xMEMS Begins Mass Production of Solid-State Cooling Chips for AI Glasses and Data Center Hardware

Putting more AI processing into a pair of glasses or a smartphone creates a problem that faster silicon alone cannot solve: the heat has to go somewhere. xMEMS Labs is moving its solid-state micro-cooling technology into mass production, targeting devices too thin for a conventional fan and small data center components that need airflow close to a hotspot.

The Santa Clara company says production has begun for its XMC-2400 and XMC-4800 µCooling chips. The first commercial products featuring XMC-2400 are expected in early 2027. That separates the manufacturing milestone from consumer availability: mass production of a component does not mean a finished phone, headset, or storage device is already on sale.

The two chips address different thermal budgets. XMC-2400 is intended for compact edge devices, including AI and XR glasses and personal SSDs. XMC-4800 provides more airflow for smartphones, enterprise SSDs, and other lower-power data center subsystems.

How a solid-state cooling chip moves air

The technology uses xMEMS’ piezoMEMS semiconductor platform to generate directed airflow without rotating blades or bearings. The company describes an all-silicon architecture that can move air through narrow microchannels and sit near processors, memory, storage controllers, and other heat-generating components.

The company’s technical overview explains the mechanism in more detail. Thin-film piezoelectric material on silicon expands and contracts at ultrasonic frequencies. Microscopic membranes and valve arrays turn that motion into air pulses, while top- or side-venting packages direct the flow toward the target.

“Solid-state” therefore describes the absence of a traditional spinning fan assembly, not an absence of physical motion. The airflow comes from tiny vibrating structures. That distinction matters when comparing the device with passive materials, which spread heat but do not actively pump air.

What the production specifications say

The supplied production release lists the following specifications, which also match the current product page. Airflow, back-pressure, and power figures are vendor specifications; the maximum values should not be read as a guarantee that every installed system will achieve them simultaneously.

Back pressure is relevant because air inside a compact product may have to travel through a restrictive path. An airflow number by itself does not explain how much air reaches the hotspot once the chip is integrated with channels, vents, and surrounding hardware. The complete cooling result depends on that system design.

xMEMS reports that customers have achieved chip- and system-level temperature reductions of 10–25°C in testing across glasses, smartphones, and SSDs. The announcement does not identify the devices or provide a common test methodology, so those figures are examples of reported results rather than a universal expected temperature drop.

Why sustained performance matters for edge AI

Short benchmark bursts and sustained workloads can produce different experiences. A device may briefly run a demanding model at high speed, then reduce processor clocks as heat accumulates. For applications such as continuous vision processing or real-time assistance, maintaining performance over time can matter more than a peak score.

In its white paper on edge AI thermal management, xMEMS argues that smartphones, SSDs, and AI glasses face a common constraint: increasing compute density within enclosures that leave little space for cooling. It positions active micro-cooling as a way to add local convection alongside existing thermal structures.

Active cooling also consumes power. The engineering question is whether the additional thermal headroom provides enough benefit to justify its electrical demand, integration space, and cost. For glasses, the design also has to account for how heat reaches surfaces near the wearer; moving air near a processor does not by itself establish that the finished product is comfortable.

Targeted cooling inside data center hardware

The data center application is more specific than cooling an entire AI rack. xMEMS identifies SSDs, laser modules, memory, retimers, and other subsystems as targets for localized airflow. Its production release describes these as components that can benefit even as liquid cooling becomes concentrated around high-power CPUs and GPUs.

That makes the technology a potential complement to larger thermal systems. A small air pump near a storage controller addresses a different problem from transporting heat away from a large accelerator. The success of either approach depends on a path that ultimately carries heat out of the component and into the surrounding cooling system.

From demonstrations to manufacturable products

xMEMS first introduced XMC-2400 in 2024. The production announcement follows a period of sampling and demonstrations, and sits alongside its broader portfolio, including XMC-1200 for smaller smart-glasses components. Our earlier coverage of xMEMS’ miniature cooling technology examined that separate product launch.

Mike Housholder, vice president and general manager of xMEMS’ Thermal Management Business Unit, describes mass production as a step toward supporting customer demand at scale. The company also presents its silicon architecture as a possible future route to thermal solutions co-packaged with edge AI processors. That is a development direction, rather than a product capability established by this announcement.

The next useful evidence will come from finished systems: sustained performance, surface temperatures, battery impact, reliability, and manufacturing yield. Entering production moves micro-cooling beyond a component demonstration. Whether it changes the practical limits of AI wearables and compact storage will become clearer as customers bring integrated products to market.