News

However, this surge in AI usage creates a substantial need for data processing. In fact, some estimates suggest that up to 70% of data centre growth in the coming years will be attributed to the capacity required to support AI workloads. If we measure these growth forecasts in terms of power, depending on the source consulted, the demand from data centres will be somewhere in the region of 200 to 300 gigawatts (GW) by 2030.

The final figure will primarily be influenced by the types of integrated circuits used and their power consumption, as well as by other factors, such as whether the AI is deployed through cloud or edge computing. Regardless, meeting this increasing demand for energy presents a significant challenge that is far outstripping the capabilities of suppliers and grid operators, and this is delaying the construction of new centres.

A potent problem

Power supply is a growing concern in areas with the highest concentrations of AI data centres, such as the western United States. Utility companies have acknowledged that they will not be able to expand their transmission infrastructure quickly enough to meet the growing demand, and, as a result, there are fears that, at some point, the demand for power will surpass the available supply.

All the indicators suggest that this situation is likely to spread to Europe, including Spain, where ambitious plans are underway to build new data centres. And although we’re seeing a sharp increase in electricity generated from renewable energy sources, concerns around supply persist. Today’s power grids face a wide range of disturbances that impact the quality of electricity supplied. Issues such as micro-cuts, voltage surges, electrical noise, and frequency fluctuations mean that the power reaching our buildings, factories, and homes is often far from optimal. Therefore, it is crucial to have the support of an uninterruptible power supply (UPS), and it becomes even more vital when powering sensitive electronic components like microprocessors and application-specific integrated circuits (ASICs) found in data centres.

Furthermore, as mentioned above, the very continuity of the power supply is being called into question due to the rapid electrification of demand and the accelerated construction of AI data centres. This uncertainty is unacceptable for critical infrastructures like data centres, and here too, UPS systems can represent a solution.

That said, a UPS destined for AI projects must meet a number of requirements, including having a modular and scalable architecture that will allow its power to adapt to the changing requirements of the facility, including its expansion. This adaptability is also facilitated by paralleling the UPS and adding redundancy to ensure uninterrupted operation under any circumstances, both externally (in the event of power failures) and internally (when performing maintenance or upgrades).

Since the aim is to minimise downtime, the UPS should ideally also have hot-swap technology, allowing modules to be replaced or installed while the system as a whole remains fully operational. These aspects, together with high energy efficiency to promote sustainability, will be crucial in shaping the protection and future of the exciting world of AI technology.