Is the AI Industry Becoming a Specialty Chemical Customer?
Yes: AI data centers and chip fabs are becoming specialty chemical customers, and they buy on qualification and compatibility evidence rather than on price per liter. The opportunity holds for suppliers that answer technical questions quickly with test data, certificates and records, and the size of the demand rests on growth that the IEA presents as a forecast.
Electricity demand from data centers rose 17% in 2025, against 3% growth in global electricity demand (IEA, 2026).
The IEA forecasts that data center electricity consumption will double by 2030 and that power use from AI-focused data centers will triple (IEA, 2026).
Wafer fabrication materials revenue rose 5.4% to 45.8 billion US dollars in 2025, with wet chemicals among the segments that posted strong double-digit growth (SEMI, 2026).
Guidelines for propylene glycol coolants in liquid-cooled racks say fluid compatibility with wetted materials should be confirmed at the system's maximum operating temperature (Open Compute Project, 2026).
What did the IEA report about AI data center growth in April 2026?
The International Energy Agency (IEA) reported on 16 April 2026 that data center electricity demand surged in 2025 and that it expects the growth to continue to 2030. In the press release for its report Key Questions on Energy and AI, the IEA said electricity demand from data centers rose 17% in 2025, and that demand from AI-focused data centers climbed even faster, against 3% growth in global electricity demand (IEA, 2026).
The forward view is a forecast. The IEA projects that electricity consumption from data centers will double by 2030 and that power use from AI-focused data centers will triple, even as power consumption per AI task falls quickly (IEA forecast, 2026). It also reported that the capital expenditure of five large technology companies passed 400 billion US dollars in 2025 and is set to rise by a further 75% in 2026 (IEA, 2026).
The release names physical limits as well. Supply chains for gas turbines, transformers, advanced chips and IT components tightened over the year before the release, and grid connections and approvals are slowing new projects (IEA, 2026). Every one of those bottlenecks is a physical build-out, and physical build-outs consume materials. The IEA does not discuss chemicals; the link to specialty chemicals is the inference this post develops.
Which specialty chemicals does the AI industry buy?
The AI industry buys specialty chemicals in three places: the cooling loops of data centers, the electrical equipment that feeds them, and the fabs that make the chips. Each is a formulated product chosen for performance, not a commodity bought by the tonne.
In direct-to-chip cooling, a common choice is an inhibited propylene glycol fluid at 25% concentration (PG25), which runs in the closed loop between the coolant distribution unit and the cold plates on processors. The Open Compute Project's 2026 guidelines describe these fluids as formulated with corrosion inhibitors and typically antifoam agents and dyes, and say the type and concentration of inhibitors used in pre-fill cleaning should be determined by a qualified cleaning company or chemical supplier (Open Compute Project, 2026). Immersion cooling uses dielectric fluids in which the hardware sits directly. One chemical maker added synthetic immersion and cold plate grades in February 2026 and describes the portfolio as designed for material compatibility with next-generation hardware (Perstorp, 2026).
The power side adds insulating fluids for the transformers the IEA lists among tightening supply chains, a choice set out in how to choose transformer fluid type. In chipmaking, SEMI reported that wet chemicals posted strong double-digit growth in 2025 as process intensity rose, within wafer fabrication materials revenue of 45.8 billion US dollars (SEMI, 2026).
Why do qualification and compatibility decide AI infrastructure chemical sales?
Qualification and compatibility decide AI infrastructure chemical sales because a fluid that fails inside a server rack or a fab tool damages equipment worth far more than the fluid. The buyer therefore asks for proof before price.
The Open Compute Project's PG25 guidelines show what that proof looks like. Compatibility with every wetted material, from cold plates and pump seals to quick-disconnect seals, should be confirmed at the system's maximum operating temperature, and any later change of wetted materials must also undergo compatibility verification (Open Compute Project, 2026). Users are told to check the material maker's chemical resistance data and the fluid's technical data sheet (TDS), and, where uncertainty remains, to run a controlled compatibility test or contact the fluid supplier (Open Compute Project, 2026). Approved fluids should meet the corrosion rate criteria of ASTM D8040, run with modifications for pre-diluted fluids (Open Compute Project, 2026).
The 2022 OCP Immersion Requirements make the same point for the glycol coolant in the facility water loop: when the mixture contains proprietary additives, compatibility with hose and tube materials may be difficult to predict, and the best way to verify it is a physical test with the specific coolant (Open Compute Project, 2022).
The commercial consequence follows. A supplier whose product is not yet qualified is not in the bid, and a qualified supplier stays in only while its answers about additives, temperatures and materials hold up.
How does an AI infrastructure customer differ from a traditional specialty chemical account?
An AI infrastructure customer differs from a traditional specialty chemical account mainly in how much evidence it expects before and after the sale. The table below compares the two on the points that decide whether a supplier wins and keeps the account.
Question | Typical traditional industrial account (illustrative) | AI data center or fab account |
|---|---|---|
What gets the product in | Often a plant trial and a buyer's approval | Compatibility testing against listed wetted materials at maximum operating temperature (Open Compute Project, 2026) |
What triggers a new review | Often a failure, a complaint or a new line | Any change of wetted materials (Open Compute Project, 2026) |
What the customer keeps on file | Often data sheets and invoices | Fluid type and supplier, batch numbers, delivery dates, certificates of analysis and fluid analysis reports (Open Compute Project, 2026) |
Who may adjust the fluid | Often the plant itself | Only additives approved by the original fluid supplier (Open Compute Project, 2026) |
Some industrial accounts, such as OEM-approved fluids, already work this way; the AI data center difference is that the OCP guidelines write the re-verification trigger into the operator's own practice.
The row that matters most is the trigger for a new review. Because any replaced pump, seal or cold plate reopens the compatibility question, the supplier's technical answers continue for the life of the system, not only at the sale. A related pattern appears when a chemical product reorder needs requalification.
The record-keeping row changes the daily work of a technical service team. An operator that files batch numbers, certificates of analysis and fluid analysis reports for each cooling system will bring those reports to the supplier whenever a result drifts. A supplier that can pull the same history for that system, and show which seal and hose materials were tested at which temperature, answers in one reply. A supplier that has to rebuild the history from email answers late, and in a qualified market a late answer reads as a weak one.
Is AI infrastructure only a market for the largest chemical makers?
AI infrastructure is not only a market for the largest chemical makers, although makers with an established fluids business hold an advantage at the qualification stage. Testing a new cooling fluid with server and equipment makers takes laboratory capacity and partner access. One chemical maker, announcing new immersion and cold plate grades in February 2026, said its original immersion fluid had been tested in collaboration with leading original equipment and design manufacturers (Perstorp, 2026). A formulator without that partner access likely starts further back.
The demand also has limits worth stating. The IEA forecast depends on how bottlenecks in power equipment, chips and grid connections resolve, and efficiency per AI task is improving at a rate the IEA calls unprecedented in energy history (IEA, 2026). Neither point removes the demand, but both make the 2030 figure a forecast rather than a booked order.
The service side favors suppliers that know the application. Under the PG25 guidelines, inhibitor levels decline in service and periodic monitoring and analysis is recommended by the fluid supplier (Open Compute Project, 2026). Monitoring, fluid analysis, flushing chemistry and the records behind them are recurring technical work. Top-up additives must be ones the original fluid supplier approves, so a specialty supplier or distributor that answers well is likely to win the monitoring and flushing work, or the top-ups when it carries the original supplier's approved additives.
What should a specialty chemical supplier have in place to serve the AI industry?
A specialty chemical supplier serving the AI industry needs a way to answer compatibility and application questions quickly, with the evidence attached. The tool or method behind those answers should meet four conditions:
Every answer shows its basis: the company's own TDS, SDS, certificates of analysis, test results and field history.
The reasoning from the customer's materials and operating conditions to the recommendation is visible, and a person approves it before it goes out.
Each compatibility question ends as one document that the next question about the same system can draw on.
Where data is missing, such as a seal compound with no test record, the answer marks it as unconfirmed instead of guessing.
The cost of waiting is set by the qualification cycle. A buyer who receives a supported answer from one supplier and a vague one from another is likely to qualify the first, and later material changes then tend to go back to the supplier already on file. The records the OCP guidelines ask operators to keep, from batch numbers to fluid analysis reports, are the same records a supplier needs to keep a concentration and pH history per site.
A supplier can start with one product line and the compatibility questions it receives. Each answered case becomes a document, and each document makes the next answer faster, the approach behind why every specialty chemical case ends in one document.