AI was on the ballot, indirectly if not directly, and the country's opinion on AI data centers was crystal clear.
AI was on the ballot last night, indirectly if not directly. The country's opinion on the future of AI data centers was crystal clear.
AI data centers were probably one of the big losers of the election. There is a lot of concern over how they are being developed, where they are being developed and the environmental impact they have. Those are good reasons to be concerned. What we can expect is that data center growth will miss its projections, and investment in new data centers will slow down. That in itself is not a bad thing.
But it changes the forecast. It will make new winners and new losers.
One of the big winners will probably be Apple. Apple is the largest provider of edge inference for consumers. It sells the machines people use to run AI locally. The reason is the architecture: a system on a chip, with the memory very close to the processor, makes a very efficient place for AI to run. Apple has also been building energy efficient chips for a long time. Together that gives you a lot of what we'll call tokens per watt.
The big AI labs will be winners too. If token supply is constrained, they can hold price on their tokens. The more expensive inference is to provide, the more they can squeeze the low cost inference providers. That works in their favor.
It also spreads out the cash they need to pay out of their balance sheets for all these data centers, because a lot of them won't make it through approval, or they will be delayed or slowed down. That keeps cash on their balance sheets. In the end, it is probably good for them as well.
The big losers will probably be the neoclouds and the data center providers outside the tech giants, the ones who have to stretch their balance sheets more and more for each build.
For consumers, inference won't get cheap as fast. That may slow deployment and keep AI more expensive to run. The more expensive it is to run, the less cost effective it is to replace people in their roles. A slower deployment of AI could give the country and society time to adjust.
But here is one thing. Electrons cross national boundaries without a problem. I would look for more data centers to be built offshore. You can move training offshore. You can move development offshore, and any inference that doesn't need to be close to real time. Electrons are fast. You still want to save hops, but I would look for offshore data centers to get a boost from this.
This is one more change that an already fast-changing industry will have to go through, adapt to and get better at. It will be interesting to see where it lands. What do you think will come from this?