Ferdinand Magellan was the Portuguese explorer who discovered the Straight of Magellan at the tip of South America, allowing his fleet to pass from the Atlantic to the Pacific Ocean and sailed (or rather, his crew sailed) around the world for the very first time. (Magellan didn't survive the trip, but his next-in-command, Spaniard Juan Elcano completed the round the world trip.)
Ferdinand Magellan and his crew charted a historical new course for trade.
And now we are in the midst of a new course - for trade, for industry, and for life in general: with AI.
As Industry (or Industree) 4.0 moves forward and mills continue to modernize and update, AI integration is surging. However, the staggering energy demands of AI reveal a significant challenge, raising surprising concerns about sustainability and resource consumption.
Did you ever realize energy demands of AI were staggering? They are. There's a whole lot to the energy side of AI data centers that isn't heavily publicized. Yet who isn't using AI in some aspect, either inside or outside, the mill?
As it turns out, no one (aside from AI employees) actually know exactly what the AI power usage is, but we can make some pretty good guesses nonetheless.
- AI has a Very Large Energy Appetite
Let's begin with a basic Google search. According to Google, a single search query uses about 0.0003 kilowatt-hours. If you got a 1-Watt LED bulb, the power used for that single search would light your 1-Watt bulb for a little over 5 minutes. On average, Google receives nearly 9 billion of these queries per day, meaning Google searches consume roughly 2,700,000 kilowatt-hours per day - enough energy used in one day to power the average American home for 225 years.
That's a lot of power.
Now let's look at AI. We're not told how much energy a search costs - that information is kept closely guarded by all AI providers, and consequently drives researchers nutty. But what we do know is that data centers are popping up all over, and being built close to each other to share energy between facilities if necessary.
For example. Virginia now has 340 data centers with 159 more planned.
The most middle-of-the-road estimate, from MIT Technology Review, reports their estimation (or best guess) of the energy usage for a single ChatGPT query was 0.3 watt-hours per message. But when you're using any AI chat, when do you stop with one query?
Using their estimates for AI, if you ask and refine 15 questions, generate and refine 10 images, plus make 3 attempts for a 5-second video, you've likely used about 2.9 kilowatt-hours of electricity--enough to ride around 10 miles in the average electric vehicle, or run the microwave for over three and a half hours.
Again, that's a LOT of power.
- Data Centers Use a LOT of Water for Cooling
Beyond the immense energy demands of AI, data centers also require cooling systems which use massive volumes of water that can put an enormous strain on local resources and significantly disrupt local communities. For example, check out how badly the water supply and quality was affected at one remote farm next to a new data center in this video.
Conveniently, cooling demands are partially met in cool weather. That's why, in Northern Europe, Denmark hosts 34 data centers, half of them located in Copenhagen.
And data centers in Nordic countries - Sweden, Norway, and Finland - benefit from lower electricity costs, due to lower cooling demand because of their colder weather. The largest actor amongst Nordic countries is Sweden, with 60 data centers. Half of them are in Stockholm, at the 59th parallel north. - Where cooler weather optimizes the operational efficiency.
- Regarding Power Supply...
Global electrical consumption is expected to grow by over 80% by 2050 due to things like Industry 4.0, electric cars, and AI, among other emerging technologies. While AI contributions to energy demands are expected to be relatively small, time will actually tell if that's the case, since the new data centers are numerous and experiencing explosive growth. - And these data centers are exclusively for AI, not for all the other electric gadgets and technologies in our lives. That energy is sourced from separate power plants altogether.
Furthermore, one obvious problem is, data centers need a constant supply of energy, and solar or wind can only go so far. To minimize the use of fossil fuels, tech giants like Meta, Amazon, and Google have responded by announcing goals to use increased nuclear power. In fact these three have joined a pledge to triple the world's nuclear capacity by 2050.
Moving Forward
Just as Magellan's historic circumnavigation opened new pathways for global trade, today AI represents a similar paradigm shift, revolutionizing industry, trade, and daily life as Industry 4.0 continues to move forward.
Yet this new AI journey comes with a steep energy cost. Also, the opacity surrounding AI's true energy costs and the strain on local resources, like water-scarce communities near data centers, underscore the need for transparency and, quite frankly, better solutions.
Like Magellan's crew persevered to complete their journey, hopefully AI's environmental trajectory will improve, companies learning from their mistakes, continuing towards improved energy production, and ensuring this amazing new technology sails toward a future leaving our natural resources in balance.
Come back next month when we discuss Quality |
