Every time an AI trains a new large model, the price of this "acid" goes up once: A full breakdown of the overlooked price surge chain in semiconductor materials

On June 29, 2026, the domestic quotes for UP/UPS grade electronic-grade hydrofluoric acid rose to 7,885 yuan/ton and 8,750 yuan/ton, up about 19% and 17% respectively from the beginning of 2026. During the same period, the market price of semiconductor-grade hydrofluoric acid from enterprises such as Duofuoduo increased by about 20% to 30%, and the capacity utilization rate remained at a high level.

While the market's attention is focused on Cambricon's market value exceeding one trillion and the consecutive daily limit increases of leading optical module companies, a more hidden clue is emerging: The accounts of AI computing power expansion are being transmitted along the industrial chain all the way to the few "chemicals" at the most upstream. This article will break down and clarify this overlooked price increase chain of semiconductor materials.

One bottle of acid can hold up a batch of wafers

Let's first explain a common sense: A chip cannot be used just by being "designed". It has to go through thousands of processes in a wafer fab, including repeated cleaning, etching, and film formation. These processes rely on a group of "electronic chemicals", such as electronic-grade hydrofluoric acid, electronic special gases, photoresists, and polishing solutions. Although they are used in small quantities and have low unit prices, they run through the entire manufacturing process and are truly "consumables".

Electronic-grade hydrofluoric acid is one of them, mainly used for wafer cleaning and etching. It has extremely high purity requirements. The UP and UPS grades correspond to different impurity control standards. The higher the grade, the more difficult it is to produce. Once the supply of such consumables is tight or the price increases, the cost and delivery schedule of wafer fabs will be affected. This is the meaning of "one bottle of acid can hold up a batch of wafers".

Where does the price increase start?

The starting point of this round of price increase is the explosion on the demand side. The training and inference of large AI models require a large number of high-performance chips; the increase in chip production forces wafer fabs to operate at full capacity; the full production of wafer fabs directly drives the consumption of consumables such as electronic-grade hydrofluoric acid and electronic special gases. The demand moves first, and the price follows.

The supply side is relatively rigid. The expansion cycle of high-purity electronic chemicals is long and the certification is strict, so it is difficult to increase production rapidly in the short term. With the supply tight and the demand loose, the price naturally goes up. In addition to hydrofluoric acid, MLCC (Multilayer Ceramic Chip Capacitors) is also increasing in price. Some securities firms judge that the price increase of this round of MLCC is expected to last for more than a year; power semiconductor manufacturers are also collectively raising their quotes.

Key data: On June 29, 2026, the price of electronic-grade hydrofluoric acid (UP/UPS grade) was reported at 7,885 yuan/ton and 8,750 yuan/ton, up about 19% and 17% respectively from the beginning of the year; the market price of semiconductor-grade hydrofluoric acid increased by about 20% to 30%, and the capacity utilization rate was at a high level; the price increase of this round of MLCC is expected to last for more than a year.

The more upstream, the more invisible

The most interesting point of this chain is that the more upstream the link is, the more invisible it is, but the more likely it is to be the "valve". The terminal chips and modules are shiny and the market discusses them every day; while the electronic chemicals that determine whether they can be mass-produced smoothly and at low cost have long been hidden in the corners of the financial statements.

However, it is precisely these "corners" that feel the heat first at the peak of the boom. The reason is that they are directly linked to the operating rate of wafer fabs: As long as the wafer fabs are in full production, the consumables need to be continuously consumed, and the price elasticity is actually more direct than some links. This is also why when the computing power narrative reaches its climax, funds will start to look for "underpriced" positions along the chain towards the upstream material end.

The following is for readers' reference. You can imagine AI computing power as a fully operational factory, where chips are products, wafer fabs are workshops, and electronic chemicals are the "fuel and cleaners" that are used every day in the workshops. When the products sell well, the first things to be snapped up and the first to increase in price are often these inconspicuous consumables. This is the general law of the industrial chain transmission. Specifically for companies and varieties, it depends on each company's actual production capacity, customer structure, and certification progress. This article does not make individual stock judgments.

Yield rate is the real battlefield

For wafer fabs, the core of competition for electronic chemicals has never been low price, but "stability and purity". A slight difference in purity may reduce the yield rate and scrap a whole batch of wafers, and the loss far exceeds the price difference of the materials themselves. Therefore, the moat of this track is long-term customer certification and batch stability.

This also determines that the beneficiaries of the price increase are those enterprises that have passed the certification of wafer fabs and can supply stably, rather than all related manufacturers. The higher the threshold, the more the price increase can be converted into profits; the lower the threshold, the more it can only make a fuss. This is two sides of the same coin as whether domestic substitution can be truly implemented.

The second half of domestic substitution

Looking at the long term, electronic chemicals are one of the most difficult and most crucial pieces of the domestic substitution puzzle. For a long time, the right to speak of high-end electronic chemicals has been in the hands of a few overseas manufacturers, and domestic manufacturers have been chasing in the mid - and low - end for many years. This round of volume and price increase driven by AI just gives domestic materials a window of "using volume to support R & D and using R & D to improve purity".

In the past few years, domestic electronic chemicals have successively achieved breakthroughs in categories such as hydrofluoric acid and special gases; if this round of boom can continue, it means that domestic materials can not only get more market shares, but also use high gross profit to support R & D and climb to categories with higher purity and higher barriers. In the end, the computing power competition is not only about whose chips are stronger, but also about who can hold the entire "invisible supply chain" from materials to consumables in their own hands.

Portable framework · Semiconductor material value chain

Materials (hydrofluoric acid, special gases, photoresists) → Consumable consumption (linked to the operating rate of wafer fabs) → Yield rate (purity determines success or failure) → Capacity expansion and substitution. In a word, to understand this round of computing power boom, in addition to chips, we also need to look upstream along this chain. Whoever holds the consumables and yield rate holds the invisible valve.

What does this have to do with you?

First layer, if you are interested in technology and semiconductor investment, the significance of this article is to extend your vision from chips and optical modules to "invisible consumables" such as electronic chemicals, and use the two criteria of "whether it is linked to the operating rate of wafer fabs" and "whether it has passed the certification" to distinguish who is the real beneficiary and who is just riding on the wave.

Second layer, if you are in the semiconductor manufacturing or materials industry, this round of volume and price increase is a window of "using volume to support R & D": The demand is rigid and the certification barrier is high. Enterprises that can supply stably are more likely to convert the price increase into R & D investment and climb to categories with higher purity and higher barriers.

Third layer, even if you are just an ordinary user of AI applications, this chain is also related to you: From one bottle of acid to one chip, the cost and self - sufficiency degree of the material end will eventually be reflected in the price of computing power, affecting the cost and availability of every AI service you use.

What do you think about the price of this "acid" increasing every time a new large AI model is trained? Welcome to share your views in the comment section.

This article is from the WeChat official account "BT Finance" (ID: btcjv1), author: BT Finance. It is published by 36Kr with authorization.