Economic Observer reporter Wang Yajie

As of the end of August, in a smart computing project in Nantong, Jiangsu, a whole batch of AI (artificial intelligence) servers have been piled up in the warehouse for more than four months without being invested in the smart computing project, with storage fees alone costing over 100000 yuan.

They are waiting for something in short supply - a special transformer.

Domestic dry-type transformer leader Jinpan Technology (688676. SH) signed new sales orders totaling 7.493 billion yuan in the first half of the year, a year-on-year increase of 69.31%. Among them, the data center field alone contributed 3.869 billion yuan.

The server has arrived, where is the transformer?

Economic Observer reporters interviewed transformer manufacturers, computing power leasing companies, data center operators, and local development and reform departments, and learned that the high-end production capacity of special transformers is being competed for by ultra-high voltage engineering, new energy grid connection, and overseas high premium orders. The gap between paper computing power and actual production computing power is widening in some local projects that focus on heavy cabinets and light power supply.

Orders Flood In

Zhao Jun is the head of the East China region at a domestic GPU (Graphics Processing Unit) server manufacturer. Some of the smart computing delivery projects he has recently been involved in are being delayed due to delayed transformer supply.

Zhao Jun told the Economic Observer reporter, "The servers are piled up in the customer's warehouse, and the packaging boxes have not been unpacked yet. Without a transformer, power cannot be turned on, and inspection cannot be carried out. Opening the boxes in advance, we were exposed to moisture and dust, and the responsibility was unclear. We will only receive the final payment after the transformer enters the site, is powered on, and passes the inspection

The popularity of computing power construction is becoming a production schedule for transformer factories.

The function of a transformer is to convert high voltage electricity from the power grid into voltage that can be used by equipment in the computer room. In the AI intelligent computing room, the power of a single cabinet has increased from several kilowatts in the past to tens or even hundreds of kilowatts today. The GPU load fluctuates greatly instantly, and the requirements for power supply stability far exceed those of traditional data centers. The production capacity of ordinary transformers is not tight, the real shortage is the large capacity, anti harmonic special model transformers customized for AI machine rooms. China is the world's largest producer and important exporter of transformers, accounting for over half of the global production. Unlike the shortage of chips and copper foils, the contradiction between chip and copper foil supply occurs before the server leaves the factory, and the contradiction between transformer supply occurs after the server arrives at the park, which is the last link before the computing power is put into operation.

The orders from Jinpan Technology are rapidly increasing. By the end of June, the total number of orders in hand of the company reached 10.697 billion yuan, a year-on-year increase of 41.87%, reaching a historical high, equivalent to three times the revenue in the first half of the year.

Overseas orders are growing faster.

In the first half of the year, Jinpan Technology's newly signed overseas orders amounted to 4.19 billion yuan, a year-on-year increase of 270.69%. The amount of overseas orders in hand was 6.239 billion yuan, accounting for 58.32% of the total number of orders in hand. The company disclosed in its investor relations activity record that its overseas order structure is mainly concentrated in the AIDC (Artificial Intelligence Data Center) field, with steady growth in new energy and a slight increase in demand for power grid updates and replacements.

The flow of orders is not uniform.

A senior dry-type transformer enterprise insider told Economic Observer that the delivery cycle of ordinary standard dry-type transformers is maintained at 2 to 3 months, with stable prices and reasonable production line utilization rates; The low harmonic special dry-type transformer adapted to AI intelligent computing scenarios requires re simulation design of the winding and iron core. The dedicated production line has been basically filled, and the delivery cycle is longer than that of ordinary models. The market is not experiencing a complete shortage, but rather a clear structural differentiation. Many customers use the expected delivery time of ordinary dry-type transformers to measure the computing power of special transformers, which amplifies the feeling of tight supply

The buyer is busy competing for a spot.

Zhang Lei, an industry insider in the computing power leasing industry, told the Economic Observer that "there is only so much capacity for special transformers. If the buyer does not sign a framework agreement in advance to occupy the position, they will not be able to get the transformer when the project is truly implemented." According to him, the common practice in the industry is for relevant enterprises to sign intention framework agreements with two or three transformer manufacturers at the same time, pay a small amount of earnest money to lock in potential production capacity, and after the project's energy consumption, financing, and land are implemented, select another manufacturer to sign a formal contract, and the rest of the framework will be invalidated.

Officials from the securities department of Jinpan Technology told Economic Observer reporters that the company's production line is already saturated, and orders from both domestic and overseas are queuing up.

Delivery Pressure

After the production schedule is filled, the real game begins.

Who should be given high-end production capacity first, and there is a clear sequence for production scheduling.

A person working for a large transformer company told an Economic Observer reporter, "The national key project of ultra-high voltage is a national strategic project and enjoys priority protection at the policy level in production scheduling. The remaining high-end production capacity can be used for market-oriented allocation between new energy, domestic computing power, and overseas orders." In some smart computing projects that his company has encountered, production has been delayed by 4 to 8 months simply due to transformer manufacturing and delivery issues.

He also said, "It's not that the factory can't produce the main transformer needed for the intelligent computing center, but the scheduling seats have already been filled with orders that took effect earlier

Overseas orders have taken up more seats.

The above-mentioned Jinpan Technology personnel told the Economic Observer reporter that overseas customers often sign long-term frameworks and have higher prepayment ratios. "When multiple large orders collide, delivery delays will occur.

A person from a leading power transmission and transformation equipment enterprise introduced that the company's overseas joint venture factory only produces transformers of 500 kV and below, serving the local conventional power grid market, and does not have the ability to produce large capacity customized main transformers required for AI intelligent computing centers. "Not only do domestic intelligent computing projects have to queue up, but high-end customized equipment with AI computing power demand overseas also needs to consume our domestic production capacity resources. The production lines capable of manufacturing customized main transformers with computing power on a global scale are mainly concentrated in China, and the gap is difficult to fill through imports.

Director Yu, who is in charge of the computing infrastructure project at the Development and Reform Commission of a central province, told Economic Observer reporters that the mismatch between heavy computing power scale and light power supply matching is quite common. "Many people think that transformers, like ordinary distribution cabinets, can be quickly fixed, but it will take nearly a year to realize after going through the procurement process. At this point, the project has already started and can only wait.

Zhang Lei revealed that the transformer framework intention orders disclosed to the public within the industry have truly been converted into formal contracts for locked production scheduling, with an overall conversion rate of between 30% and 50%. Projects in small and medium-sized areas may account for less than 30%. In order to compete for a higher production position, domestic customers either sign long-term framework agreements for more than three years, or increase the prepayment ratio from the conventional 10% to 20% or even 30%.

Queuing is becoming a monthly delivery cycle. Li Zhe, the business principal of AIDC, a data center operator, told the Economic Observer that in the era of traditional IDC (Internet Data Center), ordinary dry-type transformers were purchased on demand. Now the 110 kV main transformer (hereinafter referred to as "main transformer", which refers to the core transformer that converts 110 kV grid high voltage into usable voltage in the park) at the park side has a longer delivery cycle, and some customized high-voltage main transformers have been more than a year, "more than twice as long as before". The above-mentioned top power transmission and transformation equipment enterprise personnel remind that the causes of delayed production are often intertwined, with projects that have not implemented energy consumption indicators, delayed construction of external power grids, and transformers that cannot be powered on even if they are delivered on time; There are also a large number of projects that have been fully implemented in terms of energy consumption and access conditions, but ultimately the main transformer supply lags behind.

Solve The Problem

All parties in the queue have already started solving problems on their own.

The first step for manufacturers is to tighten their grip on the upstream.

Interviewees from multiple leading transformer companies shared the same fact: the bottleneck for expanding high-end transformer production is not in the factory building, but in materials and personnel.

The above-mentioned large-scale transformer enterprise personnel stated that the supply of high-end oriented silicon steel (a high-end electrical steel with grains arranged in the rolling direction, which is the core material of transformer cores) is tight, the procurement cycle is prolonged, and the winding of large capacity transformers highly relies on senior technicians, whose training period is three to five years.

The above-mentioned Jinpan Technology personnel stated that the factory can be built quickly, but it is difficult for skilled manpower to make up for it on a large scale in the short term. Jinpan Technology has already locked in prices and stocked up on high-end grades of oriented silicon steel and copper materials in advance to ensure production scheduling.

Building factories overseas is the second path. Jinpan Technology's Mexico factory is currently in operation to undertake overseas orders and reduce competition with domestic customers for local production lines.

A transformer company with production capacity in North America told Economic Observer reporters that the company's factories in Mexico and the United States are built according to North American standards, and the product specifications are not universal with domestic projects. Overseas production capacity is mainly supplied to overseas markets, and it is difficult to feed back domestic computing power orders.

Prefabricated cabin type substation is a solution for engineering, which involves pre installing and debugging transformers and switchgear in the factory, and connecting them to the site upon arrival, compressing the on-site civil engineering and installation cycle.

After trying this plan, Li Zhe calculated another account. The prefabricated cabin can only move the on-site installation work to the factory. The core transformer body in the cabin still follows the normal production scheduling cycle, so we have to wait.

From a technical perspective, solid-state transformers are considered a solution suitable for the next-generation 800 volt high-voltage DC power supply architecture, with higher conversion efficiency and smaller size.

The above dry-type transformer enterprise personnel supplemented the demand side situation, saying that at present, only part of the new intelligent computing clusters of the leading Internet cloud manufacturers have been implemented in the 800 V high-voltage DC architecture, and a large number of local projects are still in the pilot and planning stages.

The government is also making adjustments. Director Yu stated that the local authorities are optimizing the assessment methods, focusing more on actual production computing power, rather than the contracted scale and number of cabinets.

All these actions resulted in costs being included in the financial report. According to the semi annual report of Jinpan Technology, the company's net cash flow from operating activities in the first half of 2026 was RMB 0.24 million, compared to RMB 235 million in the same period last year; Inventory of 3.239 billion yuan, an increase of 25.08% compared to the beginning of the period; Short term loans amounted to 938 million yuan, an increase of 90.76% compared to the beginning of the period. The above-mentioned Jinpan Technology personnel stated that negative cash flow is a temporary phenomenon during the high-speed expansion stage of orders, mainly due to the increase in prepayments brought about by early price locking and stocking.

(At the request of the interviewee, Zhang Lei and Li Zhe are given pseudonyms)

(Author Wang Yajie)