- xMEMS is pitching ultrasonic micro air cooling for data centers
- Its niche play works alongside liquid cooling, not against it
- Thermal management market growth is strong, but xMEMS' long-term fit is uncertain
It’s no secret that liquid cooling is emerging as the de facto standard for high performance AI servers – just look at Nvidia’s design for its Vera Rubin racks. But air cooling isn’t dead yet, and xMEMS thinks it has a novel application for forced air inside of servers.
Founded in 2018, xMEMS has built a teeny tiny, silicon-based digital air pulse generator. In a nutshell, it uses ultrasonic frequencies outside the range of human hearing to move air in small spaces and reduce temperatures by 10 to 20 degrees Celsius. Its sweet spot is cooling components that run between 10 and 30 watts.
Mike Housholder, VP and GM of xMEMS’ Thermal Management Business Unit, told Fierce that while its tech was originally designed for the consumer electronics market, it has been “surprised” by the interest level and uptake by companies in the data center market.
It’s true that 10-30 watts is a fraction of the 1,000-plus watts modern GPUs use. But xMEMS isn’t trying to cool those chips – instead it’s focused on cooling components like SSDs and optical transceivers that are outside the reach of the liquid cooling systems being used on GPUs.
Housholder said that as customers are moving to liquid cooling, they’re removing system fans in their servers. While the GPU and CPU are happy about this since they’re still served by liquid-cooled cold plates, “the lower power stuff is now sticking out like a sore thumb as hotspots.”
For instance, on the transceiver side, it’s not focused on DSPs but rather the section of the module outside the server where optical cables plug in. Housholder said there’s about 6 watts in that little transition spot, which is critical to the efficacy and quality of the server connection.
He added xMEMS is still trying to get its arms around the data center opportunity. But while it expects this market to be just a fraction of its addressable opportunity in consumer electronics, Housholder said xMEMS already has an in with big SSD suppliers, has secured a key optical transceiver vendor as a customer and has started conversations with two hyperscalers who are interested in the tech.
Crystal ball
Zooming out, it’s important to note that the liquid cooling market – like the AI technology it serves – is rapidly evolving. Dell’Oro Group Research Director Alex Cordovil noted that “liquid cooling is expanding well beyond the GPU/CPU—memory, PCB, storage, optical components are all increasingly in scope.”
Once liquid cooling infrastructure is present at the server level, there's a natural design evolution toward larger, more complex cold plate geometries that can touch more elements within the server chassis,” he told Fierce.
While Cordovil said he wasn’t familiar with xMEMS’ offering specifically, he said this trend shapes how a company like it might fit into the ecosystem. His conclusion? Consumer electronics and edge deployments may end up being the sweet spot for the kind of solution xMEMS is offering.
“For physical AI and wearables, micro air cooling strikes me as genuinely compelling—hot spots in constrained spaces are a real challenge, and a silicon-based micro cooling chip with low power draw is an elegant approach,” he said. Cordovil added there’s also a “credible case” for using such technology for air-cooled AI racks used for edge inferencing deployments but saw less headroom in fully liquid-cooled racks.
Cooling forecast
In the wake of Q4 2025 earnings reports and 2026 capex guidance from hyperscalers, Dell’Oro Group earlier this month boosted its forecast for data physical infrastructure spending. It now expects the market – which includes thermal management – to grow at a mid-teens CAGR to hit more than $80 billion by 2030.
The thermal management category specifically is expected to grow at a 20% CAGR, with spending on direct liquid cooling tipped to surpass $8 billion by 2030.