Well, there's a lot of numbers that are circulating in the conversation, like the idea that in 40 state ballot initiatives, the anti-AI position won in all but one of them. Yes, that sounds bad. Until you find out that 18 of those were in Ohio. So we're counting the same thing again and again. The actual reality is that Governor Hinojosa, sorry, Governor-elect Hinojosa of Texas has said for the folks that already have a shovel in the ground, she's not going to get in the way. So this is not about the planned AI data center buildout that is already underway for the next year, 18 months, 24 months. That's going to continue, and that's going to bring tens of gigawatts of new AI compute capacity online. And when the AI models come online, the performance they're going to deliver is going to be jaw-dropping, and people are going to be clamoring for that.
Why are we building all this capacity? It's because right now, the revenue of these companies is constrained by supply and their ability to actually serve. So the voters have said that they don't like the data center buildout under the terms with which it was proceeding, which is like tax breaks for data center construction, NDAs so politicians can't talk about the deals that they cut. Those are things that are all addressable with more money. And if there's one thing that the $10 trillion AI infrastructure bill has, it's a lot of money that they can deploy to assuaging voters. So you're seeing through this, you don't see AI data centers crossing the border to Mexico or going to the UAE or even space anytime soon, for that matter. There are going to be states that make home for AI data centers, including many that are looking at them with a different lens.
So I think what we're after separate is what's already permitted, what's already construction underway. And that stuff is overwhelmingly skewed, tends to be that's a lot of new data center capacity that there's already a shovel in the ground somewhere bringing this online. So 2027 and 2028, those are going to be astonishing growth in AI compute capacity no matter what happens. What the governors-elect are really talking about is the future of that. And what are the terms of that new construction going to be? Are there going to be tax breaks to try and attract construction? Are they going to be building it with union labor? And are there going to be requirements that the companies guarantee that nobody's power bills are going to go up? And if you address those issues and it's just money, money, money, then the governors don't yet have any remaining complaints.
I think one thing to think about is AI polls quite negatively right now, but a lot of its polling is tied up in other issues, like the perceived connection between AI and the cost of living, those power bills going up, or perceived connections between AI and corruption, the shocking amounts of campaign contributions. When you start redeploying that river of money to communities, to bringing them benefits, rather than requesting tax breaks from them, you might see the polling on this really flip. And we've seen the data center industry weather major cost increases before. If you just go back 12 months ago, a gigawatt data center would probably cost you something like $36 billion just from memory chip price increases alone. That's up by more than 5 billion to $41 billion a gigawatt. So this is an industry that has an astonishing amount of demand for an astonishing amount of capacity underway, and an ability to weather price increases they can afford, making the communities feel a little bit better about what's being built there.
Well, Greg, as you talk about the demand picture, a week ago, the conversation we were having is whether or not this was getting to be bubble-ish, in that the market may be overestimating what ultimate demand was going to be, that maybe there was an overdoing it on the infrastructure buildout. Is there a chance that if things are slowed down, understanding your point about how much capacity is already in the works, but if things are slowed down, could it actually reach more of an equilibrium, the demand and the buildout of the infrastructure to support it?
Yes. And I think, you know, the previous commentator you had talking about folks moving off to the side, seeing things, how we're going to evolve. I think that's a very reasonable interpretation of where we are, but that mostly applies to the starting construction of what will begin and come online in 2028 and beyond. What the market is going to see is the biggest results of the bet that was made in 2025. That's what's coming online now. That's what's going to be coming online over the next 12 months. And from everything that I'm seeing in terms of the performance of these AI models, their relevance to different sectors of the economy is growing all the time. The coders experienced it first, but now white-collar professionals are experiencing it. And so many different industries are about to get a taste of just how much AI can do for them.
And that might really shift opinions on willingness to spend, return on investment, and all of this. So I think taking a pause from the investor side and seeing how things are going to develop, that can obviously affect the valuations today. But what we should expect 12 months from now is an AI buildout that is powering AI models that are so much more capable, so much more relevant to so many different sectors of the economy. And people are going to pay a lot for that capability.
