 |
| -Image by Copilot |
If cables are the arteries of the Information Economy upon which we are ever more dependent, then Data Centres are the brains and nerve centres behind it. A data centre is a physical building or group of buildings which house large numbers of computers, servers, storage systems and networking equipment which store, process and distribute digital information. They are the superstructure of the internet and cloud computing and handle everything from websites, email, streaming services, banking transactions and artificial intelligence (AI), to government records and company data.
They also face much the same challenges as other infrastructure in a changing climate -an increasing number of adverse weather events – excessive heat, flooding, more severe storms and wildfires and so on, for which the remedies remain much the same. Because we have covered this quite a bit in previous posts, this one will be not so much about how climate change will affect data centres, but rather, how to prevent data centres accelerating it, but first, a quick run down on the main ways to climate -proof them.
Reducing Weather Related Disruptions
Diversifying power supplies and ensuring there is adequate ‘redundancy’ – that is, additional capacity which can be called upon in the event of failure anywhere in the system, should be a high priority.
Regular testing and realignment with current and projected future conditions will help to prevent costly outages. Rumour has it that during Hurricane Sandy in New York (2012) in which flood waters rose 13 feet above sea level, two data centres had prepared for outages by placing their emergency generators on the 25th floor of an office tower. Unfortunately they had forgotten that the fuel and pumps for them were located in the now flooded basement. [Keeping fuel in basements was a legal requirement at the time. A similarly afflicted hospital had to resort to bucket brigades]. In another instance, emergency equipment on raised stands was blown away by the force of the storm surge. Battery backup and co-location with other services, would under normal circumstances provide additional safeguards against failure.
To protect data centres against flooding, sensitive areas and essential equipment such as control rooms and cabling should be flood -proofed with water tight doors and the premises themselves should be protected with good drainage and deep channels that direct water away from buildings. Given that data centres are expected to last 10 -20 years, flood maps should incorporate projections for the lifespan of the development and be updated regularly. Coastal areas are likely to be more vulnerable than those inland.
Smaller, dispersed data centres offer greater protection in a rapidly changing climate than one large unit which could easily be disabled by fire, storm or a power outage. Wildfires have certainly increased in recent years and a data centre does not have to be on fire to be disabled by smoke or debris, which can travel many miles ahead of a fire front. Providing one’s own fire protection may be in order too, since emergency personnel may be otherwise occupied saving lives and homes. Smaller centres may also help to reduce the growing conflict between data centre operators and the communities in which they operate.
For more on this topic see the excellent White Paper on Data Centres by Ramboll from which I have taken much of this information.
The Pushback
Mounting global opposition to data centres comes primarily from concerns about their impact on already stressed power and water supplies, but also their space requirements and general and growing contribution to global warming.
In Ireland, data centres consume 21% of national electricity — the same as all urban households combined — prompting Eirgrid to impose a moratorium on new approvals in the Dublin area. In the Netherlands, local opposition citing environmental damage, successfully derailed Meta's planned Zeewolde campus, which would have been the largest in the country.
In Chile, a court suspended a Google data centre after locals discovered it would extract more than 7 billion litres of water annually. Across Europe, at least 25 data centre projects totalling 4.7 gigawatts of demand were cancelled in 2025 due to opposition, and in Scotland, planned data centres would consume 2-3 gigawatts of power — against a national peak winter demand of just 4 gigawatts. Water use is consistently the main concern, followed by rising electricity costs for ordinary consumers and the threat to national climate targets. A summary of the main arguments follows.
Key Concerns
1.
Data centres need a vast and constant
supply of electricity. According to the IEA, their power consumption is approaching that of the whole of the UK.
2. Because the computers inside generate enormous amounts of heat, most data centres also need vast amounts of water for cooling. As temperatures rise and rainfall becomes less predictable, this is becoming more and more problematic, especially for older data centres whose design was most likely based on historic climate data. Water consumption has been estimated to be about the size of the global bottled water market.
3. Data centres currently account for around 0.5% of global CO₂ emissions, but the IEA projects this will rise to 1.4% by 2030 — a footprint comparable to that of Japan. Some analysts also suggest the true figures may be considerably higher than companies are reporting.
4. As AI is applied to more and more situations from smart management of power grids to stock inventory and railway timetables and IoT,* which connects everything to everything else, data centres are proliferating. The US alone is expected to increase its capacity five -fold by 2035.
*[IoT refers to The Internet of Things -the means by which I can tell my washing machine to start at 11 a.m., even though I am at work, my fridge could -if it was a smart fridge, order milk directly from the supermarket and my son was able to scare a cat away from his front door, even though he was 16,000 km away in Munich].
What Can be Done
We all want the services data centres provide, but not the consequences, so this section is more about mitigation — reducing the contribution data centres themselves make to resource depletion and global warming. For this reason many data centre operators are now looking at ways to manage the enormous power and water demands created by AI and cloud computing in more sustainable ways, that don't contribute to the problem.
Energy Consumption
Search engine and AI provider Ecosia, already produces twice as much renewable energy as it uses for its own operations. It also plants a lot of trees.
Among other initiatives, Claude AI’s parent, Anthropic plans to both cover the cost of grid connections so that they aren’t passed on to consumers and to develop grid optimising tools to enable it to cut its workloads during peak periods to avoid strain on the grid.
Google was initially regarded as a leader in renewable energy procurement, claiming to match 100% of its electricity use with renewable energy purchases. However, critics argued that this relied partly on “optimistic accounting” — annual offsetting and renewable energy certificates, rather than ensuring that every data centre was actually running on carbon-free electricity in real time.
The explosion in AI demand has since derailed Google’s plans to become fully carbon-free by 2030 by sharply increasing its overall emissions. Even so, the company reported that it reduced data centre emissions by 12% in 2024 despite a 27% increase in electricity consumption. Its 2025 Sustainability Report also argues that AI-enabled services such as fuel-efficient routing in Google Maps and Solar API are reducing emissions in other sectors, partly offsetting its own growing energy use.
Water
Cooling has always been a concern for data centres and has usually been supplied by water, but few were designed with today’s, let alone tomorrow’s temperatures in mind. In 2022, London’s 40°C temperatures shut down both a Google and an Oracle Data Centre, while X (formerly Twitter), went down during California’s heatwave. Operators are now looking for new ways to cool data centres which do not involve using more water.
Microsoft is leading the way on this. In August 2024, it announced that all of its new data centre designs would use zero water for cooling through a closed-loop chip-level system, saving more than 125 million litres per facility per year. The bonus for Microsoft is that it will allow it to operate in one of driest but sunniest regions in Australia, which will make running on renewables much easier. Waterless cooling may reduce some auxiliary electricity demands such as pumping and cooling infrastructure, but the trade-off is complex because conventional evaporative cooling is often one of the most energy-efficient cooling methods available.
For existing data centres, there are several ways in which water demand can be reduced. Favoured strategies include – recycling water, making more use of air cooling, relocating to cooler climates, reusing the heat which is produced to run cooling systems and various design tweaks such as placing server racks further apart, or siphoning heat away for other purposes such as district heating. Much of this is very technical and of great importance to engineers, so I don't want to go into too much detail here. If you would like to know more click here or see the brochure "The Top Twelve Ways reduce the Energy Consumption of your Data Centre."
The industry is increasingly shifting from a simple “water versus electricity” calculation toward balancing water shortages, energy demand, heatwaves and pressure on electricity grids simultaneously — and using AI to do it.
Other Ways to Reduce Water and Power Demand
Floating Data Centres
Google began experimenting with floating barges as far back as 2011. While these were to demonstrate some of its other products, they included features such as seawater cooling and harnessing of wave energy. Although Google abandoned the idea due to rising costs and difficulty in complying with building and maritime regulations, other companies have jumped on the idea and have turned it into a commercial success.
Nautilus for example, began its first barge -borne data centre -Stockton 1, on the San Joaquin River, California in 2021. Its use of river water for cooling is especially appreciated in drought prone regions such as California and it also uses approximately 30% less power. A similar facility is currently in progress for Los Angeles and also for Marseilles in France.
Undersea Data Centres
Among the latest developments to overcome the cooling problem is to establish data centres under the sea. Once again, Microsoft led the way by conducting several trials off the coast of Scotland between 2015 to 2024. Called Project Natick, it used sealed cylindrical containers on the ocean floor near the Orkneys. It proved that the concept worked, but Microsoft ultimately abandoned it because of the cost of maintenance and deployment and problems associated with scaling it. However, other companies and countries have run with it, because it offers a number of advantages. They include the following:
- Because cooling is done by seawater, very little, if any, fresh water is needed
- Energy consumption is in the order of 40% less, meaning lower operating costs too
- Because they mostly operate at between 30 and 100 metres below the surface, the cooler ocean environment allows server racks to be placed much closer together and thus provide far greater capacity than the same area on land.
- They can be located close to undersea cables which enables then to achieve much faster data transfer speeds (with less lag time) than when it must first travel long distances over cables and land stations.
- They can also be located close to population centres which also means faster speeds.
- They do not compete for scarce land and cause less disruption to local communities
- Because equipment is essentially isolated in sealed pods, there is less disruption due to things like dust, human interference and so on. In Microsoft's trials, failure rates for servers were much lower - in the order of 1 in 100 compared to 8 in 100 for conventional data centres.
China is one of the countries which is investing heavily in this technology. Beijing Highlander Digital Technology / Hi Cloud has developed the Hainan Underwater Intelligent Computing Center in the South China Sea. It is described as the first fully operational commercially scaled subsea data centre cluster in the world. Each of its modules can process 4 million high definition images in 30 seconds - the equivalent of 60,000 computers, and it plans to have 100 of those.
Phase 2 of this project opened in Shanghai's Lin -gang Special Area in 2025. This will be world's first offshore wind-powered underwater data centre. Several large Chinese coastal cities including Xiamen, Shenzhen, and Shandong have also included underwater data centres in their five year plans.
Vulnerabilities of Subsea Data Centres
While undersea data centres promise significant savings in power and water use, as well as reduced land requirements, they also have vulnerabilities. Researchers have noted that submarine data centres may face novel forms of interference transmitted through water, alongside concerns about effects on marine ecosystems, such as the warming impact of discharge waters and uncertainty over whether such systems will attract or drive away species.
Although most such facilities are located between 30 and 100 metres below sea level, making them largely immune to strong wave activity, the impacts of rising ocean temperatures, shifting currents, and corrosion have yet to be fully assessed. A third factor is that placement, maintenance, and retrieval are more difficult than in land-based data centres, and their supporting infrastructure such as wind turbines and grid connections remains exposed to the vagaries of the weather.
While some companies are looking to the depths, others are looking to the skies.
The New Frontier?
Jeff Bezos of Amazon is also keen. Speaking at Italian Tech Week in Turin in October 2025, he predicted gigawatt-scale data centres will be built in space within ten to twenty years, arguing that continuously available solar power means they will eventually outperform Earth-based facilities. He specifically noted that giant AI training clusters would be better built in space.
However, CEO of Amazon's Web Services Division, Matt Garman, disagrees. Speaking at the Cisco AI Summit in San Francisco in February 2026, he insisted that Amazon's 900-plus terrestrial data centres are exactly where they'll stay for the foreseeable future.
Why Not?
On Cooling
People think that because space is cold, cooling wouldn't be an issue, but the opposite is true. Because there is no air to carry it away, the heat would need to be pushed outwards by using enormous radiator surfaces which would add to both weight and costs.
Radiation
Hardware also needs protection from radiation, either through shielding or specialised software, and chips would need replacing every five to six years — all of which has to be launched into space at enormous, though falling, expense.
Costs
The economics only work if launch costs fall dramatically. Currently they are around $3,600 per kilogram and would need to drop to around $200 per kilogram. This is not expected to happen until the 2030s at the earliest —so brace yourselves for earthbound data centres and their attendant problems for a while yet. There is another reason for pause too.
Environmental Costs
Although space - based data centres are presumed to be greener than land based ones, that does not usually include the emissions generated by space travel itself, though over time, those too may fall, along with many of the other costs. Not that Benoist was ruling out space -based data centres altogether, rather that their time had not yet come, because many of the other issues would need to be resolved first.
Meanwhile, on a more sobering note, I wonder how many megalitres of water these posts have cost? In the course of researching for this post, I discovered that even a simple question consumes the equivalent of one bottle of water, while more complex tasks use far more. It is worth asking where your data, streaming and AI are coming from and what systems they are using.
Ultimately though, as ChatGPT assures me, while it is worth thinking about where these systems are physically located, and what they rely on for power and cooling, the real impact sits with the scale of the infrastructure and design choices, not individual queries -and that’s where the pressure for change ultimately lies.
Thanks to Copilot for the illustration, to Ecosia, ChatGPT and Claude for information and some lively discussions. Search engines and journals are great, but alas, do not answer back or fix your typos!
This ends this series of posts. The next few will be somewhat lighter as I will be away doing something else, but I am also working on how we will manage the Next Industrial Revolution, that is, without a lot of us being caught under the wheels. Feel free to comment if you have any thoughts on this or experiences you want to share. I won't publish them if you don't want me to.
Comments