ChatGPT May Soon Warm Your Home
ChatGPT May Soon Warm Your Home
As private and industrial AI applications continue to grow in popularity and adoption, so too do cloud-hosted AI computing power demands. With future power consumption from AI forecast to increase from 1GW in 2023 to 7GW in 2026 in the US alone, this raises the question: can ChatGPT heat our homes?
AI and generative AI solutions bring tremendous power consumption demands to data centres. Globally, ChatGPT consumes as much as 260MWh per day – each user query is equivalent to running a 5W LED lightbulb for 80 minutes. This consumption is only set to increase with new public and private AI offerings; with the increase in electricity required, and the subsequent increase in heat generated in data centres, it is more important than ever to capture this heat and prevent it being wasted.
The concept of reusing data centre waste heat is not a new one. However, the grade of heat has presented challenges to district heating schemes. With data centre temperatures within the 25-40°C (77-104°F) range, this heat is well outside of the traditional hot water heating temperatures of 65-80°C (149-176°F). But, with advancements in heat pump technology, it is now possible to increase the quality of this waste heat and connect data centres with local district heating networks of homes and businesses.
The technology has been successfully deployed by hyperscale and colocation data centre providers worldwide and many schemes are in development. Facebook, for example, has recycled waste heat from its Danish Odense data centre since 2019, and 100,000MWh of energy is recovered per year (enough to warm around 6,900 homes). Equinix recently announced the development of a new district heating network in Frankfurt, Germany that will heat 1,000 homes from 2025; the operator’s existing FR4, FR6 and FR8 sites will be connected to the network. Equinix has also connected its Parisian PA10 facility to a heating network and this facility is set to provide heating to the 2024 Olympics Water Centre. On a smaller scale, a “washing-machine-sized” data centre has been utilized in Devon, England to provide heating to a leisure centre swimming pool for 60% of the year, saving the operators thousands of pounds on heating bills.
Another driver of waste heat recovery schemes has been the continued pressure on global energy following Russia’s invasion of Ukraine. With volatile energy prices and Europe desperate to wean itself off Russia’s gas, green heating from data centres becomes even more attractive. Legislators see this source of green energy as essential to meet net zero targets; the German government’s proposed Energieeffizienzgesetz mandates that newly built data centres from 2025 must reuse 10% of their excess heat – and from 2028 new facilities must reuse 20%. The law also stipulates that both existing and new data centres in the country must use renewable energy for 50% of their consumption by 2024, and 100% by 2027.
As data centre electricity consumption grows for both AI and traditional cloud functions, it is imperative that as much heat as possible be captured rather than released to atmosphere. Then ChatGPT will be able to answer your complex queries while it heats your home.
For further insight on energy management technologies, including waste heat reclaim heat pumps, please see Verdantix Tech Roadmap: Energy Management Technologies.For further insight on the environmental impact of AI models, please see Verdantix: The Future Of AI Models And Their Environmental Impact.