AI’s Thirst for Power

Despite its benefits, artificial intelligence (“AI”) will accompany unintended consequences. On the plus side, efficiencies will be found, information will be more readily available, and productivity will be improved. On the other hand, jobs (and entire professions) will be eliminated, fraud will grow exponentially, and do we really want machines running our lives?

Comedian Nate Bargatze does a bit about time travel. “If I went back in time, I don’t think I would make a difference.” Surely he would want to tell someone from the 1920s talking in a phone booth about all the cool gadgets that existed a hundred years later. “Everyone will carry a phone in their pocket one day.” “How do they do it?” After thinking about the answer for a few seconds, Bargatze  says, “I don’t know.” He adds, “in fact, I don’t think I could even prove I’m from the future… I would probably be worse off in the past than I am right now.” 

Back to AI. Has anyone asked, “how will they do it?”

Data centres – large facilities with supercomputers that process massive amounts of electronic information – that enable AI will accompany an unquenchable thirst for power. Have you experimented yet with a ChatGPT inquiry? If so, be aware that your ChatGPT entry likely used ten times more energy as a Google search.

Part of the reason so much power is generated is because graphics-processing units (also known as GPUs) radiate an intense amount of heat whenever used. Power is then needed to cool the environment from overheating. This accounts for 40% of the facility’s energy use. The other 60% relates to processing the inference.

Data centres account for almost 4% of the electricity demand in the United States today. According to the International Energy Agency, that figure will jump to 6% by 2026 – equivalent to all of Japan’s electricity market! By 2030, Goldman Sachs estimates it will be 8%. This huge jump in data centres means that 20% more electricity will be demanded by 2030 compared to today, according to Wells Fargo.

Can the grid handle this inevitable surge in demand just a few short years away?

Probably not. The utility companies, which are the businesses that deliver power to their communities, aren’t equipped to deliver this kind of ask. “Across the board, we are seeing power companies say, ‘We don’t know if we can handle this,’” said Andy Cvengros, managing director of data center markets at JLL. 

Northern Virginia – where 50% of the U.S. data centres are presently located – needs multiple nuclear power plants built to satisfy future demand. Arizona Public Service has been vocal that it lacks the transmission capacity to handle the projected demand unless major upgrades are made to their infrastructure. Texas has the same problem. Even Georgia is rethinking its policy to offer incentives to lure the property tax-paying data centres to their state. 

So, can’t the states and utility companies just upgrade their infrastructure?

That’s not as easy as it sounds. While a new data centre can receive the green light in 18-24 months to start construction, approving new transmission lines can take up to a decade. It involves huge land acquisitions, endless environmental reviews by state regulatory agencies, and fights to determine who will pay for each itemized cost. Because of these factors, the entire U.S. struggles to build 1000 miles of transmission lines each year.

This all comes at a time when governments are working to reduce carbon emissions and encourage electric vehicle adoption. Today, the closure of coal plants in Kansas, Nebraska, Wisconsin, and South Carolina are being delayed because of soaring power consumption. That being said, it’s unlikely that utility companies will turn off carbon-emitting power sources any time soon. Afterall, renewables aren’t very effective when the sun isn’t shining and the wind isn’t blowing.

Natural gas will surely remain an energy source. Goldman Sachs believes that natural gas will supply 60% of the power demand growth from AI data centres while renewables will provide the remaining 40%. Big technology companies like Microsoft and Alphabet are trying to take this problem off-the-grid and build their own private nuclear power plants.

The free market has historically found solutions to its problems. Cars emit far less pollution today than thirty years ago; mailboxes are obsolete; you probably can move your television without the help of two neighbours; and no longer do we tether to the wall when talking on the phone (just don’t ask that comedian Nate Bargatze how it’s done).

However, we believe AI will integrate more slowly than many expect because of the power shortage. For that reason, our preferred exposure to AI focuses on infrastructure, electric utilities, construction, copper, and natural gas pipelines for the foreseeable future.