Economic Observer reporter Zheng Chenye

In the first half of 2026, the power semiconductor industry experienced two rounds of intensive price increases.

Starting from the beginning of the year, China Resources Micro (688396. SH) issued a price adjustment notice in February, announcing a price increase of more than 10% for all categories of power devices such as MOSFETs and IGBTs. The new prices will take effect from the date of the notice; Shilanwei (600460. SH) followed up in March and raised prices by 10% across all product lines.

Global leading manufacturer Infineon also launched its first round of price adjustments in the first quarter of this year. According to the company's notice to customers, the price adjustments involve power switches and related chips, with an increase of 5% to 15%, effective from April 1st; Jiejie Microelectronics (300623. SZ) and Xinjieneng (605111. SH) also completed their respective price adjustments during the same period.

The second round of price adjustments is concentrated around July. Infineon issued another price adjustment notice to customers on May 26th, which will take effect on July 1st. This is the second round of price increases for the company this year; Another leading manufacturer, Texas Instruments, also launched its fourth round of price increases since 2025 on July 1st.

Domestically, starting from July 1st, China Resources Micro will increase the price increase of all categories to over 15%, while Silan Micro will also increase it to over 15% on the same day. Yangjie Technology (300373. SZ) will increase the price of all products by 10% to 15%, and Starr Semiconductor (603290. SH) will increase the price of IGBT and silicon carbide MOSFET modules by over 15%.

According to incomplete statistics from reporters, along with companies such as Xinlian Integrated (688469. SH) and Lion Microelectronics (605358. SH), nearly 20 domestic and foreign power semiconductor companies have issued price increase notices in the first half of the year.

Power semiconductors are chips responsible for energy conversion and control, found in mobile phone chargers, electronic control systems for new energy vehicles, and power supply modules for data centers. Almost all electrical devices rely on them. In the past few years, this industry has just experienced a fierce price war in China.

From 2022 to 2024, the slowdown in downstream consumer electronics demand coupled with overcapacity in the domestic mid to low end market has led to fierce price competition among power semiconductor companies, with some product prices falling by more than 50%. The gross profit margin of many domestic power device companies has significantly declined, and profit margins have been severely compressed.

But the industry memory of price wars seems to have not faded yet, and a wave of price increases is coming. During the interview, the reporter learned that the delivery cycle of mainstream power semiconductors has generally been extended to over 30 weeks, and the delivery cycle of some high-voltage devices is even longer. The industry has adopted a "distribution" model of allocating goods based on customer importance.

A semiconductor researcher from a large private equity firm in southern China told Economic Observer reporters that this round of price increases will continue until at least 2027, because three conditions - incremental demand brought by AI, continuous contraction of 8-inch wafer production capacity, and rigid cost increases across the entire industry chain - exist simultaneously and will not disappear in the short term.

An increasingly complex power supply system

Infineon has raised prices for two consecutive rounds in April and July, making it one of the most active international manufacturers in this price adjustment cycle. According to publicly disclosed information, Infineon has held the top position in the global power semiconductor market share for 22 consecutive years.

Currently, this company is adjusting its organizational structure due to the impact of AI. Pan Dawei, Senior Vice President and President of Greater China at Infineon, told Economic Observer that the company expects revenue from AI data center power solutions to reach approximately 2.5 billion euros in fiscal year 2027, compared to approximately 1.5 billion euros in fiscal year 2026. In addition, starting from July 1st, Infineon will adjust its organizational structure from 4 business units to 3, with the core change being to consolidate all power related products into one business unit to more efficiently connect with AI customers.

The reason why Infineon is investing more resources in AI is simple - the power consumption of GPUs is increasing generation by generation, and every watt of electricity needs to be converted and distributed through power semiconductors. The more power GPUs consume, the greater the amount of power chips used.

Pan Dawei introduced that the power demand of AI racks is rapidly iterating, with the first generation about 200 kilowatts, the current generation about 500 kilowatts, and possibly reaching 1000 kilowatts in one or two years. This also means that the power supply chips required for each server are increasing exponentially.

In traditional servers, a machine only requires a few power chips, with a value of less than $10. But in AI servers that adopt an 800V high-voltage DC architecture (the higher the voltage, the smaller the current at the same power, the finer the cable, and the lower the transmission loss), the number of silicon carbide (SiC, a third-generation semiconductor material with high voltage resistance and low loss) devices per device has increased from 10 to over 24, gallium nitride (GaN, another wide bandgap semiconductor material) devices have increased from about 40, and silicon-based driver chips have increased from dozens to 128.

An industry insider told reporters that the value of power semiconductors for a single AI server has increased to about $200.

The increase in usage is only one dimension of change, and the other dimension is the upgrading of the power supply architecture itself. In the past, data centers commonly used 48V or 54V low-voltage DC distribution, which was completely sufficient in the era when the power of a single cabinet was only a few kilowatts. However, when the power of a single cabinet climbed to several hundred kilowatts or even approached one megawatt, the bottleneck of the low-voltage solution became apparent: under the same power, the lower the voltage, the greater the current, and the thicker the copper cable required for the current, the higher the heat loss.

So, the core logic of the 800V high-voltage DC architecture that is often discussed in the industry now is to raise the voltage and lower the current, thereby reducing the amount of copper used, lowering transmission losses, and freeing up cabinet space for computing devices.

According to a research report released by renowned semiconductor market research firm Jibang Consulting on June 25th, Nvidia is developing an 800V high-voltage DC power supply solution for the Vera Rubin platform, with plans to complete stocking in the third quarter of 2026 and provide it as an optional configuration to customers. Nvidia's AI computing platform currently uses the Blackwell architecture, with Vera Rubin being its next-generation architecture.

The first product under the Vera Rubin architecture, Vera Rubin 200, has a single cabinet power consumption of approximately 225 kilowatts, while the higher performance version of Rubin Ultra under the same architecture will reach approximately 660 kilowatts. Jibang Consulting predicts that the 800V power supply solution will gradually expand its adoption after the launch of Rubin Ultra in the second half of 2027, and large-scale deployment is expected in 2028.

The traditional 48V scheme is no longer sustainable at such power densities. According to Nvidia's publicly available technical documentation, a traditional solution requires 200 kilograms of copper busbars for a 1 megawatt cabinet, while a 1 gigawatt data center requires 200000 kilograms of copper busbars alone.

Nvidia is promoting the upgrade of the 800V architecture, with the core goal of reducing copper consumption and transmission losses. According to official information from Siemens and Schneider Electric, both power equipment giants have partnered with Nvidia to develop a matching power supply solution for the next generation of AI servers.

The automotive industry is also undergoing the same high-pressure process. According to Wang Haojun, an analyst at Jibang Consulting, the penetration rate of new energy vehicle models on the global 800V platform has reached 15% in the first quarter of 2026, a year-on-year increase of 21%. During the same period, the global installed capacity of silicon carbide main drive inverters (the core component that converts battery DC power into AC power required to drive motors) reached 1.17 million units, accounting for 29% of the total installed capacity of new energy vehicle main drive systems.

During the interview, the reporter learned that the single cost of silicon carbide is currently about 1.3 times that of traditional IGBT (Insulated Gate Bipolar Transistor, one of the most core devices in power semiconductors), and the gap between the two has significantly narrowed.

Wang Haojun stated that the application of silicon carbide is expanding from mid to high-end car models to models priced between 150000 and 250000 yuan. The 400V platform is still the main force in the current market, but 800V is the fastest growing range, with the increase mainly concentrated in the main drive inverters, as the output power can easily reach hundreds of kilowatts, resulting in the highest single vehicle usage and value of power devices.

According to the calculation of Jibang Consulting, the estimated installed capacity of inverters for new energy vehicles worldwide will reach 29.9 million units by 2026, and is expected to exceed 53 million units by 2030.

The supply side cannot make up for it in the short term

The two high-pressure paths of AI and automobiles are unfolding simultaneously, and power semiconductors are facing increasing demand in both directions, with the supply side facing more severe pressure than the demand side.

The aforementioned semiconductor researcher told reporters that the outside world often focuses on how many power devices are used in AI servers themselves, but the squeezing of production capacity is the factor that has a greater impact on the industry. AI related products are occupying the wafer capacity originally allocated to power devices: Infineon's production line is switching the capacity of traditional IGBTs to MOSFETs (metal oxide semiconductor field-effect transistors, widely used in power management) for AI servers, TSMC and Samsung continue to withdraw from 8-inch mature processes, and domestic foundries are tilting some of their capacity towards storage chips.

During the interview, the reporter also learned that a domestic power semiconductor brand had an agreement with a leading OEM factory to supply 5000 wafers per month, but the actual delivery last month was less than 3000 wafers.

The researcher determined that the proportion of AI occupying power semiconductor wafer production capacity is "absolutely over 30%".

In addition, global 8-inch wafer production capacity has been continuously shrinking in recent years. 8-inch is the main production line for power semiconductors. Unlike the 12 inch advanced process used in mobile phone chips and AI chips, power devices such as MOSFETs and IGBTs do not have high requirements for line width, but have extremely high requirements for voltage resistance and reliability. The process characteristics of the 8-inch mature process are most compatible with it. Currently, the vast majority of power semiconductor products worldwide are still produced on the 8-inch production line.

Jibang Consulting predicts that the global 8-inch production capacity will further decrease by about 2.4% in 2026, and the contraction trend will continue at least until the first half of 2027. Morgan Stanley's research report "Power Semis - Supply Driven Upcycle" released on June 18th also showed that the capital expenditures of the world's top five power semiconductor companies have been declining for two consecutive years, with only a moderate rebound of about 11% in 2026. It takes 2 to 3 years for new production capacity to be put into operation and released.

In addition, energy storage is also becoming the third growth line for power semiconductors.

Jian Jian, an analyst at Jibang Consulting, told Economic Observer that the demand for power devices in energy storage PCS (energy storage inverters, the core equipment that converts battery DC power into AC power) is rapidly increasing. After 4-hour long-term energy storage becomes a global standard, the shipment increment of high-power IGBT modules is the most rigid.

Cheng Jiayu, Senior Vice President and Head of Industrial and Infrastructure Business at Infineon Technologies, also told reporters that it is expected that by 2026, the country will add 200 gigawatts of photovoltaic capacity, about 120 gigawatts of wind power, and nearly 100 gigawatts of new energy storage capacity.

She also mentioned that traditional transformers have been out of stock globally for over two years, and the price increase of copper and other metal materials continues to push up procurement costs. Solid state transformers (a new type of transformer that replaces traditional copper ferromagnetic cores with semiconductor devices) are becoming a new choice for connecting power systems and computing infrastructure.

That is to say, AI、 The demand lines of automobiles and energy storage are all increasing, but the production capacity of 8-inch wafers is continuously shrinking, which is the reality that power semiconductors are currently facing.

The situation is different between having a production line and not having one

The last round of power semiconductor price increases occurred from 2020 to 2022, triggered by the "chip shortage" in automobiles. The concentrated outbreak of demand for new energy vehicles and photovoltaic energy storage, coupled with panic hoarding in the consumer electronics sector, further amplified the shortage signal.

Taking STMicroelectronics, another leading company in the industry, as a reference, after reviewing public information, Economic Observer reporters found that STMicroelectronics had cumulatively raised its ex factory prices four times during that cycle, with each increase ranging from 10% to 15%. The ex factory prices of large factories have nearly doubled, and the increase in the bulk cargo market has reached seven to eight times.

But after the demand for hoarding receded, a large amount of accumulated inventory was sold off, and it took the entire industry two years to digest it.

Compared to others, the current round of price increases appears much milder: from April to July, the power semiconductor industry completed two rounds of price increases, each with a magnitude of about 10% to 15%, far lower than the level in 2021.

Industry insiders told reporters that the slope of this round of price increases will not be as steep as the previous round, but the duration may be longer, and there will be no problem with the industry's prosperity in the next two years.

Pan Dawei also told reporters that cost is also the main reason for product price increases. The geopolitical tension is causing various costs in the entire value chain to continue to rise, covering multiple fields such as energy, raw materials, and transportation, which the company itself cannot fully digest.

According to data from the Shanghai Futures Exchange, the main copper contract in Shanghai rose from about 95000 yuan/ton at the end of the first quarter to about 105000 yuan/ton at the end of May, while the main tin contract in Shanghai rose from about 350000 yuan/ton to about 420000 yuan/ton during the same period.

During the interview, the reporter also learned that the current price increase of power semiconductors mainly reflects costs, and the overall gross profit margin of the industry has not yet returned to the high point of the previous round. At the peak of the industry in 2022, the average gross profit margin of three leading IDM companies (i.e. design, manufacturing, packaging and testing integrated enterprises), including Infineon, STMicroelectronics, and Anson, reached 47% to 48%. However, by the first quarter of 2026, this figure had dropped to about 37%. Infineon expects its gross profit margin to recover to 40% to 45% in the 2026 fiscal year, approaching but not yet reaching the high level of the previous boom.

Enterprises with and without production lines face vastly different situations in this cycle. For example, in the first quarter of 2026, China Resources Microelectronics saw a year-on-year increase of 297% in net profit, while Silan Microelectronics saw a year-on-year increase of 41%. Meanwhile, the Fabless model used by Sida Semiconductor saw a 74% decrease in net profit during the same period. The reason is that when outsourcing costs rise, IDM companies with their own production lines can directly convert the price increase into profits, while pure design companies face rising wafer outsourcing and packaging costs, and terminal price increases may not necessarily be transmitted, squeezing their profit margins from both ends.

During the interview, the reporter learned that among thousands of power device design companies in China, about 80% are concentrated in the low-end MOSFETs with lower technological barriers, which are highly competitive and benefit far less from the price increase cycle than the top IDM companies.

During an investor survey in early July, China Resources Micro disclosed that the company's pending orders continued to reach new highs, with order visibility reaching up to 9 months. Some popular models have a pending delivery cycle of over a year. Yangjie Technology is also accelerating its production expansion. The company's packaging base in Vietnam has achieved large-scale production, with a monthly production capacity of 1.2 billion units. The Vietnamese automotive grade 6-inch wafer fab is expected to start production in the first quarter of 2027.

But in the field of AI server power supply, which has the largest incremental growth, the situation faced by domestic enterprises is different. Several industry insiders told reporters that the customer base for AI server power management chips is highly concentrated, with high supply chain barriers. Overseas brands still dominate in this field, and the actual participation of top domestic power device companies is limited.

Automobiles and energy storage are related to the current profit recovery of power semiconductor companies, while AI power supplies are related to longer-term market share, which may redefine the competitive position among domestic power semiconductor companies in the next two to three years.

(Author Zheng Chenye)