How TenHats cut back on data center energy usage using a startup’s approach
The debate over data center growth in Knoxville is really a debate about energy, and who bears the cost of powering the AI boom. TenHats and FLUIX AI think the answer might already be hiding inside the buildings we have.
Data centers have a complicated reputation in East Tennessee and across the country. They consume big amounts of electricity, and as AI workloads surge, that appetite is only growing. For communities weighing the benefits of tech investment against the strain on local power grids, it’s a real tension.
Knoxville is feeling it directly. Earlier this month, Mayor Indya Kincannon called for a one-year pause on data center construction to evaluate environmental impacts and establish rules around land, power, and water consumption.
One data center already operating in the city, TenHats, is taking a different approach. Through TVA EnergyRight, TenHats partnered with FLUIX AI, a San Francisco startup with Tennessee ties, to make its existing facility more efficient. Early results suggest the partnership is working.
“Data centers earn their place in a community by using far less of those resources, getting more out of the existing infrastructure instead of demanding more of the grid,” said FLUIX founder Abhishek “Abhi” Sastri. “That’s what we did at TenHats.”
Making the buildings we have work harder
The core of FLUIX’s approach is a platform called A.I.M.I., which operates like an autopilot system for data center cooling.
Cooling is one of the biggest energy expenses in any data center, and it’s typically managed through old-fashioned systems that react after temperatures change, scrambling to catch up.
Instead of reacting to temperature shifts after they happen, A.I.M.I. learns a facility’s thermal patterns and anticipates changes before they occur. At TenHats, that meant cutting the lag between an IT load change and a cooling response from roughly 15 minutes to five.
FLUIX AI’s system plugged into TenHats’ existing infrastructure and, within two weeks, was ready to take the wheel, operating within guardrails that the facility’s own operators set.
The results, and what they mean
Over a multi-month deployment at TenHats, the numbers told a compelling story. The facility saw a 6.6% improvement in Power Usage Effectiveness (PUE), a standard industry metric that measures how much of a building’s total power actually reaches the computers versus how much is lost to overhead like cooling. And, critically, there were zero service disruptions throughout the entire run.
Projected across a fully loaded 2 MW facility, Sastri predicted these efficiency improvements could free up as much as 130 kilowatts of power capacity, enough to support additional computing rather than just keeping servers cool.
Sastri projects that at typical colocation pricing of around $300 per kilowatt per month, that level of potential headroom could represent roughly $470,000 to $500,000 per year in additional capacity value, though FLUIX notes that figure is a full-site projection based on observed efficiency gains and not yet directly validated revenue.
“Every watt wasted on avoidable cooling is a watt that cannot be used for computing,” Sastri said. “FLUIX is building the AI control layer that turns efficiency into capacity.”
TenHats may have started as a test site, but the facility signed on to keep A.I.M.I. in continuous autonomous control of its cooling systems following the deployment.
A regional angle worth watching
The project received support from TVA EnergyRight as part of a broader effort to evaluate how AI-driven tools can help data centers reduce waste and support more sustainable, flexible growth across the Tennessee Valley. Solutions like FLUIX’s suggest the answer may not always require new infrastructure, but rather, rethinking how existing infrastructure is managed.
“This is only the tip of the iceberg in terms of the potential AI brings to data center efficiency, compliance, and performance,” said Wade Orloski, data center manager at TenHats.
FLUIX’s TenHats deployment is an early example of what it looks like when AI starts optimizing the systems that power AI itself.
Like what you've read?
Forward to a friend!