The US stock computing power chain has exploded again, and the AI infrastructure has entered a full expansion cycle

The Zhitong Finance App learned that on September 3, local time, the US stock technology sector rebounded strongly, and the AI computing power chain collectively strengthened. Dell Technology surged 15.81%, Nvidia rose 3.21%, Micron rose 2.43%, and the NASDAQ rose 1.40%.

In this round of rebound, chip-only stocks were not leading the way; server manufacturers were more flexible. As one of the core suppliers of AI servers in the world, Dell Technology's sharp rise in stock prices has strong industrial implications. Its performance once again verifies that AI capital expenditure is being transferred from the chip sector to the wider infrastructure sector. The future development of the AI industry requires not only more powerful computing chips, but also computing power infrastructure that can support large-scale application implementation.

From chip competition to computing power system competition, the AI industry has entered an infrastructure expansion cycle

Over the past two years, market attention has focused on large model capabilities and GPU supply, and chip companies such as Nvidia are the core beneficiaries of the AI wave. Today, the focus is expanding to the back end: the importance of high-speed connectivity, AI networks, and data center infrastructure continues to rise, and companies such as Broadcom continue to benefit from AI data center construction and growing demand for high-speed connectivity. From GPUs to accelerators, from servers and high-performance storage to network devices and data centers, the AI infrastructure industry chain is becoming complete.

The reason behind the proliferation is a change in the structure of demand. If the core issue of AI was “can big models be trained” before, the question at this stage has become “how can big models be widely used”. Training requires concentrated high-performance computing power, while reasoning requires a continuous, stable, and efficient supply of computing power; as AI applications enter physical industries such as manufacturing, finance, medical care, and energy, computing power demand is shifting from phased investment to long-term infrastructure requirements.

Accelerate the layout of the domestic computing power industry chain, and deploy equipment and intelligent computing operations simultaneously

Changes in overseas capital expenditure were first reflected in domestic chip and server links. Companies such as Inspur Information (000977.SZ), IFF (601138.SH), and Zhongke Shuguang (603019.SH) continue to deploy AI servers and high-performance computing to provide underlying support for large-scale model training and industry applications. The Cambrian Era (688256.SH) is a sample of domestic chips: in the context of limited supply of high-end GPUs, its cloud-based training and inference chips continue to iterate. Customers cover leading Internet companies and intelligent computing center operators, and revenue is rapidly rising along with domestic computing power purchases — “domestic replacement” has changed from a theme to an order.

The acceleration in computing power infrastructure is being reflected in the mid-term results of domestic companies. Runze Technology (300442.SZ) focuses on AIDC business: revenue of 3,746 billion yuan in the first half of 2026, of which AIDC revenue was 1,995 billion yuan, up 126% year-on-year, accounting for more than half of the first time, with a gross profit margin of 47.3%; net profit to mother was 1,203 billion yuan, and the ratio of operating cash flow to revenue reached 79%. The company operates about 750MW data centers and has delivered about 100,000 high-density liquid-cooled cabinets, with a PUE as low as 1.08. Customers include ByteDance, Alibaba Cloud, Tencent Cloud, etc., and have achieved a heavy asset cycle exit through REITs.

Co-Creation Data (300857.SZ) is moving from smart terminal manufacturing and trading to computing power services. Its main business includes equipment sales and computing power technology services: revenue of 12.523 billion yuan in the first half of the year, an increase of 153.3% over the previous year, net profit to mother of 1,838 billion yuan, and ROE of 34.3%. The company plans to invest 8 billion yuan in intelligent computing centers, aiming to reach 50,000 P of computing power (FP16 dense) by the end of 2026, and extend to the MaaS model through platforms such as fCloud and TokenShare.

Litong Electronics (603629.SH) focuses on server resource trading and computing power services: net profit of 702 million yuan in the first half of the year, an increase of 1275% over the previous year, and a comprehensive gross profit margin of 46.6%, which is the highest in the sector; revenue from computational power-related services was 1,274 million yuan, accounting for 61% of revenue. The company holds Nvidia Preferred qualification, operates 38,000P of high-end computing power, keeps the computer room fully rented for a long time, and has signed a 5 billion yuan three-year contract with Tencent.

Hong Kong stock computing power services have ushered in new opportunities, and the business model has moved from resource construction to operational upgrading

As the AI industry enters the large-scale application stage, companies related to the Hong Kong stock market are also developing their layout around cloud services, computing power operations, and AI infrastructure services.

Jinshan Cloud (03896) has continued to strengthen AI cloud service capacity building in recent years and promote the upgrading of traditional cloud computing services to AI infrastructure services. As one of the earliest independent cloud service providers in China, Jinshan Cloud has fully embraced AI in the past two years: relying on the natural scenario of Xiaomi and Jinshan ecology, AI-related revenue continues to grow, and intelligent computing cloud services have become the main engine of growth; at the same time, it has continued to increase investment in AI infrastructure to strengthen full-stack capabilities from IaaS to MaaS. As enterprise demand for artificial intelligence applications continues to increase, enterprises with cloud platform capabilities, customer resources, and ecosystems are expected to gain new growth space in the AI industrialization process.

In the field of computing power services, Guangdong-Hong Kong Bay Intelligent Computing (01396) is gradually upgrading from a large-scale computing power technology service provider to a “super token factory”. The company's layout revolves around AI computing power services and operations, and promotes the transformation of computing power from basic resources to industrial service capabilities through the integration of computing power resources, the construction of computing power cloud platforms, and application ecosystem expansion. Revenue for the first half of the year was 2.56 billion yuan, 10 times that of the same period last year; net profit was 280 million yuan, five times that of last year. At present, the scale of computing power put into operation by the company has exceeded 50,000 P (FP16 dense), accounting for more than 2% of the country's total computing power. It plans to expand to 8 to 100,000 P by the end of the year; it is also promoting the integration of computing power resource scheduling and application services through the “quantum power” computing power cloud platform. More than 200 corporate customers have signed up, as well as more than 3,000 OPC and individual customers.

The continued strengthening of the computing power chain market is a microcosm of the expansion of AI capital expenditure, and the direction of capital expenditure is extending from chips to servers, data centers, and computing power operations. For a company, scale is just a ticket — order support, delivery capacity, resource efficiency, and profitability are variables that determine long-term value.