Do AI data centers use water for cooling? What happens inside Firebird and other facilities

Large data centers powering modern artificial intelligence systems consume vast amounts of energy, which naturally raises another question: how much water do they need for cooling?

According to Grant Khachatryan, Director of YerevanN, the cooling systems of the Firebird AI factory, the Eleveight data center, and the Yerevan State University data center do not use drinking water for ongoing cooling.

He explained that the system is filled with a specialized fluid—a glycol-based coolant, part of which is distilled water. The fluid is introduced into the system once and then circulates in a closed loop for years.

"The cooling system is filled once with a special fluid—glycol—which contains distilled water. That's it. The same fluid circulates for years," Khachatryan explained.

This is not water drawn from the municipal supply. According to Khachatryan, when the data center system is filled, the fluid is delivered in specialized containers and pumped directly into the cooling loop.

Periodic checks of the fluid are necessary. Khachatryan noted that the concentration is checked approximately once every five years, since the fluid's properties can change over time and it may require adjustment .

How much fluid is in the system

To give a sense of scale, Khachatryan cited the Yerevan State University data center as an example. He estimates that its cooling system circulates about 1–2 tons of specialized fluid.

In the case of Firebird, he says the volume is roughly a hundred times larger.

This represents a significant amount of fluid, but the crucial point is that it remains within a closed system and continuously circulates, rather than being consumed like water in traditional cooling systems.

"During operation, no new water enters this facility, and there is no pipe carrying drinking water to it," Khachatryan said.

He also noted that even the fire suppression system at the Yerevan State University data center does not use ordinary water.

The reason is simple: for such facilities, water itself can pose serious risks—especially to expensive electronic equipment.

It is important to distinguish between the volume of fluid inside the system and water consumption. A large amount of fluid in a closed loop does not mean that the data center constantly consumes the same volume of drinking water.

According to Khachatryan, this is why claims that AI centers require a constant supply of drinking water for cooling do not reflect the operating principle of the systems he described.

In the case of Firebird, the precise technical specifications of the cooling system and its actual water consumption require independent confirmation from the data center operator. But based on the description provided, the cooling system itself operates on a closed loop, rather than by continuously drawing drinking water.