
HVAC
HVAC is the umbrella term for the technology that heats, ventilates, and cools buildings. In data centers, it determines whether the computers can get rid of their waste heat – and consumes a substantial share of the electricity itself.
HVAC stands for Heating, Ventilation and Air Conditioning. It refers to all the technology in a building that ensures pleasant and stable air conditions. This includes boilers and heat pumps, ventilation ducts and fans, chillers and cooling units. In everyday use, it’s often simply called building services or climate technology. In the tech and finance world, the term today comes up mainly in connection with data centers. There, cooling is no longer a matter of comfort but the basic precondition for the facility to run at all.
Why data centers depend on cooling
Every chip converts practically its entire power draw into heat. A single modern AI accelerator can draw up to 1000 watts, about as much as a space heater. A server rack contains dozens of them. Without active cooling, the temperature would rise within minutes to the point where the chips throttle themselves or shut down.
That’s why HVAC is a first-order cost factor in data centers. The metric for this is called PUE, or Power Usage Effectiveness. It indicates how much electricity the entire building consumes, divided by the electricity that actually reaches the servers. A value of 1.5 means: for every kilowatt-hour of computing power, an additional half goes toward cooling, lighting, and other technology. Well-built facilities today achieve values around 1.1.
For investors, this matters because cooling has long since become a deciding factor in site selection. Operators build where it’s cool, where water is available, and where electricity stays cheap. Companies from the climate technology sector are therefore considered indirect beneficiaries of the AI boom on the stock market.
From recirculated air to liquid at the chip
The classic approach is air cooling. Air conditioning units blow cold air into a raised floor, from where it rises through openings in front of the server racks. The servers' fans draw it through the equipment, and hot air comes out the back. To keep the two air streams from mixing, cold and hot aisles are physically separated from one another.
This method is reaching its limits with AI hardware. Air absorbs heat poorly, while water can transport roughly three thousand times as much heat per volume. That’s why liquid cooling is gaining ground. Here, cold plates sit directly on the chips, with water or a special fluid pumped through them. An even more radical variant submerges entire servers in non-conductive oil.
The whole system becomes more efficient through free cooling. If the outside air is already cold enough, the energy-hungry chiller stays off, and a heat exchanger takes over instead. In Northern Europe, this works on many days of the year. Incidentally, a common misconception is that data centers need to be kept icy cold. In fact, many facilities run at 24 to 27 degrees Celsius, because higher temperatures lower cooling costs and modern hardware can handle it.
HVAC in quarterly earnings and at home
You mostly encounter the term in the news through numbers. When a corporation announces it’s building a data center for several billion, a noticeable share of that goes toward cooling and ventilation technology. Debates about AI’s water consumption also revolve around HVAC, since some facilities cool via evaporation and consume drinking water in the process.
In everyday life, HVAC is present in every office building, in trains, hospitals, and supermarkets. At home, a heat pump with a ventilation system performs the same task on a small scale. Such systems are increasingly controlled by software that evaluates weather data and occupancy and regulates the system predictively. This is exactly where building technology meets AI: models forecast heating demand and thereby save energy without anyone touching the thermostat.