In early 2023 we were hearing about the new power demands that will come as a result of the processes needed to make AI/machine learning work. Not much changed until Spring when news of AI companies gobbling up entire data centers to meet their requirements. Some of the 100+ MW asks almost seemed like a fantasy.
It seems like only a few years ago onsite substations became common, and everyone had signed their promises to be carbon neutral by 2030. Providers were left with a new twist to the problem to not only go green but also go bigger than ever imagined.
Why aren’t large data centers just covered in solar panels to get a green source of energy? It’s because solar just isn’t a robust enough point at the moment. Simple math says that one can achieve about 100kW per 100,000 sqft of roof-space. Furthermore, AI data centers are going to be shrinking their physical space as every 3-5 cabinets pull up to a MW of power. We could be looking at a 50,000 sqft building pulling around 50MW!
Other great alternatives while we wait for solar and wind to evolve are SMR’s, onsite natural gas generation, and even hydrogen. These are all great choice with one immediate issue… they’re not immediate. Think for a second how long licensing for your own small nuclear reactor or gas/hydrogen plant might be if a simple substation is held up for 3-5years because of transformer shortages and permitting.
The industry is going to see data center providers connecting to already built gas plants and nuclear reactors behind the meter to meet the veracious power demands of AI companies desperately needing power. This will allow speed to market. Also, data center providers won’t have their scope of responsibility doubled by now running a power plant and a data center in tandem. Attaching to one or two existing plants will bring 300 MW data centers online by Q3 of this year if not early 2025. By this time we’ll see AI evolve from a novel co-pilot / assistant closer to the more mature product that it was intended to be.
easycoloquote.net 