xMEMS Labs, a venture-backed silicon startup founded in 2018 and known for speakers for earbuds and headphones, has started mass production of two chips that cool AI hardware, SiliconANGLE's Duncan Riley reported on October 13. The company says no one has commercially sold solid-state active cooling at this scale before.
How it works
The chips have no moving parts. Nothing spins, and each device is a single piece of silicon. The technology comes from xMEMS's speaker business, where an electrical signal flexes a thin silicon membrane to produce sound; in the cooling chips a similar membrane moves air. The company says the chips build high pressure, so air can be routed through narrow channels to a processor or memory chip, and that they could eventually be packaged with the AI processors they cool.
The products
The XMC-2400 measures 7.42 by 9.48 by 1.13 millimeters, moves up to 28 cubic centimeters of air per second against up to 1,300 pascals of back pressure, and draws about 150 milliwatts. It targets personal SSDs and AI or extended reality glasses. The XMC-4800 moves up to 48 cubic centimeters per second and targets thermally demanding AI smartphones, enterprise SSDs and lower-power data center subsystems. A smaller XMC-1200, introduced in July, is designed to cool light engines and processors in smart glasses.
The claims
Customers testing the chips reported temperature reductions of 10 to 25 degrees Celsius at the chip and system level in glasses, smartphones and SSDs. These are company-reported results. In data centers, the targets are parts outside the liquid cooling that surrounds high-power processors, such as SSDs and laser modules. Mike Housholder, vice president and general manager of the thermal management unit, said conventional fans “simply cannot fit” in AI glasses and smartphones.
Funding and timeline
In October 2025, xMEMS raised $21 million in a Series D round led by Boardman Bay Capital Management to fund its move into mass production. The first commercial products based on the XMC-2400 are expected in early 2027, with live demonstrations planned for CES in Las Vegas in January.
Source: SiliconANGLE. This is an AI Business Watch summary of the original report.