Everyone building data centers today is fighting two well-known constraints: power and water. The third one gets far less attention, and it's increasingly the reason projects stall, permits get denied, and operators end up in litigation with the communities around them: noise.
That's the problem Zenbul is built to solve, and it's why Banyan is co-leading the company's pre-seed round as first money in.
Search "noise from data center" and you'll find the same pattern repeating across the country: lawsuits, community opposition, and stalled or blocked projects. At the same time, by some estimates the U.S. needs to expand its data center footprint by roughly tenfold over the next five years or so — and much of that build-out has to happen in or near the places people actually live.
Two forces are colliding. We need vastly more compute, and we need to put a lot of it close to communities that don't want to hear it.
There's a structural reason this gets harder, not easier. As workloads shift from training to inference, latency starts to matter more than raw scale, which pushes data centers closer to the end user. A recent example: Brooklyn-based colocation provider DataVerge and AI software company Mathpix deployed GPU capacity in New York specifically to cut latency for API workloads serving local customers, rather than relying entirely on distant cloud regions.
The closer compute moves to people, the more a data center's relationship with its neighbors — and its noise profile — becomes a gating factor for getting built at all.
Zenbul uses AI to design custom, per-site acoustic materials for data centers and to optimize how those materials are laid out — meaningfully lowering the cost and time of properly protecting a site from sound. The company focuses on the HVAC system, which is where the majority of a data center's noise comes from, and can work on both retrofits of existing facilities and net-new builds.
The incumbent approach is expensive, slow, and not AI-native. It's the reason a lot of mitigation today still comes down to planting trees and pouring concrete — solutions that simply don't scale in residential areas.
The hard part isn't any single material. It's the model that ties the whole problem together: designing the physical acoustic solution, fitting it precisely to a given site, and delivering it with the lowest-cost materials. That end-to-end design-to-delivery system is Zenbul's core IP, and it's genuinely difficult to replicate.
We invest in teams first, and Zenbul's is exceptional for this exact market.
Duncan, the CEO, was a founding engineer at a unicorn startup and brings a robotics background from the University of Michigan. Dr. Shai is a University of Michigan professor who runs the Biologically Inspired Robotics and Dynamic Systems Lab, is a recognized leader in data-driven mathematical algorithms, and has more than 2,500 citations on Google Scholar. This is a team built to turn a hard physical-design problem into software.
Zenbul fits a thesis we've been building at Banyan: AI applied to specialized, vertical, physical-design problems is proving to be one of the most durable uses of the technology. Our Fund I company Normal Computing is doing this in chip design and environmental testing; Zenbul is doing it for data center acoustics. Same shape, different domain.
And the wedge doesn't stop at sound. Once you've earned trust solving acoustic design for a data center operator, the same approach extends naturally into adjacent problems like thermal management — helping servers run cooler.
At the pre-seed stage, we're getting a differentiated product, a defensible model, and a team uniquely suited to an emerging market, at a moment when the noise problem is moving from a nuisance to a build-or-don't-build constraint for the entire industry.
That's the kind of risk-adjusted entry point Banyan is built to find. We're proud to co-lead Zenbul's pre-seed, and I'm joining the advisory board to help build the commercial pipeline from day one.
