By Philip Beanblossom

While consumer and business applications have utilized artificial intelligence to some degree or another for consumer applications for several years now, the AI “revolution” that’s taken hold recently is bringing algorithmic creations to a much larger audience. Previous applications of AI have largely been behind the scenes, like machine learning optimizing ride-hailing services and web searches. These certainly used large amounts of data, but the explosion in generative AI models will require a huge increase in data processing. Naturally, this processing will happen at data centers.

As the companies of “big tech” look to harness the AI revolution with their generative models, they will require ever-increasing amounts of data processing – especially as companies like Amazon, Microsoft, and Google already control the data center market through their cloud services, which independent AI models use for outsourcing their data needs.

All this data requires power – and the expansion of AI means significantly more power than previously required. A report earlier this year from Newmark found that while data centers typically required 10-14 kW per rack, server racks with powerful enough GPUs for accelerated AI adoption will require 40-60 kW per rack. With data center power consumption expected to reach 35 GW by 2030, nearly doubling the power needed in 2022, data centers will require a huge increase in electricity to ensure data centers can continue uninterrupted.

Data needs power, but from where?
Power is a finite resource as delivered in the United States. Though the AI revolution may be taking advantage of cutting-edge developments in generative models, the power grid still largely rests on technology developed in the 19th century. Buildings that require large amounts of power, like data centers, can require major infrastructure upgrades to the grid since there is a limit to how much power transmissions lines and transformers can deliver. Utilities also must ensure that once a data center is connected to the grid, there’s still enough energy to supply homes and businesses.

These issues may seem like more of a longer-term concern, and indeed, major upgrades to the electrical grid are costly and lengthy. The issue is that, as AI adoption grows and takes on ever more applications, the need for more data processing is going to outrun how quickly utilities can upgrade the grid. Cloud computing companies need to build new centers and bring them online at a rapid pace to keep up with the demand for AI technologies.

If power from the grid cannot meet this demand for new data centers, then where should the companies building the facilities source it from?

Temporary power is more than a temporary solution
Independently generated power can be the key to ensuring data centers can meet the need for data processing as the AI revolution takes hold. By providing electricity separate from the grid, modular, temporary power solutions reduce a building’s demand for grid energy and lessen the requirement for utilities to immediately expand capacity.

These solutions take the form of generators and energy storage but are more advanced than the aging, polluting generators commercial building operators may be used to. Tier 4 Final generators meet the most stringent EPA restrictions on emissions while still operating on widely accessible diesel. Compared to legacy diesel generators, a Tier 4F unit can realize emissions reductions of 94% of Nitrogen Oxides (NOX), 90% of Carbon Monoxide (CO), and 90% of Total Hydrocarbons (THC), all in a modernized unit that’s much quieter than older generators. Where natural gas is available and when emissions requirements are stricter, data centers can also rely on gas generators to provide power independent of the grid.

Modular power solutions have a proven history of enabling more immediate data center expansion compared to waiting for local utilities to expand grid power, provided this is even an option. In one instance, a major cloud provider met an increased demand for services at an existing data center thanks to several dozen 1200 kW generators, which supplied 48 MW of continuous power.

Don’t let infrastructure limitations hold back data centers
The rapid expansion of generative AI and growing applications for machine learning could cause a huge strain on power grids, but it doesn’t have to. Modular power solutions can relieve the pressure on utilities to upgrade electrical infrastructure, while also enabling the expansion of processing power that data centers need to support the AI revolution.

Philip Beanblossom is Sector Manager of Data Centers at Aggreko. He can be reached by [email protected].

Read Other Articles