Europe is enduring an unprecedented heat wave this week, and somewhere amid the news of air conditioners overloading grids and headlines warning of another blistering June, there is a less-publicized problem occurring at data centers that are fueling the AI revolution. The processors that power AI tools like chatbots and image generators are overheating along with everything else, and the industry is just starting to realize the implications.
The statistics speak for themselves. According to Zurich, one of the companies providing insurance for U.S. data center construction projects, severe weather events have grown into the leading cause of losses within its portfolio of builders’ risk insurance in the last three years, accounting for nearly a third of all claims. Another study conducted by the climate risks firm First Street revealed that 79% of global data center capacity is exposed to higher risks of acute climate events such as floods, strong winds, and wildfires.
Also read: OpenAI Codex Macro Pad: What is it, how does it work?
The problem for us comes from there. Insurers observe that a considerable part of the growing capacity of newly-constructed data centers, around 64% worldwide of what’s being currently built, is now located in so-called “frontier markets”, less developed countries with cheap land and a smaller history of weather data. India’s boom with data centers is a typical example of such expansion. Cities like Pune, Hyderabad, Chennai, and other new centers that will appear in the vicinity of Mumbai are precisely those expansion zones that insurers are talking about, just without years of hardening of infrastructure, unlike well-established hubs of Northern Virginia.
Also read: From impersonation to scams: What could go wrong with WhatsApp usernames
India doesn’t require any imported heat wave to prove its point here. We have a good experience of summers, with temperature frequently reaching 45°C in a large part of our country combined with weak electricity grids that can’t withstand the load of air conditioners’ usage every year. Cooling requires approximately 40% of data center’s power under regular conditions and rises dramatically during heat waves when grids can provide the least surplus of power. Adding an AI data center, that would consume energy similar to small-town requirements, to that grid in Delhi or Nagpur summertime becomes rather risky.
The good news is that the ability to adapt through technology already exists. Nvidia says its latest generation of AI servers can handle liquid cooling at coolant temperatures as high as 45°C, which is a significant improvement from previous limitations on that front, and will directly save on cooling-related energy costs. Hyperscalers like Microsoft are now openly discussing their efforts in climate-aware location selection and redundancy. This is not rocket science; it is simply design.
For the AI infrastructure boom taking place in India, what needs to be considered is if the builders of these facilities are designing for Indian summers right from the start, or if they will learn their lesson only once a heatwave causes an outage for them. With the huge amount of money looking to be invested into these facilities in India at the moment, getting this wrong could turn out very expensive indeed.
Also read: Top 10 most powerful supercomputers in the world: What’s their configuration?