Welcome back to the AI Power podcast. We'll get into our normal conversation regarding AI policy and the events of the day in just a moment, but obviously Tuesday was the midterm elections. Greg, I know you're from Kansas. Would you like to take a couple of minutes and kind of talk about your reflections on the goings on and certainly what happened in the state of Kansas?
Sure. I mean, I am from the same part of Kansas that Senator Elect Hamilton is from. He and I actually went to the same high school. I went to high school with one of his children. And I cannot express how completely impossible it seemed the idea that we would have a Democratic senator. I mean, not only has this not happened in a hundred years, but it has like never even come close any time, even in the previous sort of wave elections. So, when my friends back home were telling me that they thought actually it could be different this time, I was like, "Yeah, sure." You know, enjoy whatever you're smoking. But it didn't seem reasonable to me. And yet here we are and this is an incredible wave election. You can really feel just how much it has changed the policy conversation on so many topics of Washington, not least of which AI. Adam, you are a political animal. What are you hearing on Capitol Hill?
This was completely unexpected, Ray. I think everybody kind of understood that in the deep recesses of your political heart where you don't necessarily verbalize those things out loud, the House was going to flip and potentially the Senate maybe by a couple of seats. I don't think anybody had any measure of how far it was going to go. Certainly the South Carolina seat is completely unprecedented. Kansas with, you know, Senator Elect Hamilton is completely unprecedented. So this is a wave.
Mississippi, like there's a very real chance Mississippi could have a Democratic senator, which just sounds nuts to say out loud.
An extremely real chance, and the background to that particular election, where it's a senator who was nominated to a federal judge position who couldn't get the blue slip cleared by the senator that he then challenged for the seat, is absolutely crazy. Absolutely crazy. So just to put the specific numbers on this, the Watergate era shift was a D plus 16 shift, and that's really as far back as the scientific measurements go for a lot of the election data that we commonly talk about. The shift that we saw on Tuesday was a 13-point shift towards Democrats. By comparison, the Republican revolution of 1994, which was absolutely groundbreaking at the time, was a 6.8% shift towards Republicans.
Right. So the earliest era that you and I probably personally remember would have been the Republican Revolution, and that was significant. So I can't just state enough how significant this was.
So we should talk about the AI stuff in the midterms very soon. But there's one thing that I think is kind of immediately noteworthy, which is that the polls, to the extent that they were wrong, they were wrong by underestimating Democrats slightly. In the most recent elections, the polls have either been accurate or significantly underestimated Republican support. This is the first time it's underestimated Democrat support, even though broadly the polls were pretty accurate this time around. And the reason why that what that means is all these polls that have been coming out over the past 12 months showing ratcheting public discontent with AI, this election kind of legitimizes that polling and by extension legitimizes future polls that are going to come out. And so what you're going to see is companies and government affairs teams are going to start looking at those polls, as are politicians, and they're going to be making decisions based on it. They're not going to tell themselves, "Oh, whatever. You know, polling is always wrong." So, I think you can now expect that politicians and companies are going to want to start taking measures to actually shift public perceptions towards increasingly favoring AI because the expectation now has to be that if you don't fix that, you can lose an election on it.
I think that's right, Greg. I think that's certainly right. And the measure of the polls underestimating Democratic support really had to do with voters who were expected to stay home and instead of staying home voted for Democrats, right? So there's three decision points: many given two-way election vote for either the candidates or stay home. So those polls thought everybody was staying home. Those people were motivated enough, largely by kind of AI related policy issues, that they just voted in the complete opposite direction, right? So that it's going to be significantly motivating for the members of Congress that we have coming in in January.
We're leading off here kind of talking about what happened Tuesday. So, Democrats clearly swept the House and the Senate, and this is the first time that AI and data centers were in a real campaign issue. So, the three specific noteworthy things that we're talking through: more than half of the voters told exit pollsters that AI and data centers were a major factor in their vote. The second being that in nearly all cases where AI was specifically on the ballot, the anti-AI position won. And the third, AI related ad spending and campaign donations were absolutely colossal, right? Hundreds of millions of dollars associated with that. So, and to top all of this off, multiple candidates who called for a pause on Frontier AI development won Senate seats, right? That is not insignificant in these conversations moving forward. We'll go through what voters said, how the ballot measures went, who won, and what the winners said on the trail and since, what they plan now, and which AI proposals in Congress could actually move in January. So, to lead off here, before we get into the individual races, let's start with the voters. How much did AI actually matter in this election, Greg?
Yeah, so we have some good exit polling that came from NBC News. Now, exit polling is not always a perfectly reliable tool, but it definitely is an important indicator and tells you some useful information, even if you shouldn't believe it, you know, to the nearest percentage point. So, 56% of voters said that AI and data centers were either a crucial or a major factor in their vote for Congress. Now to compare that to some other issues, cost of living, which is like heavily tied to inflation, of course, 89%. The war in Iran, 74%, immigration enforcement, 72%. So AI and data centers a major issue at 56%, but very few people saying that that is the sole reason why they are voting for a kind of a thing.
Now, I think one big question here is this sort of generalized opposition to AI. To what extent is voter opinion tied to adjacent issues where AI has some overlap with the things that they care about versus what you might call more sort of intrinsic opposition to AI? What do I mean by that? Well, if you think that the Trump administration is too cozy with technology billionaires like Elon Musk or Jensen Huang and you oppose that, you see that as corruption of some kind. Well, then maybe you're saying that you oppose AI, but the actual issue that you're voting about is a corruption issue. Or similarly, the AI data center buildout is so massive that it's not only tied to overall economic growth trajectory, it's also tied to inflation. I mean, this is most obvious in the case of direct components like DRAM. But one that really affects most consumers is the cost of energy, which is really acute right now. So maybe you think you hate AI because of its effect on the cost of living issue, which again is a high priority, but those are areas where AI is sort of intersecting with the true policy priority of the voter, and that's what they care about, not AI per se.
I think the unanswered question for me is what is the share of the electorate that actually cares about AI on its own terms. This would be something like AI and safety issues, right? If you think AI is going to cause a cyber weapon or a bioweapon or child safety type issues, all of these things where like the AI is the thing you're voting for. A second way that it could be more of an intrinsic issue is AI and unemployment, right? If you think that AI is going to take your job, it's going to prevent your kids from having a good job in 5 or 10 years, those are the kinds of things where it's like we are intrinsically voting about AI. And I don't know that we actually have good polling on this issue, but I think one thing that is noteworthy to me is the gap between how many people think AI is an important issue versus how many people think that AI will hurt society. And that gap is incredibly small. 57% of people in this NBC News exit poll said that they think AI is going to do more to hurt society. And the framing there of that question hints at it being more of an intrinsic kind of an issue. Like the problem is the AI.
And then by contrast, if you look at the data center polling, three in four voters oppose a new data center in their own community. Well, that could be opposition that is more tied to corruption, right? If that data center construction got tax breaks that voters think is illegitimate, or cost of living, if they think that that data center is going to raise their overall energy prices. And so what that suggests to me is that kind of both things are true. On the one hand, voters really don't like what they see as the intersection of AI and other issues that are big important priorities for them like the cost of living, like corruption. But then also there is this building intrinsic concern about AI related to safety, related to jobs.
I think the biggest story here, however, is not where we are at this moment in time, although that's extraordinary. It's actually how far we've come, right? AI barely even registered as an election issue in 2024, and here it is as one of the major priorities that people are focusing on. And a lot of people, myself included, see a plausible scenario whereby AI is like the top issue or at least one of the top three issues in the 2028 presidential campaign. I mean, no less an individual than election forecaster Nate Silver has said that he thinks it's totally plausible that AI would be the number one election issue of 2028. So, that's my sense about where we are in terms of AI and voters' opinions on it.
It's a solid take, Greg. That's a solid take. So, clearly after every election cycle, there's a whole bunch of spin that occurs, right? In this case, we've got some parties, including AI companies, saying that it wasn't really an AI election because AI is rarely anybody's top topic, even though it's got adjacency to all of those different issues. Is that a fair read of voter sentiment from your perspective?
Yeah, and I think actually OpenAI had sponsored some polling that suggested that only 3% of voters said that AI was the number one deciding factor in their overall vote. I think that that might well be true, right? That relatively few people are saying this is the number one issue that decided my ballot right now. But I think that you would be ignorant, right, if you focused on that as the message from this campaign. I think we've seen two really noteworthy things. Number one is again the shift, the trajectory in polling on this, and the second one is how individual candidates perceived AI as a useful thing to focus on.
So Democrats made AI and data centers like a central part of the campaign message. In more than 130 federal and governor's races, they aired anti-tech industry or anti-data center ads. I mean, from 2020 to 2024, they ran two. And so, basically, what you have to conclude from that is that Democrats perceived this as an issue that they were gaining traction on, that it was when they were talking about it, they did better. To the extent that they surfaced this as an issue in voters' minds, they were more likely to win elections. And according to reporting by Semaphore, this is four times as frequently as Republicans mentioned it in their own advertising. So I think basically if you just look at the campaign's behavior, they think this really matters. And I guess I would also come to the Michigan senator-elect Abdul El-Sayed. He actually said that AI was the number one issue that his now going to be constituents would bring to him and say that they're concerned about, they would come and want to talk to him about AI. Now, a lot of that is anecdotal, but if you believe that the actors in this system are wisely acting in their own best interests, I think that behavior tells you a lot.
It absolutely does. So data centers specifically were literally on the ballot in dozens of towns and states, mostly in the form of bans and moratoriums. Let's go over how those votes went. Greg, how did it shake out?
Yeah, so there was a ton of places where data centers were in some way, shape, or form a part of a ballot initiative. Now ballot initiatives might be binding, right? You're like amending the charter of your city and like this is actually now part of the law. Or it might just be an advisory opinion to your elected officials where you sort of say, elected officials, we voted on this topic and we expressed our will 75-25, we do or don't want something to happen. There was 41 of these, and sometimes the initiative was prohibit the construction of data centers. Sometimes it was prohibit the construction of data centers above a certain size. Sometimes it was require that any data centers be constructed but only under certain kinds of provisions like energy usage or water usage kinds of conditions. But other initiatives were actually pro-AI, pro-AI data center construction, right? Like advising the local elected officials to encourage data center development and to negotiate on the community's behalf to have data centers be constructed.
So across those 41 ballot initiatives, sometimes yes meant pro-data center, sometimes yes meant no, anti-data center. But this is the crazy thing. In 40 out of 41 cases, voters chose the more anti-data center position of the two options available to them. And perhaps just as extraordinary, it was by a huge margin. We're talking like 75-25.
So connect that back to that poll that we talked about a moment ago where like only 3% of voters perceived AI as their number one issue. Well, that doesn't necessarily mean AI companies are safe, right? If this is not a top priority of voters because they hate it so much that if they're given the option to vote against it, they will choose that option. So, what does this mean for the buildout itself, Greg? Just nailing down on this. Will it slow down the buildout or kind of stop data center construction in these areas? What does that really look like?
Yeah. So, this is a really important distinction here. You know, I have previously said on this podcast that I did not think the New York state data center moratorium, which is pretty old at this point, we're talking like 6 months ago, was that big of a deal in the grand scheme of things because other states like Texas, Virginia, these are the places where data centers are really getting built at extraordinary numbers. Well, guess who just elected not just a Democratic senator in James Tyler Rico, but a Democratic governor in Gina Henosa. So she specifically ran part of her campaign for governor was on a moratorium, a pause on data center construction until they come up with, you know, what is the set of criteria that is going to make data centers a good deal for Texans, as she put it.
This is a big deal. Not in that I think it will actually stop data center construction, but I think it could slow. Right, as SemiAnalysis put it, you know, Texas is such an important part of the data center buildout that even a brief pause in approving new permits and so on can actually impact the national numbers. So even if these are short-term data center moratoria, you know, they could actually have implications. I sincerely doubt that the governor-elect is going to jeopardize Texas's reputation as a business-friendly state by slashing projects that are already underway. So I think this mostly affects projects that are proposed but have not yet received their permits or begun construction, that sort of a thing.
But I want to talk about like what actually might happen here aside from the delay. If you look at some of the things that are talked about, we haven't really seen evidence of intrinsic AI opposition. You know what I was saying before on safety grounds, the kinds of criticisms that Governor-elect Hinosa is making are about, for example, subsidies and tax breaks for data center construction. There's issues about like NDAs being signed with politicians, which makes, you know, the tax breaks seem corrupt because they want to be secretive about the tax breaks that they're given, and they want proof that power bills or water bills are not going to rise as a result of that data center construction. And then some groups also talk about, you know, wanting this construction to take place with union labor, which in many states it already is. But if you add all that up, no subsidies, tax breaks, no NDAs, no electrical bill or water bill price increases, union labor, all of these just translate to money, right? All of that is not an intrinsic opposition like we don't want AI to happen, the kind of thing that you see coming out of, for example, Senator Bernie Sanders. It's much more like a we don't want data centers to happen in our state under the current conditions.
Like the tax breaks for data centers is kind of a weird one from my perspective because normally you want to have tax breaks when you're looking to make your state more competitive. Like think about Amazon's HQ2, which is actually in Crystal City, Virginia, not too far from where I live. Amazon basically had a bake-off among many different metropolitan areas around the country, and the local governments bid to Amazon about what types of incentives that they would have to have Amazon locate its second corporate headquarters there, create all the jobs, create all the construction associated with that second headquarters. Well, you can kind of justify tax breaks on that ground because you can say, "Well, look, we only have to lure Amazon once, but then they'll have tens of thousands of high-paying jobs for a very long time afterwards." Tax breaks for data centers are kind of weird because almost all of the employment benefit is upfront in the construction. Takes an awful lot of people to build a data center. Takes way fewer people to operate a data center. So, like why we're offering massive tax breaks in the first place doesn't make a ton of sense to me. It goes against the traditional logic of when you want to incentivize corporations to make decisions with those kind of tax breaks. And now I think voters are basically saying, yeah, like why would we want to do that?
So the question then becomes if Governor Hinosa, if the governor-elect in Georgia, if the governors in some of these other states are now thinking about what kinds of conditions they want for the data center buildout to take place, I think the vast majority of them do just come down to money.
Per your point, Texas wasn't alone, right? What other governors have talked about this, right? Let's expand the conversation outside of Texas.
Well, I think another big one would be Georgia, where Governor-elect Kesha Lance Bottoms also talked about a statewide pause. Now, Georgia is a big factor in the construction of data centers. There's 2.9 gigawatts of data center capacity under construction right now. I mean, that's overtaking Northern Virginia, my home, as the biggest US construction hub. So, that would be a big one. Also states like Ohio, where you saw Amy Acton defeat Vivek Ramaswamy for the governorship of Ohio, and one of the things she ran on was a conditional freeze on projects that don't meet strict energy and water criteria. One thing to note, the 41 different ballot initiatives on data centers that we were talking about a moment ago, 18 of those were in Ohio. So this was a very salient issue in the state of Ohio. And now when you put all of that together, more than 40% of the planned data center capacity sits in states where incoming governors have pledged to impose some kind of pause on data center construction.
But now I said that this really all comes down to money. You know, how much is it going to cost for what they want? I think one interesting example in this is what's happened to memory prices. So if you think about a data center, a gigawatt data center, the sort of ballpark cost for constructing that data center is about $41 billion. And about 20% of the price of the data center is going to memory chips of some kind. Whether that's DRAM or NAND or high bandwidth memory, HBM, about 20% of the total cost of the data center is going to memory chips. And the price of memory chips has exploded over the past year such that if you were to build a gigawatt data center at last year's memory prices, instead of $41 billion, you'd be paying $36 billion. So the price has gone up by $5 billion to build one of these data centers. And that has not slowed down data center construction hardly at all. It is still proceeding at a rampant pace.
And so coming back to like what these new Democratic governors might want, union labor, no subsidies or tax breaks, guarantees that power and water bills won't rise. All of that is just money, money, money, money. And so far we've seen that this $10 trillion AI infrastructure buildout that is expected to unfold over the next few years has enough money. So a way that you could think about this, there's three prices that the AI companies could potentially pay for their data center. They could pay the subsidized price. That's the one they've been getting in many cases. They would love to pay the subsidized price forever. Probably not going to happen. Then there's the market price, which is no subsidies, but you're bidding against all of your competitors and it's a competitive market. Okay, they'd be willing to pay that price. But now I think they're going to face a third price, which is the political price. What is it going to take to make communities excited at the prospect of having data centers as a neighbor? And that's like all of those conditions that we just talked about. And you could even think about specific community benefit negotiated agreements being a part of the contract.
In Louisiana, all of the teachers in that state just got like a $50,000 one-time bonus because of how Louisiana handled the construction projects that Meta is undertaking in that state, building out all those new data centers. I bet you can bet that the Louisiana Teachers Union kind of likes data centers. And in some states, the cost of guaranteeing that power bills is not going to go up would be like nothing compared to the price of DRAM. In some places, it might be pretty expensive. Maybe even Texas, it would be pretty expensive. But like in North Dakota, you know, if you have to guarantee that a few hundred thousand people are not going to pay, you know, such and such more on their heating bill, I do think that that is a price, a political price that this industry can afford to pay, at least for now.
Yeah. And Greg, one of the points that a number of stakeholders, including Senator Dave McCormick from Pennsylvania, have made is that if the data centers are implemented in the right way, they potentially could lower property taxes for the surrounding areas. Right? He said that multiple times over the past year. It's tough to do that if you got a significant tax break going on for the data center at the same time. So rationalizing these things makes a lot of sense. So let's turn to Congress. Which Senate and House races turned on AI and data centers?
Well, I think the poster child for all of this has to be Ohio, where absolutely now Senator-elect Sherrod Brown ran against John Husted, as he tagged him as this quote, "the face of data centers in Ohio," end quote, and he won by five points in that race. And now you might remember on this podcast we talked about a leak...
National Republican Senate Committee memo where they said this quote: 'More than any other thing in this race, data centers are the anchor hanging around Huey's neck.' End quote. And Brown was facing like an avalanche of money coming from the technology sector. That includes $64 million in crypto-related campaign contributions, the most the industry has ever spent on a single race. And he still won. While he was lieutenant governor, Huey oversaw $2.5 billion in tax breaks for data centers for 226 data center construction projects over the past 7 years. So that is the kind of thing that Republican voters, Democratic voters, independent voters are recoiling against, right? Is like, why are we in our taxes subsidizing this construction? These people are already billionaires, or in the case of Elon Musk, trillionaires. Really, shouldn't they be doing something to help us rather than us doing something to help them? And I think that is a message that is going to be heard very loudly and clearly, not just by Democratic politicians who clearly took it to heart as part of their campaign message, but Republican politicians. Right? You saw that memo from the Republican Senate campaign committee. And I guarantee you this is going to shape the behavior of Republican candidates who in some cases are going to be thinking about the 2028 presidential election. Another state where this was a huge point was Texas. So Elon Musk alone spent $224 million trying to elect Texas Republicans who have been supportive of data center development. James Talarico beat Ken Paxton by a mile. And it wasn't just that, you know, Talarico was some kind of remarkably talented candidate. Democrats won a lot of the statewide election offices in Texas.
17 offices that Democrats...
