Here is a fact about a modern data center that sounds like a mistake but isn't: a large share of the electricity it consumes never touches a server. It goes to keeping things cold. Chillers, pumps, fans, cooling towers - an entire second machine running alongside the computers, quietly turning power into removed heat. The computers do the famous work. The cooling does the expensive work. For a long time nobody optimized the second machine very hard, because it was somebody else's department and it mostly worked.
etalytics is a company built on the premise that "it mostly works" is leaving an enormous amount of money and carbon on the floor. It makes software - a platform called etaONE - that watches an industrial energy system in real time, builds a digital twin of it, forecasts what it is about to do, and then, if you let it, takes over and steers the whole thing. The pitch is not "use less." The pitch is "use exactly what you need, when you need it, and let a model figure out the rest." Customers who have handed etaONE the keys to their cooling report electrical cooling power dropping by 30 to 50 percent, which is the kind of number that makes a plant manager ask you to prove it several times before they believe you.
They did ask, repeatedly. Co-founder Thomas Weber summarizes the early sales cycle as "Nice idea. Now prove it. Ten times." This is the correct instinct from anyone running a facility that cannot go down, and it shapes how etalytics sells: start with a small deployment, show the savings on real equipment, then scale. It is a slower go-to-market than a viral app, but it is the only one that works when the customer is a pharmaceutical cleanroom or a car plant.
The company was founded in 2020 by three PhDs - Niklas Panten, Thomas Weber, and Björn König - who came out of TU Darmstadt's "ETA Factory," an actual working research factory built to study energy-efficient production. This is a useful origin detail, because it explains why etalytics does not look like most AI startups. Its founders spent years in a building whose whole purpose was measuring where energy goes in industrial processes. When they say efficiency should be treated as a property of the system rather than an afterthought, they are describing the environment they were literally trained in.
The technically distinctive choice is that etalytics does not just throw a neural network at the problem and hope. It uses a hybrid approach: machine learning combined with physics-based model predictive control. The physics matters because industrial customers need guarantees - a chiller has real limits, a cleanroom has real tolerances, and a purely statistical model that has never seen an edge case is exactly the thing that gets you a compliance incident. Wrapping the AI in a physical model of the plant is what lets etalytics hand over control instead of just producing a dashboard. A dashboard tells you the problem. etaONE makes the decision.
There is a nice irony in who is funding this. In October 2025 etalytics doubled its Series A to €16 million, and the new lead was M12, Microsoft's venture fund. Microsoft is, of course, in the business of building some of the most power-hungry data centers on Earth. So one of the world's largest sources of soaring cooling demand just invested in a company that makes cooling demand smaller. That is either a hedge or a strategy, and it doesn't much matter which - both point at the same future, in which AI-driven optimization becomes standard plumbing for large energy systems rather than a niche upgrade.
The customer list reads like a who's-who of energy-intensive operations: Volkswagen and Audi in automotive, Merck and Grünenthal in pharma and chemicals, and a wall of data-center operators - Equinix, NTT, Digital Realty, Telehouse. These are not companies that adopt experimental software for fun. They adopt it when the payback is obvious and the downside is contained, which is a reasonable summary of etalytics' entire value proposition: the savings are large, the equipment is the equipment you already own, and the AI is bolted to a physical model so it behaves.
The company remains relatively small - around 73 people - with roughly €2.3 million in revenue and a stated goal of growing recurring revenue tenfold over two years. Whether it hits that number is the usual startup question. But the underlying bet is unusually easy to state. Everyone is worried about how much energy AI consumes. Far fewer people are building the tools to make the rest of the industrial world consume less. etalytics is quietly doing the second thing, and its investors and customers happen to be the exact people driving the first.
