Happy earnings season! What’d you get me? I’m joking, but I did get you this newsletter. We’ll talk about the bummer summer big banks had, pull apart UnitedHealth’s rising profits, and figure out who sets all those quarterly earnings expectations in the first place. But first, my colleague Caleigh Wells is here to explain why data centers use so much water. — Tony Wagner, newsletter editor

Up until a couple years ago, data center cooling was just like cooling your computer. When it gets hot, a fan comes on to blow cool air across the hot chips. In cold places, that might come from outside. In hot places, it might require air conditioning.

But artificial intelligence is using more and more power, and chips are getting jammed into tighter and tighter spaces, all in the name of harder, better, faster, stronger computations. All of which makes more heat.

“The heat flux is going to be, in the near future, too high for air cooling to always be a satisfactory solution,” said Jonathan Boreyko, a professor of mechanical engineering at Virginia Tech.

That’s where liquid cooling comes in. Instead of fans, some data centers might rely on pipes, filled with water and ethylene glycol (also known as engine coolant). Those pipes send liquid through the CPUs, which grabs their heat, and shoots it into a cooling tower.

“During that process, some of the water is evaporated, and that's how the heat is actually moved away,” said Eric Masanet, a professor of sustainability in emerging technology at the University of California, Santa Barbara. “It's the same thing that happens when we sweat after a workout. The sweat evaporates, and it takes the heat from our bodies and moves it into the atmosphere.”

Masanet says with air cooling, data centers might have to tack on an extra 30% of their power draw just to keep cool. Switching to liquid drops that cooling tax down to 10%.

That might not sound like much, but new data centers are now ten times the size of the old ones.

“That 10% of power that goes into cooling a gigawatt data center is equivalent to the total power it took to run some of the largest data centers on the planet before AI,” Masanet said.

In addition to the power use, there’s the water use. Researchers at University of California, Riverside, figured out that generating a 100-word email with AI uses about a bottle of fresh water. A hyperscale data center can use up to 5 million gallons of water every day. Which is about the same as a town of 30,000 people.

That water gets used up in two places.

Those sides would be health care providers and insurance companies, Silvers said, as each tries to squeeze more money out of the other over the course of your care. Who’s winning? Silvers said better tech is part of why UnitedHealth reported higher profits with fewer enrollees today.

We often talk about publicly traded companies “beating” or “missing” expectations on their quarterly earnings reports. Who are these people setting expectations, and why do they have so much power? You’ve always wondered, and we have the answer