Boosting the energy efficiency of data centres

IEC E-tech

article here

The exponential rise of artificial intelligence (AI) demands a rapid growth in innovative power solutions to store and process data more sustainably and efficiently. The IEC provides the standards required.

AI-powered virtual assistants require ten times (2,9 watt-hours) more electricity to run a query than traditional search engines (0,3 Wh), according to the International Energy Agency (IEA). Multiply that many times over for the data crunched by big tech and start-ups seeking to develop AI and apply it on an industrial-scale and it is easy to see why the race for artificial intelligence is also a race for energy.

Power is an issue…

If, as many believe, AI is the engine of our economic future, then the means to power it must be prioritized by governments worldwide. In turn, the acceleration in demand for AI is leading to greater power density in data centres, the facilities housing the servers, storage systems and networking equipment used to train large language models (LLMs).

The IEA estimates that data centre (DC) electricity consumption will grow more than four times faster than the total electricity consumption of all other sectors between now and 2030. US data centres could consume 17% of all electricity in the country by 2030, the Electric Power Research Institute (EPRI) calculates.

Today, an average hyperscale AI DC is capable of consuming upwards 100 megawatts (MW) - as much power as 100 000 homes. That figure is set to be dwarfed as dozens of projects in the multiple gigawatt (GW) range come on stream. Data centres of 1 GW are being built in India; 6 GW in Riyadh, and 10 GW in Louisiana.

…but so is water!

Not only does this surge threaten energy supply across national grids, there are titanic knock-on effects on local water systems. A single 40 MW facility can consume over 1 million tons of water annually for the cooling of electrical components “putting vast amounts of fresh water resource under increasing pressure,” warns the World Economic Forum.

Concern that AI demands could lead to water shortages have been raised in multiple countries including the UK, US and Uruguay. A third of DCs under construction today are projected to face water scarcity in two decades time. Chile has hosted more than 30 DCs over the past decade but there is now a “backlash as datacentres have drained water from drought-stricken wetlands, consuming billions of litres annually.”

There are also concerns about the CO2 emissions generated by these massive facilities. DCs fuelling AI could be forced to invest in sustainable and decarbonized power generation if state governments in Australia and the EU have their way. That means an obligation to invest in renewables and nuclear sources of energy. Another incentive is to reuse excess heat from data centres to heat homes and businesses, according to the EU Energy Commissioner. Improved energy efficiency is becoming imperative together with the increasing resort to decarbonized and sustainable energy to power these mega facilities.

Making data centres more energy efficient

Data centre energy efficiency must shift from optimizing individual components to “optimizing whole systems,” according to Philippe Vollet, Secretary of IEC SC 23K,  a subcommittee which standardizes electrical and electronic energy efficiency devices, for instance for smart buildings. He is also the Chair of the IEC Advisory Committee on Energy Efficiency (ACEE). The committee has published a guide which helps IEC Technical Comittees include energy efficiency aspects in their standards.

Vollet argues that the first and most decisive factor in data centre efficiency is location. Cooling alone accounts for 30–50% of total energy consumption, making climate a fundamental design variable. “A data centre in Norway will never have the same cooling burden as one in Dubai,” he says. Location also determines the carbon intensity of the electricity mix: a facility in France, with its largely decarbonized grid relying essentially on nuclear power, may have a very different footprint from one elsewhere.

“Proximity to the grid matters too; building hundreds of kilometres from a connection point is inherently inefficient,” he says. Cooling remains the “battlefield” of efficiency gains. Traditional air cooling is reaching its limits at around 40 kW per rack, pushing operators toward liquid cooling and now immersion cooling, where components are submerged in dielectric fluid. (For more on liquid cooling read: Data centres to bear the brunt of climate change | IEC e-tech). Vollet describes immersion as “the best way” to remove heat, though it will require updated safety and performance standards as adoption grows.

IECEE, the IEC System of Conformity Assessment Schemes for Electrotechnical Equipment and Components, is also a key enabler for energy efficiency. As one of the world’s most recognized and trusted multilateral certification systems based on international standards, it offers third party testing and certification for a wide range of electrical and electronic products that align with over 3 000 standards. The use of IECEE certificates helps to ensure consistent quality of products and services and facilitate access to international markets by aligning products in any country with global requirements. Its portfolio includes a number of standards for energy efficiency.

It also runs the IECEE Electrical Energy Efficiency (E3) programme, a globally standardized approach to testing and verifying energy efficiency for electrical and electronic equipment, based on IEC International Standards. 

Reusing heat from huge server facilities

A major emerging opportunity is the reuse of heat. Across Europe, pilot projects (such as one in Bulgaria) are feeding waste heat from data centres into district heating and cooling networks. Recycling half of the waste heat from DCscould meet the heating needs of nearly 4 million European households, according to the EU. While data centres are not designed to be energy producers, Vollet sees heat recovery becoming a “strong secondary use” of their energy footprint, with new standards needed to integrate facilities into urban energy systems.

Beyond cooling, electrical architecture needs addressing. “Every conversion step wastes energy, so operators are moving toward higher‑voltage distribution and direct current power, reducing the number of conversion layers. Data centres may also become grid‑flexibility assets, using their large battery reserves to support the grid during peak demand,” Vollet explains.

Metrics and standards are important

Environmental metrics are broadening too. Once dominated by Power Usage Effectiveness (PUE), which is still key, the sector now also tracks water usage (WUE) and carbon usage (CUE). “Harmonized global calculation methods are essential to this effort,” Vollet says.

That’s where IEC Standards come in: ISO/IEC JTC 1/SC 39, the joint technical committee of ISO/IEC working on sustainability, information technology and data centres, offers a range of design and practices for building and managing DCs. These include KPIs for water usage effectiveness (ISO/IEC 30134-9); measuring server energy effectiveness (ISO/IEC 21836); and a methodology to calculate and present the renewable energy factor of a data centre (ISO/IEC 30134-3:2016); while the document ISO/IEC TR 23050:2019 describes the treatment of data centre metrics in circumstances where electrical energy is stored and exported from within the data centre boundaries.

The IEC also plays a crucial role in the achievement of the UN’s Sustainable Development Goals (SDGs) by providing standards that ensure safety, compatibility, and performance across global markets. Of significance here is ISO/IEC TR 21221:2025, a report which describes the delivery of functional, economic, environmental, social, intellectual and personal benefits by AI systems as perceived by their stakeholders. ISO/IEC SC 42 prepares standards in the field of AI and has published a technical report which looks at the environmental sustainability aspects of AI systems, ISO/IEC TR 20226.

The EU is backing efforts to drive a more circular and efficient energy system. It plans to introduce a ratings scheme marking the performance of data centres regarding energy and water use and sustainability. Ironically, AI-based operation and maintenance optimization plays a role here by potentially saving up to €94 billion a year by 2035, according to the EU.  “Regulation and standardization are proven drivers of energy efficiency,” Vollet concludes.

Nuclear energy is becoming a player

Renewables remain the “fastest-growing source of electricity for data centres”, on track to meet half of the demand by 2030, estimates the IEA. Nuclear energy is also viewed as an important source of decarbonized energy and is expected to become even more so over the next decades. Small modular reactors (SMRs) will come on stream helping to double current global nuclear operational capacity by 2050, according to the International Atomic Energy Agency (IAEA).

Pre-fabricated SMRs could be manufactured and assembled far quicker than the average decade it takes for traditional nuclear plants but start-up costs, regulatory hurdles and getting community buy-in remain challenges. However many pundits stress nuclear’s advantages. “Only nuclear energy can meet the five needs of low-carbon power generation, round-the-clock reliability, ultra-high power density, grid stability and true scalability,” argues IAEA Director General Manuel Grossi.

As evidence of demand, one US SMR developer founded in 2019 was recently valued over USD 9 billion with investors including a hyperscaler with which it plans to add 5 GW of new nuclear power by 2039.  Provided there is “stronger government support” there could be more than 1000 SMRs deployed by 2050, with a total capacity of 120 GW.

IEC Standards provide the global framework for the safe use of this energy. IEC SC 45A  is a subcommittee inside TC 45: Nuclear instrumentation, which was set up to develop standards for the instrumentation, control and electrical power systems of nuclear facilities. It cooperates with the IAEA which sets global safety standards for nuclear energy. The subcommittee's standards cover the entire lifecycle of electrical and electronic control systems of nuclear power plants, from design to decommissioning. It has just published the third edition of IEC 61513, which establishes the general requirements for control systems that are important for safety.

Conformity assessment can help floating data centres

Despite obvious technical challenges, floating data centres are emerging as one of the energy efficient solutions to solve both water scarcity and excessive land use problems. Their main advantage is their significantly lower cooling costs (by being surrounded by water) and consequent reduced carbon emissions.

A number of projects are being devised which include plans by a consortium of Japanese companies to pilot a data centre off the coast of Yokohama; France’s first floating DC launched in Nantes, on the Loire River; a floating data centre park in Singapore is due to open in 2028 and a joint venture to develop floating data centre infrastructure has been signed between a US AI developer and a South Korean electronics device maker.

Off the coast of Shanghai an underwater data centre, claimed to be the world’s first, is now in operation. Windfarms on the water’s surface generate the power to run servers 10 metres underwater. According to the project developers, the system reduces electricity drain by 22,8%, eliminates water use, and cuts land use by more than 90%.

An even more ambitious innovation is being developed by a US company focused on harnessing ocean energy for clean power which will drive AI compute onboard. “While traditional wave energy systems tend to be located close to shore so electricity can be sent back through cables, the waves with the strongest and most continuous energy are further out in the open ocean,” says Garth Sheldon-Coulson, co-founder and CEO of Panthalassa. “Capturing energy there effectively could solve a major part of the global energy problem.”

Its floating ‘node’ consists of a large white sphere mounted to a vertical structure extending down below the water's surface. The repeat motion of water inside the tube generates a high-pressure jet of water which is released through a turbine, which spins a generator.

Instead of transporting energy to power a land-based data centre, the idea is that thousands of floating nodes would directly power onboard GPUs with satellite links transmitting data between the nodes and customers. This is one of the project’s Achilles heel, since relying on satellite transfer means dealing with limited bandwidth, signal delays and complications if multiple nodes must coordinate to handle larger AI workloads.

Another is making the transition from experimental to industrial scale marine energy generation. Various marine energy projects exist around the world, some at more advanced stages than others. None yet can claim to be fully commercial. To move into that space, they rely on a strong standardization framework, with specifications developed by IEC TC 114: Marine Energy Conversion Systems and certified through one of the IEC four conformity assessment systems, IECRE, the IEC System for Certification to Standards Relating to Equipment for Use in Renewable Energy Applications. It was established a little more than ten years ago to help provide third party certification and testing services for all power plants producing, storing, or converting energy from wind, marine and solar photovoltaic (PV) energy. 

Space is the final frontier

An even more extreme approach is to put data centres in orbit where solar energy is unlimited and round-the-clock. “These giant training clusters will be better built in space, because we have solar power there, 24/7. There are no clouds and no rain, no weather,” said the CEO of a hyperscaler at a tech event in Italy last October. “We will be able to beat the cost of terrestrial data centres in space in the next couple of decades.”

A US firm claims to have successfully tested edge processing tasks in space, including data handling for a Texan AI developer, from a data centre “the size of a hardback book” as part of the payload on a US satellite launch in 2025. “This is where the future begins for this new resilient layer of critical global infrastructure,” says the firm’s CEO. “By proving that our technology can operate in space, we are one step closer to establishing the [area between Earth and Moon] and the Moon as the ultimate off-Earth storage and data resiliency solutions.”

In December 2025, another US company claimed that its satellite was the first to run a version of Google Gemini in space and the first spacecraft to train an LLM. It envisages gigawatts of compute will be deployed in space in the near future.

The European Commission has also explored the feasibility of orbiting data centres. A report published by a French and Italian aerospace group determined that deploying data centres in space “could transform the European digital landscape, offering a more eco-friendly and sovereign solution for hosting and processing data.”

The project aims to deploy one gigawatt of capacity before 2050 and suggested it would return “several billion euros” on investment between now and 2050. According to a white paper from one of the space data centre developers, the continuous illumination allows orbital solar arrays to achieve a 95% capacity of solar energy generation compared to 24% for terrestrial farms, while peak power generation in space is 40% higher due to the absence of atmospheric losses.

Nonetheless there are clear risks. Commercial exploitation of low earth orbit is a frontier science with multiple challenges including the difficulty of remote maintenance, the possibility of launch failures, and the need for a solution to cool equipment because conventional cooling systems don't work well without gravity. The cost alone could be prohibitive with every kilogramme sent into space costing at least USD 3000.

But who knows? The development of tech solutions is advancing at such a rapid pace that predictions that seemed a little mad only a few years ago are now becoming reality.