Using Arbino
Reuse the heat that AI releases
By Arbino ·
Your laptop gets warm just from watching Netflix. Now imagine thousands of computers giving off heat in one room. That's a lot of heat to waste while people are trying to keep their heating bills down.
An October 2026 survey by Hart van Nederland found that roughly 88% of respondents hadn't turned on their heating since summer. For 25%, cost was part of the reason for keeping it off.
Most of the electricity used to run a data centre ends up as heat. When that heat goes to waste, we're not using the energy as efficiently as we could. With the right equipment and a connection to a heating network, that warmth can be reused again. Infomaniak's D4 project in Geneva shows how computing and household heating can work together.
That's one reason we chose Infomaniak for Arbino's hosting and AI processing. Its AI infrastructure uses local renewable electricity and recovers heat for household heating. If you want everyday AI help with more thought behind the infrastructure, Arbino is our pick.
How do you get heat from a server into a home?
Think of a district heating network as a shared heating system for a neighbourhood. Pipes carry heat to connected buildings, where it can warm rooms and provide hot water. A data centre can feed heat into that network.
At Infomaniak's Geneva D4 site, the journey has three main steps:
- Collect the warmth. Air moving through the servers picks up heat. A heat exchanger transfers that warmth into a water circuit.
- Raise the temperature. Heat pumps use electricity to bring the heat to a temperature the district heating network can use.
- Send it where it's needed. A separate water circuit carries the heat into Geneva's heating network.
The circuits exchange heat while keeping their water separate. The same process helps cool the servers, ready for more computing. You can explore the equipment and diagrams in the D4 technical guide.
For a data centre connected this way, computing produces both a digital result and heat the neighbourhood can use.
The heating network matters as much as the servers
Collecting heat is only part of the setup. There also needs to be somewhere to send it, with demand for that heat.
During operational testing in 2024, Infomaniak recovered 65% of the electricity consumed at its newest Geneva data centre as heat, as documented in its impact report. The district heating network was still ramping up, which limited how continuously the site could export heat. The design aims for full heat recovery.
That makes the location and connection part of the engineering. You need the servers, the recovery equipment and a heating network ready to accept the output.
A new project in Bulle puts heat reuse in the plans
On 13 November 2026, Infomaniak and Gruyère Energie announced a proposed data centre in Bulle, Switzerland, in a joint press release. Planned for 2028, it would supply recovered server heat to Gruyère Energie's existing district heating network.
The project has entered public consultation, with construction planned for the first half of 2027, subject to the necessary approvals. The partners aim to recover all the heat produced by the servers. Their planned responsibilities include Infomaniak building and operating the data centre, and Gruyère Energie providing the heat-recovery installations and renewable electricity, subject to final contracts.
Choose your AI with the infrastructure in mind
You probably choose an AI chat because it helps you understand something, tidy up an email or get past a blank page. You can also choose what kind of infrastructure sits behind that help.
With Arbino, we chose a provider that buys local renewable electricity for its AI infrastructure and puts server heat to use. Renewable power addresses the electricity used to operate the service; heat recovery makes that energy useful beyond the server room.
Try Arbino with a question you've been meaning to work through. For a broader look at price, privacy and environmental choices, our previous published post compares green ChatGPT alternatives.