Economic Observer reporter Wang Yajie

May 2026, Wuxi, Jiangsu. Zhang Lei, the business leader of a leading accounting leasing enterprise, has just signed a three-year long-term agreement worth 5 billion yuan. The customer is a leading Internet company. The rental rate of his high-end computing cluster has reached 100%, and orders have been scheduled until 2028.

1300 kilometers away in Zhongwei, Ningxia, Ma Hongyuan, a small business owner with computing power, is staring at the server monitoring on his phone. He only has a few hundred units of domestic computing power (1PFLOPS, which is equivalent to one quadrillion floating-point operations per second), and his clients are small entrepreneurial teams and individual developers in fields such as artificial intelligence and software development. He calculated that if the utilization rate of computing power is below 70%, he will lose money.

Looking back at Wuxi, Lee, the business manager of AIDC (Artificial Intelligence Data Center) at Wuxi Lianyun Century Technology Co., Ltd. (871315, hereinafter referred to as "Lianyun Century"), a company listed on the New Third Board, described the current industry as follows: "The domestic computing power leasing industry is entering a period of explosive growth dividends, with market demand continuing to expand, but the supply of high-quality computing power resources is relatively scarce, and industry competition is showing a white hot pattern. Faced with fierce competition from a large number of homogeneous service providers, we must take the lead in matching resources

Public information shows that Lianyun Century is a computing power leasing service provider. After obtaining computing power resources in bulk from upstream, it provides them to enterprise customers in need in the form of leasing.

In a government building in a city in the Yangtze River Delta, Mr. Zhou, who is responsible for attracting investment in the computing industry, is reviewing materials from another computing enterprise applying for household registration. In the past two years, he has eliminated more than a dozen speculators who ran away after obtaining subsidies, and has also witnessed several smart computing centers that invested over a billion yuan in construction funds but ultimately failed. The wind is a bit too strong now, we can't let everyone rush in, "he said.

The rise of computing power leasing began with the global big model craze ignited by Chat GPT in 2023. In the past three years, the application of artificial intelligence (AI) in China has shifted from a "hundred model war" to industrial landing, and the demand for computing power has exploded accordingly. According to data from the China Academy of Information and Communications Technology, the size of the computing power leasing market has jumped from 32.8 billion yuan in 2025 to 68 billion yuan in the first quarter of 2026, and is expected to exceed 260 billion yuan for the whole year.

The computing power trend has arrived. But the wind doesn't blow evenly onto everyone. In the field of computing power, participants are all competing for resources, customers, and time. And the government, in the words of Lao Zhou, wants to seize 'order'.

Grab Resources

The underlying logic of computing power leasing is first and foremost whether there is a card or not.

Card "refers to GPU (Graphics Processing Unit, here referring to the computing chip used for AI training). High end GPUs, including H100, B100, A800, etc., are the "hard currency" for training large models. However, due to US export controls and chip production capacity limitations, these cards have long been in short supply. According to data from the China Academy of Information and Communications Technology, the high-end intelligent computing gap reached 35% in the first quarter of 2026, with a shortfall of 430000 units for just one H100 model.

Whoever has a card has the right to speak.

Zhang Lei's company is the type of player standing at the top of the resource chain. Its qualifications are unique, with a card acquisition cycle that is 6 to 12 months shorter than its peers, and a cost that is 30% to 50% lower than that of small and medium-sized players.

His strategy is very clear: only focus on high-end training computing power, without touching on low-end reasoning computing power and the domestic card market. When he crossed over into the computing industry in 2023, he set this iron rule. The low-end market is oversupplied, the price war is fierce, and the gross profit margin is as thin as paper. We have no advantage in entering. On the other hand, the high-end training computing power is subject to export controls and chip production capacity limitations in the United States, resulting in long-term supply shortages

This determination is first reflected in customer selection. When many small and medium-sized business operators worry about getting customers, they only serve a few leading Internet manufacturers and top model enterprises, and small and medium-sized customers will not accept them.

Top customers have sufficient budget, stable demand, and zero default risk. They can accept high pricing and long contracts, "Zhang Lei said." The core of what we sell to top customers is not 'computing power', but 'certainty' - stable high-end computing power, reliable operations and maintenance, continuous technical support, and uninterrupted services

This certainty has earned his company a three-year long-term agreement worth billions of yuan from a major manufacturer. In the first quarter of 2026, his company's dispatchable computing power reached 38000 P, including 10000 P owned and 28000 P subleased, with over 7 billion yuan in orders on hand and a stable gross profit margin of 53% to 60%.

The company has also developed its own immersion liquid cooling technology, which has reduced the PUE (Power Utilization Efficiency) to 1.09 to 1.1, far below the industry average of 1.5. The density of a single cabinet reaches 100kW (kilowatts), which is twice that of traditional air cooling and reduces power consumption by 30%. Relying on the precision machining capabilities of traditional manufacturing, we produce liquid cooled hardware with a gross profit margin of over 60%, forming an integrated closed loop of "computing power+liquid cooling".

But not everyone can grab the resources.

Ma Hongyuan's computing team consists of only 8 people, holding hundreds of domestic computing power, including domestic AI chips such as Ascend 910 and Haiguang DCU. He dare not touch H100 and A800.

Ma Hongyuan said that firstly, he couldn't get the quota, and secondly, it was too expensive, with one card costing hundreds of thousands of yuan, so he couldn't throw it at his hands.

His survival logic relies on the cost advantage of the West. The electricity bill in the eastern region ranges from 80 cents to 1.2 yuan per kilowatt hour, while in the central area it ranges from 30 cents to 40 cents per kilowatt hour, with a difference of more than half. 100P computing power, the electricity cost in the western region is less than 20000 yuan per month, while in the eastern region it is 30000 to 50000 yuan. Combined with the policy subsidies of "counting from east to west" and venue reductions, these "hidden profits" constitute the foundation for his survival.

But he calculated it: he had 50P of domestic computing power in hand, and if all of it was rented out, at a rate of 3000 yuan per P per month and a utilization rate of 70%, the monthly income would be 105000 yuan. But the cost of obtaining the goods is 100000 yuan, plus the cabinet electricity fee and maintenance bandwidth, the monthly cost exceeds 110000 yuan. If the utilization rate is below 70%, there will be losses; if it is above 80%, there will be a small profit; if it is above 90%, there will be some profit

Caught between top computing equipment suppliers and small and medium-sized computing leasing enterprises in the western region is Lianyun Century, headquartered in Wuxi, Jiangsu. In 2025, the company's revenue from computing related businesses will be approximately 13.88 million yuan, accounting for 12.05% of the total revenue. The enterprise is currently focusing on integrating upstream computing resources, supplemented by its own computing power. It is gradually increasing its investment in self built and self owned computing power, adopting a service model of "computing power leasing+supporting operation and maintenance+customized solutions".

Lianyun Century has built its own intelligent computing center in Wuhu, Anhui Province, with a planned 10000 card cluster to be delivered and put into operation by the end of 2026.

"The outside world defines CALI Leasing Enterprise as' CALI Second Landlord ', and this positioning is only applicable to our business model before 2020." Lee said, "From 2021, we will no longer only do the simple light asset sublease business, and continue to invest heavily in AI computing field. At present, Lianyun Century is building its own data center in Wuhu at the node of' East Counting West Counting ', with an overall planning of 120 mu. The first phase of the project is expected to deliver 40 MW by the end of 2026, and the second phase is expected to deliver in 2027. After the project is officially put into production, it will completely break away from the traditional single model of' Second Landlord '. ”

Grab Customers

Resources are the supply side, while customers are the demand side. The wind is blowing, and customers are also dividing.

The top customers are the top Internet manufacturers and top AI model companies. Their needs are "certainty", stable high-end computing power, reliable operation and maintenance, continuous technical support, and uninterrupted services. They have sufficient budget and are willing to pay a high premium for this certainty.

With this "selling certainty" strategy, Zhang Lei's company has won a 3-year long-term agreement worth billions of yuan from a large factory, with a 100% occupancy rate and orders scheduled until 2028.

The second layer consists of small and medium-sized enterprises in the Yangtze River Delta, including digital transformation of manufacturing, implementation of AI applications, financial technology, and healthcare. These customers have computing power needs, but they cannot afford to buy, maintain, or operate the entire cluster. What they want is' flexibility, affordability, and someone to manage '.

This is the main battlefield of the United Cloud Century. Lee's customers cover financial payment enterprises, quantitative private equity companies, medical manufacturing leaders, Internet platforms, and game companies. He serves a financial payment company with over 30 million daily transactions; A quantitative private equity firm with a management scale of approximately 4 billion yuan, relying on AI strategies for investment.

Lee's strategy is to achieve the ultimate goal of "localization and customized services". The services provided by large factories are standardized and remote, and the response is slow. We have a localized team, and customers can come to our doorstep in a timely manner if they have any problems, "he said.

Among his customer base, AI startup teams account for about 40%, small and medium-sized enterprises account for about 35%, and government enterprises and research institutes account for about 25%. AI startup teams have strong demand, large fluctuations, short cycles, and tight budgets, pursuing high-end cards and mainly doing model training; Small and medium-sized enterprises have stable demand, long cycles, and pursue cost-effectiveness, mainly engaged in AI reasoning and data processing; Government and enterprise clients have sufficient budgets and pursue stability, security, and compliance, mainly focusing on scientific research and government AI.

In the Yangtze River Delta, another private computing power service provider has also sensed structural opportunities. The company is rooted in Hangzhou and provides local 7 × 24-hour operation and maintenance, on-site docking, and small batch customization. The computing power service price is more than 30% lower than that of large factories. Zhou Bo, a representative of the company, told the Economic Observer: "In the past, when talking to clients, they asked 'what is computing power?' Now when talking to clients, they come up and ask 'how many H100 do you have, can you lock it for three months?'. This kind of gap is only understood by those who have experienced it firsthand. ”

Zhou Bo's clients are mainly small and medium-sized enterprises in the Yangtze River Delta. The pressure he felt came from two directions: one was that upstream large factories were sinking down to grab small and medium-sized customers. "They have funds, resources, brands, and technology, which our medium-sized companies cannot compete with," he said; The second issue is that the supply of goods is becoming increasingly unstable. For example, we once locked up a batch of A800, but before delivery, the upstream cut 30% of the quota and the price increased by 15%. We had to fulfill the contract to the customer while bearing the cost increase, and we had to bear the price difference ourselves. ”

The third layer is "scraps", consisting of local small businesses in the west, small entrepreneurial teams in the east, and individual developers. Their budget is limited, requiring short-term rentals and flexible payments, which are not favored by large companies, and even medium-sized companies find it troublesome.

This is Ma Hongyuan's market. His client haggled fiercely, saying that other companies only charge 2800 yuan per P. He can only explain that 'we provide maintenance and technical support, with a 7x24 hour response time, while other companies rent without paying attention to any issues'. After grinding for an hour, the final transaction was 2900 yuan and a one month short contract was signed. It's better to earn some hard-earned money than to be empty, "said Ma Hongyuan.

He is also trying to provide value-added services, helping customers with 7x24 hour on duty, fault repair, model deployment, parameter tuning, and charging on a per use or monthly basis. Subleasing earns hard-earned money, service earns profit. Without service, small players cannot survive, "he said.

Three tier clients, three-tier strategy. Ma Hongyuan said, "Eating meat on the head and drinking soup by oneself, there is still sand in the soup, which needs to be filtered slowly. But once it is filtered clean, it can also quench thirst

Race Against Time

Both resources and customers have been seized, the rest is time. Because the industry changes too quickly, if it's slow, it's gone.

Zhou Bo used one word: 'snatch'. Grab time, grab landing, grab customers.

Lee described it this way: "At present, the whole computing lease industry is in a stage of blowout growth and highly active market.

His team is able to promptly visit customers when they have problems, a response speed that cannot be achieved by standardized services provided by large companies.

Zhou Bo added, "The thing that big companies find troublesome is our rice bowl. But this rice bowl needs to be more stable as it gets more and more high-end

The way Zhang Lei's company competes for time is different. It doesn't seize a moment, but locks the future 3 to 5 years with long orders.

Small and medium-sized players pursue fast in and fast out, short-term rental arbitrage, and we only do 3 to 5-year ultra long contracts, "Zhang Lei said." The core value of long-term orders is to lock in high gross profit. In the first quarter of 2026, high-end computing power rent increased by 20% to 30%, and long-term orders directly lock in the high gross profit after the price increase. Secondly, the cash flow certainty is extremely strong, customers prepay 30% to 50% of the deposit, receive monthly payments, and the prepayment can cover about 60% of the equipment cost

But he also has anxiety. The iteration cycle of high-end GPUs is only two to three years, and once new cards are released, the prices of old cards plummet. At the end of 2025, Nvidia will release a new B100 card, and he has signed all long-term orders for H100 in advance for 3 years, while prioritizing the B100 quota to smoothly pass the technical iteration period.

Short term is a game, long-term is a monopoly. Resources are concentrated at the top, and ultimately a few leading companies monopolize the high-end market. The computing power leasing industry is a natural monopoly industry, and scarce resources determine that it is impossible for all to bloom. It is highly likely that the leading companies will take all, "Zhang Lei judged.

Those who cannot seize the time are being eliminated.

Ma Hongyuan's biggest fear is idleness. The most tragic thing is that high prices lock up resources, customer projects become yellow, computing power is idle for two or three months, and they lose money every day. His response is to prioritize risk control, and small customers try to make short-term contracts and prepay deposits as much as possible; Idle computing power should be immediately posted on the industry community to quickly find the next one.

The industry reshuffle is already accelerating. Zhou Bo said, "After 2024, outsiders and capital will rush in, thinking that computing power is the trend and can make quick money. But they don't understand the industry or risk control, and only rely on throwing money and low prices to grab customers, causing market chaos. The most common is blind expansion leading to the breakage of the capital chain, heavy investment in locking resources, renting computer rooms, recruiting people, resulting in customers not keeping up, serious idleness, and direct bankruptcy when the capital chain is broken

His judgment is that from 2026 to 2027, the industry will accelerate its market capture and reshuffle, and low-end pure second tier landlords will die. Only mid-range service providers with services, technology, and localization capabilities can survive.

Lee also holds the same view: "In the current market environment, the extensive 'second landlord' model that relies solely on computing power subleasing and hardware leasing is no longer sustainable and is gradually being eliminated by the industry. With the continuous tightening of industry regulation and the acceleration of standardization, industry competition has officially shifted from resource reselling to deep competition in comprehensive services and compliance capabilities

Grab Order

According to Lao Zhou, the trend in the field of computing power is too strong now, and everyone wants to rush in. The role of the government is to find a balance between supporting innovation and preventing risks.

Lao Zhou is a government worker responsible for the digital economy and computing power industry in a city in the Yangtze River Delta. He has been rooted in the front line of computing power investment, industrial planning, and risk control for five years. He has seen the frenzy of capital pouring in and companies flocking to the scene, as well as the loneliness of capital chain breakage and project abandonment after blind expansion.

Lao Zhou roughly divides the computing power companies in the market into two categories: one is the "gold diggers", companies like Lianyun Century that deeply cultivate the local market, do not blindly expand, and make solid money. "They are not greedy for short-term profits, control gross profit at 25% to 30%, and do not engage in vicious price wars. These companies are the ones we value and are most willing to support." The other is the "impetuous", a large influx of outsiders and cross-border capital during the boom period. "They do not understand the industry or risk control, only see 'huge profits from computing power', invest money to enter the market, blindly expand, and speculate and arbitrage." Lao Zhou gave a real case: a non local enterprise landed in early 2024 and received a subsidy of 5 million yuan. I rented a 1000 square meter venue, but I didn't have a local team or customers. I just sublet the computing power from other places, and after six months, I ran away due to severe idleness and a broken capital chain, Leave behind abandoned sites and a pile of debts. We have already shut down more than ten companies of this kind. Most of them do not survive for more than a year

The response mechanism of Lao Zhou is a combination of "screening+assessment+supervision".

In the investment promotion process, strict due diligence is required: check the team background, industry experience, resource channels, and financial strength; Check if there is self owned computing power, localized team, and local customer resources. Lao Zhou said, "For pure arbitrage and idle trading, no subsidy will be given

Subsidies are not issued in one lump sum, but are assessed and implemented over three years: in the first year, 30% of local revenue, employment promotion, and compliance standards will be implemented; In the second year, the industrial chain will drive compliance and pay an additional 40%; In the third year, all targets will be completed and the remaining 30% will be fulfilled. Do you want to run halfway? The venue will be reclaimed and the subsidy will be recovered, "said Lao Zhou.

Compliance is the bottom line that Lao Zhou values the most. He introduced that in national hub nodes and energy intensive areas such as the Yangtze River Delta, the construction of new large-scale data centers requires PUE (power utilization efficiency, which is an indicator for evaluating data center energy efficiency, and the closer it is to 1, the more energy-efficient it is) ≤ 1.25 and a green electricity proportion of ≥ 80%. Our energy consumption indicators are particularly tight, and many companies have started construction without obtaining energy consumption approval, and have been ordered to shut down in the end. ”

In 2025, the city where Lao Zhou is located will shut down two illegal enterprises, one with a PUE as high as 1.8, seriously exceeding the standard; Another company's data security is not compliant. Both companies were shut down, and the initial investment of several million yuan was basically wasted.

Lee also confirmed this trend: "This year, the requirements for industry compliance supervision are more refined and strict than in previous years, and the entire industry is undergoing standardized rectification

Establish rules in this trend and let companies that truly want to work stay, "said Lao Zhou.

(At the request of the interviewee, Zhang Lei is a pseudonym in the article)