For The Next 2-5 Years There’s An Extraordinary Imbalance In Favor Of Demand For Power

By Eric Peters, CIO of One River Asset Management

“I’ve got fifteen examples of companies where AI is expanding profit margins from between 5-40%,” said Lone Star, a Top-5 best performing US endowment CIO. “And yet, no one seems to believe this story,” he said, long, confident, convicted. “There’s a company whose revenue is up 20% and costs are down, that never happens.” Costs nearly always scale up with revenue, but not now. “When I meet Wall Street analysts, they tell me that cost savings will accrue to software companies and consumers. But no way. This will mostly boost margins.”

“Let’s say you’re a CEO with 14% margins that jump to 16%,” continued Lone Star. “Your rivals don’t compete that extra 2% away. And let’s say another CEO has 20% margins that jump to 28%.” 40% increase. “Maybe some of that gets passed to customers and some gets competed away by rivals, but you still end up at 23%, or something like that,” he said. “And all this is incredibly deflationary, allowing the Fed to probably cut rates twice this year,” said Lone Star. “But even so, the S&P probably chops around between now and the election.”

* * *

“Hard to make long-term predictions about something of this magnitude,” said the investor, a business builder, iconoclast. I’d called to check in on the latest in AI. He was early to the theme, as usual, and has been racing to build out the energy infrastructure to support this new form of intelligence. “For the next 2-5 years there’s an extraordinary imbalance in favor of demand for power. But 5yrs out to 10yrs and I’d be inclined to bet we have a power surplus,” he said. “My focus is to live in the 5yr window where I know the supply/demand imbalance” (See “The Next AI Trade“).

“Trying to predict electricity usage for AI in the out-years is difficult,” the investor continued. “You can do lazy linear analyses around megawatts needed, and they’re not necessarily wrong, but capitalism is clever.” If copper prices rise enough, aluminum can be used as a substitute. “But I strongly believe that the AI itself will begin to use its intellectual capacity to improve its own efficiency. Right now, we’re just throwing power at AI in a geometric fashion. But at some point, we’ll be asking the machine to be more judicious in its use of power.”

“The parable is the US; in its early industrialization, power usage moved in lockstep with GDP,” he continued. “Then it shifted, GDP would grow at 5% and power usage would grow at 1%.” Power demand from 2000 to 2022 was flat, but GDP rose 50%. “It’s lazy intellectual thinking to slavishly apply linear or geometric projections for power ad infinitum. If you’re a trader, it can make you a lot of money doing just that. But as an investor in long dated assets, it can get you into a lot of trouble even if this is the greatest investment boom of my 30yr career.”