Microsoft backs German AI firm etalytics as datacenter power costs bite

Microsoft’s venture arm doubles a German startup’s Series A to €16 million—software that cuts datacenter cooling costs by up to 40% as North American power prices climb and AI workloads multiply.

M12, Microsoft’s venture investment arm, led an €8 million extension to etalytics’ Series A round, bringing total capital raised to €16 million. The Darmstadt-based company deploys AI to optimize cooling systems in datacenters, pharmaceutical plants, and automotive facilities in real time—infrastructure that typically consumes 40–50% of a facility’s non-compute energy budget.

The investment centers on North American expansion. Etalytics opens a Bay Area office in 2025, targeting the world’s largest datacenter market as operators face rising electricity costs and mounting sustainability pressure. Pilot projects with major U.S. datacenter operators are already in motion. That’s not a lab demo. It runs in production at Equinix, Digital Realty, Audi, Volkswagen, Stellantis, Merck, and NTT Data.

What’s actually new

This isn’t just more capital for European climate tech. The new money comes with a push into North America, the world’s largest datacenter market and the epicenter of AI demand growth. Etalytics will stand up a San Francisco team this year to support pilots already underway with major U.S. operators and to scale deployments as those tests convert. It’s a go-to-market ramp. And it’s timed for grid reality.

The original €8 million round was led by Alstin Capital, joined by ebm-papst and the Technologiefonds Hessen (BMH). Microsoft’s participation doubles that total and signals a practical focus: efficiency software that adds effective capacity when new megawatts are hard to find. That’s the crux.

The capacity squeeze, explained

AI workloads swell, but utility interconnects don’t. In Northern Virginia, Silicon Valley, Phoenix, and Dallas, power delays stretch into years. Operators need to free headroom from the infrastructure they already have. Every kilowatt not burned by chillers, towers, and pumps can be reassigned to GPUs. It’s compute by subtraction. It also softens Scope 2 emissions and helps with ESG audits and buyer scrutiny.

Cooling is the biggest lever on the non-compute side. If you can reduce cooling energy by, say, 40%, you’re taking a meaningful chunk out of total site load without touching servers. That’s budget and carbon in one move. It’s also speed. Software deploys faster than concrete.

Software meets thermodynamics

Why believe the savings? Because the control problem is tailor-made for hybrid AI. Pure neural nets struggle with physical causality in safety-critical systems. Etalytics blends physics-based models (digital twins of the plant) with machine learning for prediction and optimization. The system continuously adjusts valves, pump speeds, and compressor loads against goals and constraints. No hardware swap required.

The platform—etaONE—emphasizes autonomous control over dashboard advice. That distinction matters in plants that run 24/7 and in datacenters where temperature excursions carry real risk. Explainability also counts. A physics-grounded controller can show why it moved a setpoint, which helps with audits and standards like ISO 50001. It’s designed to be deterministic where it must be, adaptive where it can be. Safety first.

From lab work to field results

👉 Etalytics spun out of TU Darmstadt’s ETA program, where co-founders Dr. Niklas Panten and Dr. Thomas Weber spent years testing control approaches on real energy systems.

👉 The company now has roughly 70 employees across a dozen countries and plans to grow past 120 within two years as it scales pilots into portfolio-wide rollouts. That headcount is aimed at deployment, not just R&D.

Cross-industry breadth is a feature, not drift.

👉 Automotive paint shops, pharma cleanrooms, and hyperscale datacenters share the same mechanical building blocks: compressors, evaporators, condensers, towers, and variable speed drives. The control strategies travel, with parameters tuned for each sector’s risk tolerance and uptime requirements. Reuse is the economic engine here. It keeps integration costs sane.

Why Microsoft cares

Cloud economics and enterprise optics converge. Azure faces the same thermal and energy dynamics as its peers; its customers increasingly ask for efficiency and emissions transparency tied to their AI usage. A portfolio company that can shave megawatts from ancillary loads offers practical leverage across both dimensions. It’s also a partner story. Microsoft brings distribution, compliance muscle, and enterprise relationships that can compress sales cycles in conservative industries. That reach matters.

There’s a second hedge: supply chains for new power and new cooling hardware remain constrained. Software that wrings more out of installed plants helps bridge the gap until substations, on-site generation, or heat-recovery projects arrive. This is stop-gap plus step-change. It stacks with future upgrades instead of competing with them.

The deterministic AI question

Call it the other AI. Not generative chat, but agentic control with guardrails. Datacenter cooling is safety-critical. It requires predictable responses to known states and bounded behavior in unknown ones. That’s where hybrid controllers earn their keep: they keep temperatures within strict limits while trimming kilowatts at the edges. Fail-safes are built-in. Night shifts still sleep.

This is where the market is going: more autonomy across thermal, electrical, and mechanical layers, with humans supervising fleets and approving policy rather than micromanaging plants. Etalytics is carving out the thermal lane first. It’s the biggest, fastest win.

Why this matters

• Datacenter growth is now constrained by power, not land; cutting cooling load effectively adds compute without new megawatts.
• Software-first optimization delivers immediate savings and emissions reductions while bigger grid and hardware projects catch up.