Scott Caravello: That sounds great. And, you know, I think we will be able to really interestingly connect all of these topics to the White House's evolving position on AI, which we have talked about many, many times on the podcast.
Katherine Forrest: Yeah, absolutely. And, actually, now that I've given us the order of the agenda, I want to actually change the order again, because I think it's actually useful to start with data centers, because the data centers are sort of the, part of the physical infrastructure. And then we'll go into sort of the regulatory initiatives. And, you know, I spend a lot of time talking about data centers.
Scott Caravello: You absolutely do. And that's talking about why they're so important. They are responsible for training an AI model, handling the inference needs of AI models and tools, and all of that takes these really large facilities that we call “data centers.”
Katherine Forrest: Right, they consist of, first of all, acres of land and warehouses or large structures that have racks of GPUs and CPUs and all kinds of networking equipment and wires and whatnot.
Scott Caravello: And they need tons of power, electricity, and that can compete with the community's own needs. And they also need incredible amounts of desalinated water in order to cool all of that equipment you just mentioned and to keep it running.
Katherine Forrest: Yeah, you know, one of the things I think we've mentioned before is why it's desalinated, which is actually important and why you can't just sort of pump it out of the ocean. Because if you do pump out salt water from the ocean, it can actually corrode some of the equipment. So, you do have a need for the same kind of water if it were otherwise sufficiently, you know, pure enough that would be competing with drinking water or other water that also needs to be desalinated. And right now building data centers requires a lot of cooperation from different authorities, land authorities, power authorities, water permitting authorities, and some would argue that there are environmental impacts, and, so, there might be some environmental reviews that have to be studied. And, so, we've got a number of states that have data centers today and other states which are considering the possibility of building data centers and Maine, where I have, as we all know, my summer home, which I'm looking forward to a whole other, this is a whole other topic, Scott. I'm really looking forward to getting back up there and being able to talk about like Maine events, from Maine. But anyway, Maine has been at the forefront recently of leading efforts to try to prevent a data center entry into the state. And there are other states that have tried this and we'll talk about those in a few minutes, but the Maine legislature had actually passed a bipartisan bill that was then presented to Governor Janet Mills to prevent data centers. It went by name LD307. But on April 24th, Maine's Governor Mills vetoed that bill. And so there will not be a Maine moratorium on data center development. So, I think that's an interesting sort of moment in this back and forth community, sometimes support and sometimes opposition to these data centers.
Scott Caravello: Absolutely. I think we will maybe talk a little bit more about it, but it is, you know, the furthest along that one of these bills have gotten. So, it is notable. But, as for some more details about that bill, it would have paused permitting for any new data centers requiring more than 20 megawatts of power until November 2027, while a newly created data center coordination council studied the impacts of that construction on ratepayers, the grid, and the environment. But you have to put that 20 megawatts in perspective, that's enough to power about 16,500 homes. And, before you ask, Katherine, yes, I just had that statistic right off the top of my head. And so, you know, we're also not necessarily just talking about those data centers that are really important to AI, because that figure, 20 megawatts, is really quite a bit shy of what you'd see at major AI data centers with specialized AI training facilities. Their power requirements reach into the hundreds of megawatts and even up to a few gigawatts, which is where the biggest AI campuses are.
Katherine Forrest: Right, good flag. And it's worth mentioning that Governor Mills agreed, in principle, with the pause. She actually said in her veto letter that the moratorium is appropriate given the impacts of the massive data centers in other states on the environment and electricity rates. So, she was recognizing those concerns, but she said she vetoed it because the final version did not include an exemption that she'd requested for a specific $550 million data center project in a town in Maine that had experienced significant job loss a few years back when a paper mill had closed. And, anyway, there are strong feelings on both sides of this. The legislature tried to override the veto, but it fell, you know, short of the two-thirds majority needed. So, the moratorium is dead for now and the data centers may go forward. We'll have to wait and see.
Scott Caravello: Yeah, and to put this whole bill and data center opposition in context, there are about a dozen other states that have introduced similar moratorium bills just this year. So, that includes Georgia, Maryland, Michigan, New Hampshire, New York, and so on and so forth.
Katherine Forrest: Well, I think the so on and so forth is important because it's also Oklahoma, South Carolina, South Dakota, Vermont, Virginia, Wisconsin, and others. But the reason I think it's important just to interrupt you, Scott…
Scott Caravello: No, sure, sure.
Katherine Forrest: …is because again, these are red states and blue states, we’ve got concerns from folks all across the spectrum. And, you know, part of it may be education, part of it may be allocation. But there are real concerns by communities about the impact on sort of water access, the amount of power that will be taken, whether or not rates are going to go up, whether there can be rate guarantees. You know, again, maybe part of it's going to be education. Part of it may be some additional support from the companies wanting to build the data centers. But there certainly is a fair amount of opposition. And, well, you know, this is why I had to sort of jump in—you were just going to go “so on and so forth.”
Scott Caravello: Exactly.
Katherine Forrest: I thought the ones that, the states you were leaving out, the “so on and so forth,” were important to the story, you know. None of those states that I've just mentioned, the so on and so forth, by the way, had advanced as far as Maine's opposition did. You know, Maine had that bipartisan bill that was right before Janet Mills that she then, Governor Mills, had vetoed. But most of the other states have actually had bills that have stalled in committee or been voted down. And, as listeners may recall, we've discussed the White House's AI action plan for March 2026 in prior episodes, and they had emphasized, it had emphasized, and actually the prior executive orders from July and from December of 2025 had also emphasized accelerating data center buildout because it's a key component. That infrastructure is a key component to providing what's needed for the AI model. So, you know, in the July 2025 executive order, the administration had issued some, I guess it would be really statements around streamlining federal permitting and suggesting that there really ought to be streamlining around federal permitting and in the December and the March executive orders, both of those made clear that there should be things that went beyond general permitting reforms for data centers, but that federal legislation should not preempt state laws on AI compute and data center infrastructure, which could include authority over zoning, land use, utility regulations, which is where most of the action is. I need to pause. Now, the pause, this is going to be struck by Juliana and Robert. So are we right? Is this, made clear that beyond these general permitting reforms, federal legislation should preempt state laws? Should not? Okay, good.
Scott Caravello: That’s an important point. And it brings us to our second topic where the federal government is getting involved with state-level AI laws. And, like we mentioned before, that's in Colorado where the future of the state's landmark AI law has been a topic of discussion for quite a while now, or honestly, even since it was first signed into law. Though we should probably caveat that we are recording this episode on May 13th. And by the time the episode comes out next week, the future of the law could finally be settled.
Katherine Forrest: Right. So we're talking about Colorado's SB 24-205 and it's frequently referred to as the Colorado AI Act and it was signed into law way back in May of 2024, so two years ago now. And it was supposed to take effect originally in February of this year. And it deals with potential algorithmic discrimination in situations in which AI tools are involved in what are called consequential decisions, which could involve things like healthcare, housing, credit, education, and there's a whole sort of list of these. And, at the time the governor of Colorado signed the bill, he actually made it clear then that he wanted to see some amendments before it went into effect.
Scott Caravello: Yeah, and that signing statement was motivated by what he viewed as a “complex compliance regime” that the law would create.
Katherine Forrest: Right, so before the original effective date, the lawmakers negotiated a host of amendments to the bill, but when they couldn't agree, they kicked the can down the road and they pushed the effective date eventually to June 30th, 2026, which would be coming up, really, not too far away. And, as written, and as it would have taken effect, on June 30th, 2026, the law would have required impact assessments, risk management programs, annual reviews, consumer notifications, actually a whole host of things which some other so-called algorithmic discrimination laws in other states also require, though Colorado's AI Act was the most far-reaching.
Scott Caravello: And, so, all of that brings us pretty close to present day and where we are right now.
Katherine Forrest: Right. So, about a month ago, actually it was April 9th, a little more than a month ago, April 9th, 2026, XAI owned by Elon Musk filed a lawsuit that challenged the constitutionality of the Colorado AI Act. And it sought a preliminary injunction to prevent it actually going into effect. The company actually was arguing that there were First Amendment violations by compelling AI systems to adopt state-defined viewpoints.
Scott Caravello: So, then just about two weeks later on April 24th, the US Department of Justice sought to intervene in that lawsuit siding with XAI. And so there the DOJ argued that the law jeopardizes the United States' position as the global AI leader by requiring AI systems to incorporate discriminatory ideology that, “prioritizes preferred demographic characteristics and outcomes over merit-based outputs.” They were arguing that through the guise of what's being called prevention of algorithmic discrimination, AI developers and deployers are actually being made to discriminate. And so then they also argue that because the methods to prevent the algorithmic results would require accommodating certain messages, it censors speech based on content and chills that speech.
Katherine Forrest: Yeah, and it's worth pausing on to really think about and worth people going and looking at the complaint that the DOJ filed in Colorado. But in pausing on what all of that means for a whole host of algorithmic discrimination laws and this marks the Trump administration's first litigation effort aimed at limiting state-level AI regulation and it took aim at the most common, I would say, state AI regulation, which is the algorithmic discrimination legislation. And, you know, I thought we would be able to see, or that we would see, something like this this year, but the language that's used in the complaint, I think, is very interesting and worth taking a look at.
Scott Caravello: But… there is another beat to this saga.
Katherine Forrest: Well, that beat is that on April 27th, the federal magistrate judge in the court where this lawsuit was pending with the DOJ's intervention piece before it also issued an order pausing enforcement of the law. And actually, all the parties, both sides agreed to it.
Scott Caravello: And, so that brings us to where we are today, which is that there might be replacement legislation on its way. There's a rewrite of the Colorado AI law on the governor's desk right now. And we should note that the new bill really represents a scale back of the Colorado AI Act. And, among other obligations, the bill, which is referred to as SB-189, scraps the current law's impact assessments and annual reviews.
Katherine Forrest: So, what's left is, it's more than this, but it's essentially, I want to just boil it down for a moment, this new AI law in Colorado would have a transparency and notification framework, but also a 60-day period to cure violations before any enforcement. It would be, actually, a very different sort of scheme altogether.
Scott Caravello: Totally, and so what I have been really focused on and I think is really interesting about it is just how the bill also narrows the technology that's in scope under the law. It's no longer the same broad definition of AI, but it's scoped to “automated decision-making technology.” And I think that there are two quick notes that we should cover to highlight just how far the bill has been scaled back in changing that covered technology.
Katherine Forrest: Right, so first is that the covered automated decision-making technology in the prior Colorado AI Act is now limited to technology that materially influences a consequential decision in one of the areas we mentioned like employment or lending. And if you take those words at face value that reads like a higher bar than substantial factor, which was a hook under the original law, and which scoped in technology that assisted in a decision that is capable of altering that decision.
Scott Caravello: And so then second, the bill is explicitly carving out ChatGPT-like systems. If there is no intent, advertising, marketing, et cetera, for the system to be used to make a consequential decision. And if the system is also subject to an acceptable use policy that prohibits its use in a consequential decision. And so as a final note on the Colorado bill, even if it is passed, it wouldn't take effect until January 1st, 2027.
Katherine Forrest: Yeah, so the short point is that the June 2026 Colorado AI Act is put off. It will be put off who knows how long, but if the new bill were to actually be signed and to go into effect, it wouldn't go into effect until early 2027. But before we wrap up, there's one more development that we have to flag now on the federal side, and that was happening in the first week of May.
Scott Caravello: Right, so on May 5th, the Commerce Department Center for AI Standards and Innovation announced new agreements with Google DeepMind, Microsoft, and XAI. And under those agreements, the companies are going to voluntarily provide their frontier models to the center for pre-deployment evaluation and research.
Katherine Forrest: Right, you know, this really comes after, in my view, it's not surprising that it comes after the whole Claude Mythos piece, but what's interesting is that there are certain frontier labs that are not part of at least being publicly out in front on this initiative, including Anthropic, as well as OpenAI or Meta. I don't know where they are in terms of their involvement or conversations that's not public. But the ones that are public are, as you'd mentioned, Google's DeepMind, Microsoft, and XAI.
Scott Caravello: Definitely. And so two important points about the nature of the initiative that some of these companies did agree to. And that first is that it is entirely voluntary. And second, providing a model to the Center for AI Standards for its review doesn't prevent the company from releasing it. It's basically an arrangement for first access, but not a gate to actually releasing the model.
Katherine Forrest: And the press on this has been pretty interesting. The New York Times reported that the White House is considering an executive order that would create an AI working group of tech executives and government officials to examine some other potential oversight procedures. Again, I think it's, you know, tied both in time and substance to the Claude Mythos situation, as well as some of what the March 2026 AI framework and executive order had previewed.
Scott Caravello: And that would all be a significant shift from the administration's earlier hands-off posture on AI. But for now, what we have is a voluntary research and evaluation framework.
Katherine Forrest: Right, and so taking a step back, we've got Maine and Colorado that have got their sort of situations bubbling along, and we've got some federal testing. And so that's really looking at different kinds of governance and regulation or regulatory adjacent things related for AI, you know, being worked on across every level of government in real time, and the relationship between federal policy and state regulation is very much in flux. So, we'll continue watching and looking, but I want to talk about space orbital stuff next time, all right?
Scott Caravello: Fantastic.
Katherine Forrest: Okay, all right, I'm Katherine Forrest.
Scott Caravello: And I'm Scott Caravello. Don't forget to like and subscribe.
