The morning after the midterms, Rep. Yassamin Ansari of Arizona called for a pause on advanced AI development and a ban on artificial superintelligence, and she wanted Democrats to make it a day-one priority. Bernie Sanders went bigger on November 5. He posted that Democrats must enact a national moratorium on AI data centers, pause advanced AI development and ban superintelligence. Three demands, one post.
RedState says investors were already dumping AI-linked stocks as more Democrats came out against the technology, and it blames the whole agenda for the selloff. That’s a partisan site, so I wouldn’t trust its explanation of why. The timing though is hard to ignore.
I read both posts twice and then did what I always do. I set the politics aside for a minute and asked a boring question: does my API bill go up, and does my next job exist? Nobody in those posts answered that, and the replies mostly shouted past each other. So this is my attempt to sort out who is saying what and what the bills actually say, and then to work out what it could mean for people who write software for a living.
Two bills, and a lot of people mixing them up
“Democrats want to stop AI” sounds like one thing. It’s at least two bills plus a growing pile of state and town fights, and they do very different jobs.
The older one is the AI Data Center Moratorium Act. Sanders and Rep. Alexandria Ocasio-Cortez announced it on March 25, 2026, and the idea is simple: no new large AI data centers until national safeguards are in place. The law firm Troutman Pepper Locke, summing up the conditions, listed proof that data centers don’t push up household electricity bills or damage the environment, plus a community veto and union labor on the job. One developer-focused write-up says the freeze would cover facilities above 50 megawatts, but I couldn’t find that number in anything official, so treat it as unconfirmed. Rep. André Carson put out a release on June 29 saying he’d joined the effort, and he named job losses as one worry. The states are moving faster than Congress. New York’s governor, Kathy Hochul, has paused new hyperscale data centers, a state bill (S9144) would pause development more broadly, and Troutman counts lawmakers in at least ten other states with similar bills. Towns are doing it too, East Greenbush in New York and Oldham County in Kentucky among them.
The second one is newer and a different animal. It’s the Ban Artificial Superintelligence Act from Sanders and Rep. Greg Casar of Texas. Forklog reports they first announced it on September 3, and the AP covered the formal rollout on September 23. It would permanently ban building or deploying superintelligence, which the bill describes as a system that beats human cognitive ability or can resist being shut down. It would also pause advanced AI development until federal rules exist. A new Department of Artificial Intelligence would run all this, and it would work with the State Department on international bans, with export controls as one possible tool. Gizmodo’s subhead says Sanders wants superintelligence treated like enriched uranium.
Penalties are in there too. A post quoted in RedState’s coverage says up to 20 years in prison, and I couldn’t confirm that from the bill itself, so don’t quote me on it.
Two holes I couldn’t fill. Roll Call described the proposal as unnumbered back in September, and I never found a clear line for what counts as “advanced AI” in the pause. That second gap matters a lot for developers, so I’ll come back to it.
Who’s actually saying this, and who isn’t
It isn’t all Democrats. Sanders is an independent, for one. The loudest voices are Sanders, Casar, Ocasio-Cortez and Ansari, and Common Dreams (an advocacy outlet, not a news report) adds Reps. Frank Pallone and Maxine Waters to the data center moratorium side and says more than 500 groups have joined a call for a full national freeze. Sanders told Al Jazeera in March that lawmakers are “way behind” on understanding AI, and Ocasio-Cortez said Congress has a moral duty to stop expansion until the existential harm is addressed. Casar’s argument on the superintelligence bill is more like an industry-regulation argument. He told Roll Call that voluntary commitments based on each company’s own risk tolerance aren’t how any other industry works. On NPR he compared it to the Cold War nuclear agreements with the Soviet Union and said Trump is wrong that slowing down means losing to China. Sanders was shorter with the AP: “It doesn’t take a genius to say, ‘slow it down.’”
Then there’s the pushback inside the party. Sen. John Fetterman called the data center moratorium “China First.” Axios asked nearly twenty Democrats seen as 2028 contenders whether they backed Sanders’s call to stop building machines humans can’t control, and only one answered directly. Common Dreams complains that several governors still welcome the industry, and says Schumer rallied Democrats to kill something called the Ratepayer Protection Act, which he called a fraud. I haven’t read that bill, so I can’t tell you what it did.
The public numbers are messy. Casar told NPR that Data for Progress, a progressive think tank, found about 70% of Americans back an immediate superintelligence ban and about 25% oppose it. Another write-up says 68% overall, with 72% of Democrats and 63% of Republicans in favor, but I couldn’t trace who ran that poll. Question wording matters a lot in these, and I haven’t seen either questionnaire. The Republican number is the odd bit, because it suggests this isn’t purely a left-wing idea. The AP noted that Sanders attended a conference on the issue alongside Republicans, Steve Bannon among them. And an earlier statement calling for a superintelligence moratorium had right-wing media figures and tech pioneers on it, per a CO/AI write-up.
The industry reaction is split in a way I didn’t expect. The AP says multiple employees at leading AI companies endorsed the Sanders-Casar bill, and OpenAI safety researcher Juan Felipe Cerón Uribe said superintelligence could go extremely right or extremely wrong. On NPR, the host said leaders at Anthropic, OpenAI and Google had backed a coordinated slowdown over the weekend, and Casar agreed that their CEOs were talking about a pause. I couldn’t find those statements myself, so I’m not building anything on them.
Small tangent. The NPR transcript ends with a note that Anthropic and Google financially support NPR, and I liked that they said it out loud.
The other side is just as clear. Trump told the UN General Assembly that AI would be renamed “super intelligence,” repeated his opposition to new rules, and said the Justice Department could step in if needed. Racing China is the central argument, and the AP said AI safety was expected to come up in Trump’s talks with Xi Jinping that same week (Sanders had urged him to push for limits). Rep. James Comer, a Republican from Kentucky, talks about the race with China and about whether the grid can handle the demand. Elon Musk answered Ansari with four words, “It’s a Brave New World.” Some people who take the risk seriously still won’t sign on. Gary Marcus would accept a temporary pause and a dedicated regulator, but told Forklog’s readers that a permanent unilateral ban is too broad. Brendan Steinhauser of the Alliance for Secure AI credited Sanders and Casar for taking the issue seriously and stopped short of endorsing the ban. An analyst named Castro warned that a ban in one country could simply hand the lead to China.
I’m not going to tell you who’s right on the politics. The best case for the ban is that people inside the labs say the risk is real and voluntary promises have no teeth. The best case against is that America slows down alone, China doesn’t, and the danger is still a prediction. My opinions from here on are about money and jobs, where the evidence is easier to weigh.
The argument underneath all this
Strip away the names and both sides are arguing about one thing, which is who gets to decide how fast this goes.
Casar’s version, from his NPR interview, is that no company should be allowed to build AI that humans couldn’t control or contain, and he put it in the same category as nuclear weapons and other weapons of mass destruction. The AP adds that some leaders inside the industry have been urging Congress to set guardrails, which is the point supporters keep coming back to. If the people building it say they want rules, the argument goes, then waiting for a voluntary fix is strange. Casar also says the international piece is doable, since the bill only tells the new department to go and pursue agreements.
The skeptics’ version is about enforcement and about who else is in the race. A ban in one country doesn’t stop work in the others, and the AP calls the China argument one of the central points against sweeping limits. There’s also the plain fact that the risk Sanders and Casar describe hasn’t happened yet. Forklog’s write-up calls the targeted risks largely speculative, while the benefits of today’s AI are real and already here. That’s Forklog’s framing, and the bill’s backers would say that waiting until the risk stops being speculative is exactly the problem.
What it could do to software work
Here’s how I split it. There are three dials, the ban, the pause and the data center freeze, and they don’t hit developers equally.
Ban first. Going by the AP’s description, it targets systems that outthink humans or can resist being switched off. The coding tools I use today don’t claim to be either one, so a ban by itself shouldn’t change daily work much. If I had to put odds on it, 70/30, this is the piece with the smallest effect on software jobs. That’s mostly true, though I’m oversimplifying. If the pause defines “advanced AI” loosely, it could reach products people use right now, and I never found where the line is.
The pause is fog. Nobody has said which models, which companies or which date, so any claim about what it does to developers is a guess, mine included.
I can still sketch two versions, as long as it’s clear they’re guesses. If the pause is narrow, meaning only the biggest frontier training runs and only for a set period, most developers would see slower model releases and not much else, since the tools they use today keep working. If it’s broad, meaning anything labeled advanced AI, then product teams might wait on sign-off from a regulator that doesn’t exist yet, and that would hurt small companies far more than big ones because big ones can afford lawyers. Neither version is written down anywhere I could find.
The data center freeze is where the money question lives, and the argument goes like this. A freeze limits supply while demand keeps climbing, so compute gets pricier and the biggest buyers win the bidding. DevX makes that case and says independent developers, small firms and schools end up at the back of the line. MindStudio, which sells tools to AI builders and so isn’t neutral, lists mid-size SaaS companies built on third-party APIs, university researchers and small model providers who rent GPU time as the people who’d feel it first. It also points out that startups don’t own data centers, they rent from AWS, Azure, Google Cloud, CoreWeave or Lambda, so any price change reaches them through the bill. Both pieces add that companies could just build abroad, taking jobs and tax money with them. MindStudio says the major cloud providers and labs argue the same thing, that a freeze would cost more jobs than it protects. For history there’s Amsterdam’s 18-month pause in 2019 and Singapore’s near three-year pause from 2019 to 2022, and I couldn’t find what either one did to prices. My lean is that prices would probably rise at the margin for new capacity, slowly, and not next quarter, since buildings that already exist keep running. But that’s a prediction from interested parties, not a measurement.
I tried to put a number on it. Thursday night I opened spreadsheet, made one column for my monthly API spend and another for a 10 percent rise in compute costs, and then just stared at it. I don’t know how much of a per-token price is electricity and how much is chips, and without that, any percentage I typed was made up. So I closed the file.
There’s another price in this fight that gets less attention in developer circles, and it’s your electricity bill. Salon’s reporting says these projects are expected to push power prices up a lot while creating few permanent jobs, and the ratepayer point shows up in Common Dreams over and over. Even Comer, who wants America to win the AI race, raises grid capacity as a limit. So whether prices go up depends on which price you’re counting. Supporters of the freeze would say the household one matters to far more people than the API one, and I can’t wave that away in a sentence.
Now jobs, where I part ways with the easy version of the pause argument. Sanders has warned that AI and robotics could eliminate millions of jobs if deployment continues without safeguards, per TechSpot, and Carson’s release says something similar. Fine. But a pause on future models doesn’t un-ship the tools that are already out. Teams that have cut hiring because of them won’t rehire because the next model is delayed. That’s the weak link in the “pause protects jobs” story, at least for software. Supporters would answer that slowing the next jump buys time to write rules on retraining, which is reasonable, though only if the rules show up. Meanwhile the offshoring risk is real. If training and inference move to countries with fewer rules, the work moves with them.
What would this look like for different kinds of developers? For a solo developer or a small SaaS shop built on someone else’s API, the price risk is the whole story, because you can’t negotiate with a cloud provider the way a big company can. For an engineer inside a large company, the freeze matters less directly, since your employer is one of the buyers that wins the bidding, though a pause on new models could freeze whatever tools your team was about to roll out. Researchers and students depend on cloud credits, and MindStudio flags that group specifically. And for people who run models themselves, on rented GPUs or their own machines, I’d guess demand for that route goes up if hosted prices rise. That last one is only my guess, nobody I read said it.
One more guess, and it’s only a guess: a brand new Department of Artificial Intelligence would eventually write rules, and software companies would have compliance work to do. I found nothing that says what those rules would look like.
Past the code editor
Software isn’t the only place this lands.
Construction is the first. Salon quotes someone involved in these projects saying that once a data center is built, most of the people who worked on it have to find new jobs, unlike a power plant where electricians stay on. The National Taxpayers Union takes the opposite view and warns that New York’s pause could cost jobs and investment, while Hochul held a roundtable where community and business leaders backed her decision. So the same projects get called job creators and job-light, depending on who’s talking. RedState’s writer asks what happens to the people already hired to build these things, since their bills come due long before a new regulator writes any rules. It’s a fair question, and I didn’t find a direct answer from the Democrats pushing the freeze.
Then there’s power. Common Dreams treats ratepayer anger as a winning issue for Democrats, and the fight over who pays for grid upgrades looks like it will outlast any one bill.
Creative work is getting squeezed from a different direction. RSL Media launched an AI consent standard backed by Hollywood stars to protect voice and likeness, according to a newsletter I saw. It’s separate from Congress, but it points the same way.
Investors are watching all of it. Crusoe paused one of its campuses, and an analyst named Rajkumar said that should barely dent overall demand for AI chips. I mention it because buildouts already stop and start for business reasons, long before any law gets involved.
Medicine and drug research is the area I’d worry about next, but that’s my guess. I found no source on how a pause would hit it. Gizmodo also mentions another Sanders proposal that would pay Americans $1,000 a year out of the wealthiest AI companies’ money, and I didn’t dig into that one.
Will any of it pass
Not soon. The AP says the superintelligence bill faces long odds in a Republican-controlled Congress, Data Center Dynamics called the federal moratorium highly unlikely to pass, and Trump can veto anything that does move. Casar called the election result an early tremor of a much bigger AI earthquake, per RedState, which also says Democrats are still divided on details. The real action is at the state and town level, where a moratorium needs a county board and not sixty Senate votes.
For developers the state fights may matter sooner than anything in Washington. If a state like New York or one of the ten others pauses hyperscale builds, the cloud regions you rely on don’t change overnight, but the next expansion might land somewhere else, or not at all. That’s my reading, not something I saw spelled out, and I found nothing specific from the cloud providers on it.
Two things are still open as I write this. One is the bill number. The other is a hearing. In early August Casar led a group of House Democrats asking for one where tech CEOs would testify under oath about recent AI hacks, per Gizmodo, and I couldn’t find that a date was ever set.
What I’m changing on my side
I’m not changing plans because of a bill that probably won’t pass. I’m changing them because compute prices just became a political topic, and I’d rather not be surprised.
Put every model call behind one function so swapping providers takes a day, not a rewrite. And write down tokens used per feature each month, with the price per million tokens and the date you checked it, because when somebody says the bill went up, you’ll want to know by how much. I’m also watching two things, whether anyone ever writes down what counts as advanced AI, and whether the 50-megawatt number turns out to be real.
I added that last column Thursday night. The first date in it is November 5.