# Ferveret CEO on Cooling, Power Pains, and ‘Waterless’ Compute Gains — retained record

> Retained news record from news:datacenterknowledge.

Canonical: https://highsignal.app/data/records/16e4522ae4b5cb8d

[← sources](https://highsignal.app/data)


source: [news](https://highsignal.app/data/news) (news:datacenterknowledge)  published 2026-04-29 16:30 ingested 2026-04-30 08:09

[original source ↗](https://www.datacenterknowledge.com/infrastructure/ferveret-ceo-on-cooling-power-pains-and-waterless-compute-gains)

## extract

Ferveret CEO on Cooling, Power Pains, and ‘Waterless’ Compute Gains
Ferveret says fanless, waterless cooling can unlock more compute from fixed power, but deployments remain in pilot stage.
Cooling has emerged as one of the defining challenges of the AI infrastructure buildout, as operators push more power-dense systems into facilities already strained by grid limits, water availability, and rising energy costs. Into that environment steps Ferveret, a young startup positioning cooling not as overhead, but as a lever to unlock more compute from fixed power budgets.
In recent months, the company has drawn attention with benchmark results tied to its “adaptive phase cooling” technology, including claims of roughly 15% higher compute efficiency at the server level and larger gains when combined with lower facility overhead. Those figures are based on testing and modeled scenarios rather than full production deployments.
Ferveret’s approach draws on thermal engineering techniques developed for nuclear reactors, aiming to improve heat transfer efficiency while reducing reliance on traditional infrastructure such as chillers and water-intensive systems. The company describes its approach as “waterless” at the data center level.
More broadly, CEO Reza Azizian argues that improving cooling alone may not be enough, pointing to the underlying inefficiency of silicon-based computing as a longer-term challenge.
Still, Ferveret remains in the pilot phase, working with customers at the rack level and building toward broader deployment through OEM and ODM partnerships.
In a conversation with Data Center Knowledge, Ferveret’s Azizian discussed the company’s approach.
The following interview has been lightly edited for length and clarity.
Data Center Knowledge: What does Ferveret sell today?
Azizian: We invented a new category of cooling technology called adaptive phase cooling. The beauty of this cooling is that it’s rack-mounted. It’s adaptable to the current infrastructure of data centers, making it extremely easy for folks to try our solution and implement it in their facility. And by using our solution, they would be able to get a lot more compute out of the same power envelope, because the cooling is much more efficient.
DCK: What’s the primary draw for customers?
Azizian: Performance. As the data centers get more and more power-limited – right now, it’s like a nightmare to get access to more power – customers want to get more compute out of the same power envelope, and that’s basically what we offer them.
DCK: You’ve cited about a 15% improvement at the server level. What’s behind that?
Azizian: The majority of that 15% improvement at the server level comes from the fact that even in the case of direct-to-chip cooling, you still have fans at the server. Our system doesn’t have any moving parts, so by removing the fans, you get the same amount of compute with much less power. So, I would say from the 15%, about 4% to 5% comes from the fact that we run the chip cooler, so you have less leakage, and the chip runs a little bit faster. The other roughly 10% comes from removing the fans.
DCK: You’ve also talked about significantly higher gains at the data center level. How do you get there?
Azizian: We would be able to get to very low PUEs of around 1.03, 1.04. That’s a huge improvement in how much overhead you have in your data center. So, if you combine this PUE improvement with the 15% improvement at the server level, in total, you would be able to get roughly 35% more compute out of the same power envelope.
DCK: Where is this deployed today?
Azizian: We have done a lot of testing at the server level, and now we are doing pilots at the rack level with different customers. We are running pilots in data center environments, but it’s not in full production.
DCK: What does this mean for operators working within a fixed power budget?
Azizian: In a typical data center, you waste roughly about 20% to 30% of your electricity on cooling. With us, that number significantly goes down. We save a large portion of that. So, in a fixed power envelope, you would be able to feed more servers because more of your total power actually reaches the rack.
DCK: Where do you compete with direct-to-chip liquid cooling on cost?
Azizian: Our base cost is very comparable to direct-to-chip. We are not far off. And depending on the deployment, we don’t need all the chillers and cooling towers that you typically need for traditional cooling. That can be a huge amount of savings on the capex side.
DCK: Who carries operational risk, and how complex is the system to maintain?
Azizian: The fact that this is a rack-mounted solution makes maintenance easier. If there is any issue, you can take one unit out and service it, similar to what operators do today. In previous deployments of similar technology, we had cases where systems lost fluid. But even then, the temperature rises gradually. It’s not a sudden failure. You can see something is wrong and fix it.
DCK: What does deployment look like?
Azizian: At the pilot level, we’ve been doing these in a matter of weeks. The main work is repackaging the server into our solution. For scaling, we are working with OEMs and ODMs because they have the capacity to integrate and deploy at hyperscale.
DCK: What do you think people are getting wrong right now in AI deployments?
Azizian: I truly believe people think they need these gigascale data centers to get their work done. But there are a lot of smaller data centers with 5-10 MW that are left over and not used. I think there’s a path where compute could be more distributed, where it doesn’t matter what scale you run, and you can still contribute compute into a broader network.
DCK: Looking ahead, what will we realize we all got wrong?
Azizian: The way data centers are today is just not sustainable. If you compare the efficiency of the human brain to silicon, it’s completely different. The energy we need for silicon and the heat we generate are just not sustainable in the long term.
Bottom Line
Ferveret is positioning its cooling technology to reduce overhead and unlock more compute from constrained power budgets. Its approach – eliminating server fans and improving thermal efficiency – shows measurable gains at the server level, while broader data center improvements remain based on modeled outcomes. With deployments still in the pilot phase, the next test will be whether those gains hold at scale.
Even so, Azizian frames cooling as only part of a larger challenge – one that may ultimately require rethinking the efficiency limits of silicon itself.

## retained text

No document body was retained for this record — only its metadata was stored.