Home News UPS for artificial intelligence

UPS for artificial intelligence

UPS for artificial intelligence

Es the great protagonist of current technology Thanks to its recent advances and the rapid rise in the use of tools such as ChatGPT, Copilot, and Gemini, generative artificial intelligence (AI) is making its way into numerous professional, educational, and leisure activities, and also needs Uninterruptible Power Supplies (SAI/UPS).

AI is generating a huge need to process data. Some estimates suggest that 70% of data center growth in the coming years will be due to the capacity needed to run the workloads required by AI. If we measure these growth forecasts based on their power, and depending on the source consulted, Data center demand will reach between 200 and 300 gigawatts (GW) by 2030.

The final figure will be influenced primarily by the integrated circuits used and their power consumption, as well as other factors such as the AI ​​implementation method, whether in the cloud or distributed (edge). In any case, meeting such energy demand represents an enormous challenge that is far beyond the capabilities of suppliers and grid operators. This is delaying the construction of new centers.

A powerful problem

Power supply is already a problem in areas with a higher density of AI data centers, such as the western US, where power companies acknowledge they are unable to expand their transmission infrastructure quickly enough. There is, therefore, fear that at some point, demand will significantly outstrip supply.

Everything indicates that this situation could soon reach Europe and also Spain, where ambitious data center construction projects are underway. And while electricity generated from renewable energy sources is also experiencing a sharp increase, doubts persist.

On the one hand, the electrical grid already routinely suffers from a wide variety of disturbances that affect the quality of supply. From micro-outages and surges to noise and frequency fluctuations, the signal reaching buildings, industries, or homes is far from optimal. That's why it's important to have the added support of an uninterruptible power supply (UPS), whose role is even more important when powering particularly sensitive electronic components, such as microprocessors or application-specific circuits (ASICs) that house data centers.

Furthermore, as noted above, the continuity of the power supply itself is in jeopardy due to the rapid electrification of demand and the accelerated construction of data centers for AI. This uncertainty is unacceptable in critical infrastructures like data centers, and UPSs are also the solution for this.

However, a UPS designed for AI projects must meet several requirements, such as its modular and scalable architecture, allowing the UPS's power to adapt to the changing needs of the facility, including its expansion. This adaptability is also facilitated by the parallel connection of the UPS and the addition of redundancy to ensure uninterrupted operation under any circumstances, both external (in the event of power failures) and internal (during maintenance or upgrades).

Since the goal is to minimize downtime, it's advisable for UPSs to also incorporate hot swap technology, so that modules can be replaced or installed while the entire system is fully operational. These aspects, along with energy efficiency that increases sustainability, are destined to define the protection and future of a technology as exciting as AI.

Source: https://www.salicru.com/los-sai-para-la-inteligencia-artificial.html