1972 Air Vent Sees 80% Demand Jump From AI Data Centers

Orders for a 53-year-old air vent design jumped 80% year on year, and Armstrong International does not know who is buying most of it. The Michigan manufacturer reported batches of more than 1,000 units arriving this year for a part that normally sold in lots of a few dozen, according to The Wall Street Journal, which first reported the story on the 11th.

The product is an air vent that releases air trapped in piping. Armstrong, founded in 1900 and now employing about 3,000 people at 15 sites worldwide, has supplied the part for decades to customers such as oil refineries. The units sell for several hundred dollars each.

Who is buying the vents?

Andrew Witek, head of North American sales at Armstrong, tried to trace the buyers. Distributors declined to answer, citing nondisclosure agreements. After asking around, Witek heard from an engineer that the parts were headed for data centers.

The vents cool semiconductors. AI data centers pack in high-performance chips that handle enormous computing loads; excess heat can degrade performance or cause failures. Liquid-loop cooling (a method that circulates fluid to carry heat away from chips) has grown more important, and air trapped in the piping reduces its efficiency. Armstrong's vent expels that air.

The company has added a second shift at its plant in Three Rivers, Michigan, to boost output. Kyle Higginbotham, who leads North American sales and marketing, said large orders that once would have been discussed months in advance now arrive abruptly.

Why does a 1972 design still work?

Charlie Reynolds designed the vent in 1972 at age 27, using a drafting board and a slide rule rather than a computer. He chose stainless steel for the body because it resisted corrosion better than the cast iron versions then in use. That material decision, made five decades before the AI boom, now underpins demand from the most compute-intensive facilities being built.

Reynolds, now retired, told the WSJ that he had become famous only in old age. He could not hide his surprise at the revival of his invention.

What does Armstrong not know?

Almost everything about its end customers. Data center builders guard supply chain information closely. If parts and volumes become known, rivals can estimate a facility's size, cooling method and construction schedule.

Witek told the WSJ that supply chain contacts reveal only a destination region. Armstrong then searches for data center projects in that area to guess where its vents end up.

That blindness carries costs. Without knowing the end customer, the company struggles to forecast demand, improve the product and provide maintenance support. An 80% demand increase built on anonymous orders is real revenue, but it is also fragile visibility; one long sentence captures the tension: a firm can hire a second shift for buyers it cannot name, yet it cannot plan a product roadmap for them.

There is friction closer to home as well. Reynolds lives near the Three Rivers plant, and he said nobody around the area wants a data center now. Local opposition to data center construction is growing even as the facilities drive Armstrong's order book.

What is the takeaway?

A component designed with a slide rule is now a constraint item for AI infrastructure. The numbers frame the shift: from dozens of units per batch to more than 1,000, an 80% demand rise, and a customer list the seller cannot see. Whether the boom sustains depends on data center construction that Armstrong can observe only indirectly, region by region.