Powering Up for an AI Future

AI makes a range of demands on data center power topology and management. Future-proof prime power, backup load-stepping and the use of renewable fuels are three of the most critical.

With no sign of a slow-up in the phenomenal pace of AI infrastructure investment, data center operators need to be on the winning side and implement the smart and sustainable upgrades required for AI and HPC readiness. 

According to a recent report from IDC, AI data center capacity is projected to have a CAGR of 40.5% through 2027, with AI data center energy consumption forecast to grow at 44.7%, reaching 146.2 Terawatt hours (TWh) by 2027. AI workloads will consume a growing proportion of total data center electricity use.

Given this growth in AI-driven demand, it’s not surprising that, in a recent report, CBRE observed a growing price disparity between data centers that have been built or upgraded with the high-density power demands of AI in mind and existing facilities that still lack adequate infrastructure for AI.

Densities are rising fast and will continue to do so, reaching as high as 100 kW/rack, possibly higher. At AVK we advise customers on how best to configure their power infrastructure to position themselves for AI in a resilient and sustainable way.

Related:How AI Can Lead to Operational Transformation in Smaller US Companies

This is not just a matter of environmental responsibility; it is also a strategic necessity for the long-term viability of the AI industry. There are of course several considerations and pitfalls to upgrading next-generation power systems effectively and quickly.  If businesses fail to take them into account they can lose market opportunities, risk downtime and undermine sustainability targets at a time when renewables and emissions reduction are key purchasing criteria for customers.

Putting Prime Power in Place

AI is putting demands on the grid that it can't cope with. This is clearest in the established hubs -  including Dublin, Frankfurt, Amsterdam and, increasingly, in London. Waiting times for capacity are several years long in some cases, creating major headaches for operators. Where power simply isn’t available, an alternative solution is needed, something that bridges the gap between supply and demand and between data center operators and grid requirements.

This is where a smart microgrid comes in.  If customers want to build a 100 MW microgrid to power their facilities it must be harmonized with the local grid. It should offer seamless dual-fuel systems (or gas only, depending on standby setup and local regulations) that deliver stable power and that are compatible with renewable feeds and fuels. Ideally, when local authorities unlock heat and power grids, they should also deploy dispatchable power that can give back to local communities and grids via smart control systems. Importantly, it needs to be implemented fast to keep ahead of the competition.

Related:Role of FD-SOI in Powering AI-Driven IoT Applications

Maybe the best thing about adding a smart microgrid is that it tackles all these challenges and covers both the short and long-term needs of the operator. For example, when a microgrid is designed and built well, it makes data centers net contributors to local communities, with the potential to become part of the national infrastructure, contributing to a smoother energy transition.

Backup: The Load-Stepping Challenge

Both existing and new facilities must also pay close attention to their backup solutions, which may not be able to keep up with even smaller increases in power density. As customers decentralize their UPS to increase power densities up towards the 50/100 KW level, they need smooth load-stepping, which means the ability to gradually increase the electrical load on a backup power system, such as a generator.

Smooth load-stepping ensures that the backup power source can handle the growing demand without overloading or causing instability. It's like driving a car, if you just let go of the clutch and slam down the accelerator, you will stall. If backup engines cannot cope with the high, variable and unpredictable energy demands of AI they will stall, causing power failures as well as increasing emissions and wasting fuel.

The point of failure is the generator, and many are not currently up to the job. At AVK, we work with the world’s leading generator manufacturers, enabling data center organizations to handle AI-level load-stepping with ease, because they have industry-leading load-step capability with extremely fast ramp-up and minimal frequency and voltage deviations. 

Even when full AI demands are not yet fully planned this load step capability gives peace of mind. What this means for the data center customer is more available power and less risk as they add higher densities.

Buy Into Renewable Backup Fuels

Another key consideration when adding new backup generator sets (gensets) to scale up power requirements is to think of the future. Generators are often diesel engines but not all can run on renewable fuel. At AVK, we always recommend engines that can run on Hydrotreated Vegetable Oil (HVO), in fact, 90% of the diesel generators our organization currently works with run on it.

Future-proof fuel selection will make a huge difference to data center operators’ sustainability goals as they step up their power requirements. HVO is fossil-free and generates 90% less CO2 emissions than diesel.

Unlike biodiesel, it is fully renewable as it is made from feedstock and cooking oil. It also performs better. Unlike traditional biofuels and diesel itself, it doesn’t suffer from performance issues in cold weather, nor is it hygroscopic - attracting moisture from the air - which would cause the fuel to degrade over time. Biofuels can start to degrade within months, and diesel within a year, but HVO can be stored for up to a decade, easily offsetting any price premium. It also has International Sustainability Carbon Certification (ISCC) and meets the international fuel standard EN15940.

The clinching argument for HVO is how straightforward it is to move to it from an existing diesel setup. If an operator has half a tank full of diesel, HVO can simply be added into the mix, with no impact on performance. This makes it an excellent and practical interim fuel solution, while new fuels are developed and tested.

Understanding Your Power Needs

Sustainable, high-density infrastructure can be assembled piece by piece as long as organizations apply the same principles at every step and build for the future. Microgrid design, load management, and renewable fuels are just a few of the more important elements to consider as power levels increase to support AI, but there are many more. The market is also changing fast. New solutions are being developed constantly, and a lot of them now rely on AI to work effectively, making it part of the solution as well as part of the challenge.

Perhaps the most important principle for tackling these and other issues is that the more data center businesses understand power the easier it is to scale it and manage it sustainably.

AVK has implemented thousands of power projects across Europe delivering some 3.5 GW of capacity, and our organization is technology-agnostic, meaning each project has its own unique challenges and requires a fresh approach to design and implementation.

Understanding the specific local needs of the facility and keeping abreast of the latest innovations will help to create long-lasting low-impact solutions for the data centers of tomorrow.