three pillars of the country's new information infrastructure. Moreover, the Government Work Report has, for the first time, incorporated "computing-power-electricity synergy" into national-level new infrastructure projects. This means that computing power is no longer just a supporting foundation for the tech industry, but a national lifeline of equal importance to water conservancy, electric power, and transportation.

Subsequently, in February 2026, the Ministry of Industry and Information Technology (MIIT) officially deployed the "National Computing Power Infrastructure Project." The core is to build a "1+M+N" three-level node system: 1 national computing power internet service node, M regional nodes, and N industry-specific nodes. This will form a "single ledger" for national computing power, promote the coordinated scheduling of computing resources, and truly realize the vision of "using computing power like using water and electricity."

Data confirms the effectiveness of this strategic layout. As of the end of March 2026, the total scale of national intelligent computing power has reached1882 EFLOPS (FP16), the number of standard racks in use in computing centers reached14.45 million units, which has more than doubled compared to two years ago.

Capital arms race: 100-billion-level investment becomes the threshold

After the policy signals were issued, the response from the industry side was almost instantaneous.

ByteDance (字节跳动) has raised its AI-related capital expenditure for 2026 from the 160 billion yuan planned at the beginning of the year toOver 200 billion yuan, an increase of over 25%, of which the budget for AI chip procurement is approximately 85 billion yuan, while the remaining funds are mainly invested in global data centers and network infrastructure; Alibaba Group (阿里巴巴) CEO Wu Yongming publicly stated that capital expenditure over the next three years will "far exceed the previously promised 380 billion yuan," and the scale of computing center assets held by Alibaba Cloud (阿里云) will be more than ten times that before the AI outbreak in 2022; Tencent Holdings (腾讯) computing-related capital expenditure reached 31.9 billion yuan in the first quarter of 2026, a year-on-year increase of 16%. The three major operators contributed approximately 35% of domestic computing investment, serving as the ballast stone for the construction of the computing network.

Behind this competition lies a clear industrial logic: in the AI era, computing power reserves determine the upper limit of intelligence. Whoever builds the infrastructure first will gain the first-mover advantage in future model competition and application competition.

It is worth noting that the competitive landscape is shifting from a competition of "quantity" to the optimization of "structure".In 2026, global AI infrastructure spending is projected to reach 450 billion USD, with inference computing power accounting for over 70% for the first time.(Wu Hequan, an academician at the Chinese Academy of Engineering, pointed out that China's inference demand is even 8 times that of training demand). This means that the business logic of computing power has quietly shifted; in the past, it was about whether one could train large models, but now it is about whether one can withstand the peak inference load of hundreds of millions of users using AI simultaneously.

Computing-Power Coordination: An Underestimated Hidden Barrier

If computing power is the "new oil" of the AI era, then electricity is the indispensable fuel for the refinery.

The power consumption of a single cabinet in a high-density AI data center has already surged from 5-10kW in the traditional IDC era to 50-80kW or even higher. Behind the country's 14.45 million racks in use lies an increasingly heavy energy bill. The National Data Bureau has explicitly required that the proportion of green power consumption in newly built data centers at national hub nodes must be further improved on the basis of 80%.

"Computing-Power Coordination" has therefore become the new issue receiving the most policy attention in 2026. Its essence is to resolve the structural contradiction between the explosion in AI computing power demand and power supply constraints: through the two-way interaction between clean energy and computing facilities, allowing large-scale photovoltaic and wind power bases to supply power to western computing centers, and allowing computing centers to actively increase loads during power troughs to participate in grid peak shaving, ultimately achieving a "green, economic, and safe" computing power energy system.

Following the roadmap of the "15th Five-Year Plan," by 2027, an energy security system featuring computing-power-electricity synergy will be initially established. By 2030, the clean energy support capability for AI computing facilities and the effect of two-way empowerment between AI and energy will reach a world-leading level. This timeline implies that niche sectors such as liquid cooling, energy storage supporting facilities, and smart grids will remain in the spotlight of policy and capital for the coming years.

The market speaks: Capital has already voted in advance

The dual resonance of policy and industry has formed the strongest main market trend in the capital market in 2026.

The optical module sector can be described as the most direct beneficiary. Represented by the "Yi-Zhong-Tian" trio—Zhongji InnoLight (中际旭创), Eoptolink Technology Inc., Ltd. (新易盛), and TFC / Suzhou TFC Optical Communication Co., Ltd. (天孚通信)—the three leading optical module stocks have seen their share prices rise by over 65%, 32%, and 76% respectively since the beginning of the year. Their combined market capitalization once surpassed that of Kweichow Moutai, and the ChiNext board also saw the birth of the second 1,000-yuan stock in its history. The hard logic behind this is: the internal connection density of AI data centers is growing exponentially, leading to a surge in demand for 800G and 1.6T high-speed optical modules, and Zhongji InnoLight's (中际旭创) exclusive orders from NVIDIA are already scheduled through 2028.

The computing power chip sector is also welcoming a moment of performance realization. Montage Technology (澜起科技) has risen by over 100% year-to-date; Cambricon Technologies Corporation Limited (寒武纪) achieved its first annual profit five years after its listing in 2025 (net profit of 2.059 billion yuan), with a net profit of 1.013 billion yuan in the first quarter of 2026, indicating a significantly accelerated pace of profitability; Hygon Information Technology Co., Ltd. (海光信息) has become an important core holding target for institutional funds thanks to a balanced combination of scale and growth rate. The capital side confirms this judgment: since April this year, margin investors have net bought over 10 billion yuan in AI chip concept stocks, with Cambricon Technologies Corporation Limited (寒武纪) and Hygon Information Technology Co., Ltd. (海光信息) ranking as the top two in margin buying.

Overall, the optical chip concept sector has seen a massive surge over the past 120 trading days (January to May).134.51%, ranking first among all 714 concept sectors in the A-share market.

A new narrative that is taking shape

This computing power race is, in essence, a reconstruction of the infrastructure of civilization.

Over the past 20 years, cloud computing has transformed computing power from "dedicated resources" into "public resources." In the next 5 years, the construction of AI computing infrastructure will drive computing power from "public resources" to "factors of production." Just as electricity is to industrialization and the internet is to informatization, it will become an indispensable underlying support for the operation of the economy and society.

This is not a story about a certain company or a specific technical route, but a proposition of the times regarding the paradigm shift of infrastructure. Policy has sounded the clarion call, capital has completed its layout, and the industry is also delivering results.

For researchers and observers, the question truly worth pondering might be: when computing power becomes a basic resource like water and electricity, who is controlling the "gate," and who is setting the "water bill"?

Data sources: Ministry of Industry and Information Technology, National Data Bureau, SEMI, public financial reports of Zhongji InnoLight Co., Ltd. (中际旭创) / Montage Technology Co., Ltd. (澜起科技) / Cambricon Technologies Corporation Limited (寒武纪), Tencent Stock Quotes data, as of May 19, 2026.