Sad, unlikely pairing: AI & coal

The resource that powered the first great industrial revolution, coal, is somewhat ironically/sadly still helping to power the most recent industrial revolution — AI.

There are around 3,000 data centers in the US (33% of the world’s total) that consume vast amounts of electricity (2% of US total) to power them and water (3–5m gallons per day) to cool them. The contribution of greenhouse gas (GHG) emissions from data centers is roughly equivalent to that of the aviation industry. These figures are only set to exacerbate as the “AI wave” morphs into a tsunami. AI workloads will account for ~20% of total data center demand by 2028, pushing the amount of energy data centers are expecting to use to 10% of the world’s electricity by 2030.

“We do need way more energy in the world than we thought we needed before,” Sam Altman, chief executive officer of OpenAI, whose ChatGPT tool has become a global phenomenon, said at the World Economic Forum in Davos, Switzerland. “We still don’t appreciate the energy needs of this technology.”

Stepping back from the AI boom for a moment, this is all occurring at the same time when a great electrification initiative is underway to reduce the dependency on fossil fuels across sectors: electric vehicles, heat pumps, and green hydrogen all require electricity.

The sectoral growth in US demand

The demand for electricity is forecast to rise at 2.4% CAGR between 2022–2030.

So, where will all of this new electricity come from?

Unfortunately, from many of the same resources that powered the first industrial revolution hundreds of years ago. Power companies are scrambling to satisfy the needs of data centers and new factories in a country where the grid is already strained. To cope with the surge, some power companies are reconsidering plans to mothball plants that burn fossil fuels.

“Soaring electricity demand is slowing the closure of coal plants elsewhere. Almost two dozen facilities from Kentucky to North Dakota that were set to retire between 2022 and 2028 have been delayed, according to America’s Power, a coal-power trade group.” (source). In the Kansas City area, a data center along with a factory for electric-vehicle batteries that are under construction will need so much energy the local provider put off plans to close a coal-fired power plant indefinitely.

There is a cruel sense of irony that prehistoric carbon created from dinosaurs hundreds of millions of years ago is combusted to power cutting-edge AI applications being used to discover new drug combinations, map the galaxies (oh, and serve you the next personalized funny cat video on your Tik Tok feed).

Powering servers with a renewable form of electricity is critical. The challenge is that renewable energy is intermittent (e.g. the sun isn’t always shining / the wind isn’t always blowing), but there are ways to shift certain interruptible compute workloads (like training AI models) to the times that it is sunny/windy. Think about it this way — we can continue to build more dirty sources of 24/7 baseload power or adapt the way we do computing to one that more closely matches intermittent clean sources.

Enter stage left — Build AI.

We’re a team of specialists that wants to make AI training more accessible and importantly environmental. Not only are we deploying data centers in parts of the country with the cheapest, cleanest power (e.g. W. Texas, N. Dakota), we’re operating our data centers in conjunction with the grid. This means pausing AI models being trained at their checkpoints and turning off our data centers during the day (~5–8pm) when the grid is most strained, when power prices are the highest / the power is predominantly coming from fossil fuels. We’re working in partnership with utilities to reduce their peak loads and we pass those energy savings (in the form of lower GPU prices) to our customers.

Lower cost & lower environmental impact — start training your AI models with Build AI today. To get started, please complete our short questionnaire detailing your requirements. It is time for the new AI industrial revolution to not be powered the same way as the last one.