Why the Data Center Fight Has Little to Do With AI

0:00Today on the AI Daily Brief, why the data center debate has less to do with AI than you think. Before that, in the headlines, the White House unveils its plan to vet AI models to a very selected few. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI.

0:24All right, friends, quick announcements before we dive in. First of all, thank you to today's sponsors, KPMG, Blitzy, Robots & Pencils, and HyperAgent. To get an ad-free version of the show, go to patreon.com slash aidailybrief or you can subscribe on Apple Podcasts. And to learn more about sponsoring the show, send us a note at sponsors at aidailybrief.ai. By the way, while you're on aidailybrief.ai, you can find each episode turned into a convenient set of cutdowns. So if there is some particular quote or number or part of the episode that you are super interested in sharing, chances are good that it has a shareable card that might make it a little bit easier.

0:56And every episode gets that sort of treatment. So check it out at aidailybrief.ai. We also have our summer adventure going, which is a choose-your-own-adventure-free, self-directed education program. It's got a full range of skills you can develop from very basic to more advanced. You can check it out again at summeradventure.ai. To be honest, guys, I thought that today's main episode was going to be all about this new White House AI framework that they hauled in all the different companies yesterday to discuss. But it appears that while they do have a new safety testing framework, the rest of us don't get to learn the details.

1:29Representatives from the leading AI companies attended a meeting with the White House Office of the National Cyber Director on Tuesday to discuss the new framework, but there are very strict limits on who gets to know anything about it. In fact, even AI companies that weren't invited to the meeting won't be told about the policy. Now, as far as we know, representatives from OpenAI, Anthropic, and Google attended, but the exact guest list also hasn't been divulged. As for the policy itself, this is the culmination of the voluntary safety testing framework that was proposed in the June executive order.

1:56Companies are invited, quote-unquote, to submit frontier models to the government ahead of release for a maximum of 30 days of safety testing. Only state-of-the-art models that could pose a threat to national security are expected to be covered, although we also don't know exactly how that criteria is determined. It sounds like there is some discussion about the administration and the AI companies failing to reach a common definition which could muddy the waters for future releases. During safety testing, certain trusted partners, quote-unquote, can gain early access to the models, but again, we have no details on who these partners are or how the system will operate.

2:26Specifically, we don't know whether early releases will be limited to U.S. firms and exclude allied governments and foreign companies as we saw with the Mythos rollout. Much of the coverage has focused on how open-weight models would be treated under the policy. Many have been concerned about the administration cracking down on near-frontier open models, either through this safety testing framework or through a standalone policy. Sources have said, though, that open-weight models would be exempt from the safety testing regime. Although there was some disagreement in the press, with the Wall Street Journal publishing that the exemption would only apply to American-made open models, and a few hours later, Bloomberg reporting that Chinese open models would also be exempt from testing.

3:01That could ultimately be a distinction without a difference, as even if Chinese models were subject to safety testing, what exactly are they going to do? It's unclear how Washington could enforce a release ban on companies that are based in China. Prins on Twitter summed up, To the surprise of no one, the US government went with the approach that provides it maximum flexibility to deal with any risks posed by AI that might arise in the future. Now the challenge, of course, is that maximum flexibility for the government comes with minimum knowability for the industry. Former FTC chief technologist Neil Chilston said, A secret process by which the government gets early access to cutting-edge AI?

3:34What could go wrong? There may be good reason to classify the benchmark used to test those models. There is no good reason to hide how the program works. Remember what this program does. It gives the government early access to cutting-edge tools and a potential veto over whether the rest of us ever use them. Doing that in secret is no way for a democracy to govern what may be the most important technology of our lifetimes. Secrecy invites abuse. The rules will shift with each new administration, and they will weight national security over economic growth, even though lasting security depends on a strong economy.

4:04Congress must write any necessary rules in public and in law. Certainly seems like there will be more to this story to come. For now, let's move on to more hacking incidents that have come to light in new third-party testing of OpenAI's models. On Tuesday, OpenAI published the results of evaluations run by the UK AI Security Institute and security firm Irregular. Each firm reported an incident where the agent gained access to the internet and carried out an exploit on a live website. Irregular's incident seemed like human error, with the agent's sandbox mistakenly still having access to the internet.

4:33The agent was tasked with hacking a fictional website within the simulated environment, however after the agent found its way onto the internet, it hacked domain with the same name. Basically, it sounds like Irregular left the gate open with fairly predictable results. The UK AISI disclosed a perhaps more concerning incident. They were running both Mythos 5 and GPT 5.6 sold through what they described as routine cyber evaluations, and said that both models took, quote, sustained, unsanctioned actions directed at real people and organizations. In the most serious case, they wrote, the agent engaged in social engineering, creating fake online identities and using them to pressure the project's maintainer to approve the code.

5:09A human maintainer caught and refused to approve the malicious code. Now, an important asterisk to this is that the incidents occurred under the AISI's standard cyber testing conditions, which include full access to the internet and the removal of all guardrails. They've now engaged with both Anthropic and OpenAI, along with Meter, for independent review. Now to some, this is scary. Tenebris writes, Mythos is not aligned. Models without strong cyber refusals will, in fact, frequently take significant steps to commit crimes in the real world when presented with eval setups. But others said this is just kind of the expected behavior.

5:39Shannon Sands of News Research posted, This is like handing an actor a loaded gun and then complaining that they shot someone during a scene. Oh my god, we told it was an eval and had no guardrails, but actually it was live. It is the fakest form of misalignment I can imagine. Now, while Chinese open models might not be subject to these new policies, the Trump administration does appear to be preparing a ban on Chinese data center components amid espionage concerns. Reuters reports that the Federal Communications Commission is drafting a ban on certain components, including optical transceivers.

6:09These are the devices that convert digital data into light signals for transmission over optical fiber. Now, the networking between data center chips is now entirely fiber optic, and because these transceivers are commodity electronics, much of the supplies come from Chinese suppliers. Sources said that the installation of these components is seen as a national security risk, possibly allowing the injection of malware, data theft, or disruption of service at AI data centers. It is unclear whether the risk has manifested or this is purely a precautionary measure. Still, analysts believe it's a legitimate concern, with AI policy analyst Divyanch Kashuk commenting, transceivers definitely pose a risk.

6:42As the data center build-out scales up, you want to make sure the data center supply chain is secure from the get-go. The ban also comes as the House Select Committee on China prepares to release a report on the Salt Typhoon hacks uncovered in 2024. These hacks targeted telecommunications companies and allowed Chinese spies to access phone calls from high-profile political figures in the U.S. for several years. Sources said that the report will find that Chinese data center hardware was a weak link, allowing the hack to take place. On the other side of the argument are those who say that transceivers as an attack vector doesn't make a ton of sense, and instead claim that this is a protectionist move to boost a few domestic suppliers.

7:15Certainly right now, there are only a couple of US companies that make this component and they appear to be ripping on the news. Lastly today in the headlines, tech earnings continue as SpaceX delivers their first earnings report as a public company. This was the first chance for investors to take a look under the hood and gauge the company's growth trajectory after they raised$86 billion in a record-setting IPO in June. All things considered, the numbers were pretty good. Top-line revenue grew to$7.8 billion for the quarter, up 92 % from this time last year. Of that,$4.2 billion came from Starlink, with satellite internet subscribers doubling to$12 million over the past year to produce 66 % revenue growth.

7:49The AI side of the business brought in$2.6 billion across both Grok subscriptions and datacenter rentals. That's triple the revenue from a year ago before SpaceX AI had signed datacenter deals. Notably, the cursor deal hasn't closed quite yet, so won't appear in earnings until later this year. On the other side of the ledger, SpaceX had$15.8 billion in AI-related capex for the quarter. That represented 86 % of total capex, implying that data centers are now much more capital-intensive than a fully-fledged space program. Overall, SpaceX recorded a$541 million quarterly net loss, excluding capex, down from a billion-dollar loss in the first quarter, but not enough to quiet critiques that SpaceX is a massive cash incinerator.

8:26Still, as with everything SpaceX, the narrative is about where the company is going rather than how expensive it will be to get there. During the call, CFO Brett Johnson reported that the company has a further$6.7 billion in cloud services revenue in the pipeline over the next six months, presumably referring to the Anthropic and Google deals. He also said that the integration of Cursor should take SpaceX AI to a$100 billion run rate by the end of the year. That would be more than a 5x increase from their$18.7 billion in revenue for 2025. Elon Musk added, The$100 billion ARR in December is not a question mark.

8:56That's what we would achieve if we basically did nothing. So I think it may be higher than that. It probably will be higher than that. The other big question is what the shape of SpaceX AI's data center build-out will look like moving forward. Johnson said that CapEx will remain relatively flat through the third and fourth quarters, bringing SpaceX to around$60 billion in data spend for the year. That puts them significantly behind Google and Amazon, who are each committing to$200 billion in spend this year, and starting to creep up on the heels of Oracle, who have forecast$95 billion for this year.

9:23Still, Elon added, we're building AI compute capacity at a scale faster than anyone else we believe, and we're significantly improving our AI models. Overall, this was an earnings beat, but probably not enough to get anyone who's critical on board. The stock was down as much as 45 % from its all-time high as of last week, falling meaningfully below the$135 IPO price. It had been ramping back up in recent days, gaining 10 % to begin the week, but following earnings, the stock rolled over and lost 7 % in after-hours trading. Importantly, this earnings also marks the end of the first round of lockups, giving early investors their first chance to sell into the market and further test the price level.

9:58More to keep an eye on, but for now, that is going to do it for the headlines. Next up, the main episode.

10:07One of the most important AI questions right now isn't who's using AI, it's who's using it well. KPMG and the University of Texas at Austin just analyzed 1.4 million real workplace AI interactions and found something surprising. The highest-impact users aren't better prompt engineers. They treat AI like a reasoning partner. They frame problems, guide thinking, iterate, and push for better answers. And the good news? These behaviors are teachable at scale. If you're trying to move from AI access to real capability, KPMG's research on sophisticated AI collaboration is worth your time. Learn more at kpmg.com slash US slash sophisticated.

10:45That's kpmg.com slash US slash sophisticated. Here's why most legacy modernization projects fail. The AI doing the work can't understand codebases at scale. It sees a small slice of context, examines syntax, and misses years of decisions distributed across the global application ecosystem. Blitzy solves this the way it solves everything. Grounded in your code before any migration begins, Blitzy's agents reverse engineer the entire legacy system into a persistent knowledge graph. Every dependency, every constraint, every piece of tribal knowledge that used to live in one engineer's head. From that understanding, Blitzy autonomously executes language migrations, framework upgrades, and monolith-to-microservices transformations, all validated end-to-end.

11:24One Blitzy customer modernized a$10 million monolithic insurance stack in 16 weeks against a 137-week baseline with coding agents. That's 9x compression. Retire technical debt while accelerating your roadmap. See how at Blitzy.com. That's B-L-I-T-Z-Y dot com. I cover the capability gap between AI potential and AI reality every day on this show. Most companies are still figuring out how to start. Robots and Pencils is already launching and scaling. Agendic and generative AI in production at large enterprises in weeks. AWS Advanced Tier, Pattern Partner more than doubled in a year. And they're hiring.

12:0050 open roles. If you're someone who knows this moment is different, who wants to be inside it, not watching it, this is worth a look. At Robots and Pencils, the best ideas win, and the team is purposefully kept super high quality. This is the kind of place you look back on as the best decision you ever made. Take a look at robotsandpencils.com slash careers. This episode of the AI Daily Brief is brought to you by HyperAgent, where you run fleets of agents your team can manage together. New users get$1 ,000 in inference. Forget local agents and chat workflows waiting on your laptop to be prompted.

12:32HyperAgent deploys always-on agents in the cloud, doing real work across the tools your team already uses. Marketing's agent turns competitor moves into landing pages. Sales's agent enriches leads, drafts emails, and updates the CRM. Ops agent chases the paperwork and tracks the budget. Every agent has access to shared context and follows your rules about scope and approvals. It's time you add agents that feel like teammates. Hire yours at HyperAgent, built by the team at Airtable. Claim your$1 ,000 in inference at hyperagent.com slash AI Daily Brief.

13:05Welcome back to the AI Daily Brief. Today we are talking about the latest in the AI data center debate, and I'm going to make a contention, as presumably you've already seen in the title of this show, that the data center debate has a lot less to do with AI itself than it appears on first glance. It would certainly be going too far to say that it has nothing to do with AI, for some data centers are the easy visual manifestation of a thing they don't like. But I believe that that is only one, frankly, kind of small component of the backlash. What I think it is way more about is a loss of agency, and a feeling that people have that a world that they didn't choose is being imposed upon them.

13:42For all sorts of different reasons right now, people feel unable to control their lives. And this is of course a broader political setting than just AI faces. They feel skeptical of anyone in any position of power, and don't believe that those people in power have their best interests as a priority. And within that, people no longer believe that technology is designed to serve them. More and more, in fact, they think it's just about serving the oligarchs that build it. Data centers then represent both last straw and something tangible to fight. Now, the companies building these data centers don't have much control over the context that local constituencies are operating from, even if in the case of a company like Meta, they might have had a hand in shaping it.

14:18But they do have control over how they interact with it, and they are doing an absolutely abysmal job. This week, we got a great new 6 ,000-word piece from journalist Jasmine Sun and a follow-up interview with Ezra Klein of the New York Times that explores a lot of these themes. However, to situate a little bit of recent news around this for those who haven't been paying attention closely, let's talk a couple of data center stories that have made headlines recently. One of those stories comes from, of all places, Texas. Texas Governor Greg Abbott has thrown the brakes on data center construction with stringent new energy audits.

14:48Governor Abbott has instructed the Public Utility Commission of Texas and the statewide grid operator, ERCOT, to begin verifying and auditing new data center proposals. In his letter to the utilities, he wrote that the measures were necessary to, quote, keep the grid stable and reliable. Texas data centers will now be required to provide information on state and local incentives they've received, disclose how much they will rely on the local grid, submit expected water consumption, and a supply plan. They will also need to prepare a plan for how they will track and respond to community concerns, including things like noise complaints.

15:18Now, it is not clear how much this red tape will slow down approvals and whether the policy is intended as a de facto moratorium. What is clear is that Texas is straining under a wave of new data center permit applications. Texas is the second largest data center market behind Virginia, and the state is experiencing an unprecedented wave of applications. ERCOT currently has 474 gigawatts of new connection requests, with 90 % of those requests coming from data centers. This figure has doubled in the past six months and is now five times the record peak capacity of the grid. In other words, it is, at this point, completely impossible for all of this to get built.

15:53This backlog is in fact around four times the total current installed capacity of the US. Even if ERCOT could quadruple their power supply to service all these projects, bottlenecks and chips, components, and labor would prevent everything from getting built. Now one potential explanation is multiple applications from the same project. Often, a data center developer won't know which application will be approved, so they submit multiples to increase their chances of at least one going ahead. LAN that's ready to build with full approvals also attracts a huge premium, so some developers are simply making applications with the intent to sell the land at a later date.

16:25This has all resulted in a bureaucratic nightmare, where ERCOT has no real means to sift the genuine projects that are likely to go ahead. Dealing with this issue could be part of the reason for Governor Abbott's directive, rather than a desire to sneak in a data center moratorium. Still, the impact is the same, with new applications in Texas halted until audits can be completed. Kevin Hsu from Interconnected Capital makes the obvious point here. Texas is ground zero for AI data centers, poised to overtake Virginia as the number one data center state, so any disruption there is important. New York Times economy reporter Lydia DePilis wrote, as the rest of the country has grown skeptical of data centers, the answer has been just build them in Texas.

17:02Well, today, Greg Abbott paused new applications, saying they threatened to overwhelm the grid. Now, one similar story from last month was New York Governor Kathy Hochul signing that state's first data center moratorium. Around the signing, Hochul said, there are so many applications in primarily small communities across the state of New York, that the localities aren't quite sure how to manage this. Continuing, she said, they've asked me for help and at first I thought this is more of a local decision, but then I realized they don't have the negotiating ability, the clout, the wherewithal to be able to negotiate the best benefits or make sure that these companies are either bringing in their own power or paying a premium into the grid to cover the cost.

17:37A data center moratorium tracker built by Interconnected Capital is currently tracking 219 local moratoriums. Most of these are not state-level, there are 23 state bills that they're following, but it's clear just looking at the map that this is a large and growing area of concern. Now, as an aside, one interesting challenge with data center moratoriums is that they introduce a free-rider problem to the larger economy. The states that don't want to host data centers and are taking pretty strong measures to prevent them being built aren't talking about not using AI on a statewide level. They still want the service provision, they just don't want to support the infrastructure.

18:11Indeed, when Illinois Governor J.B. Pritzker signed their AI bill, he made a big deal over the fact that Illinois, New York, and California are 40 % of the AI market, despite only having 20 % of the population. Basically, we're now getting to the point where some of the largest U.S. markets want other states to deal with the externalities of running AI infrastructure. Now, if you just look on social media, some of the discourse around data centers feels absolutely insane. One person who has emerged as a perhaps surprisingly sensible voice is Taylor Lorenz. Now, Taylor's long-term relationship with the tech industry goes way beyond the scope of this particular episode, but suffice it to say that she's not necessarily getting invited to a lot of Silicon Valley parties right now.

18:50And yet, with increasingly force she is arguing, in short, that there are enough good reasons to be opposed to data centers that some of the more dramatic, hyperbolic, and conspiracy theory-type claims are actively getting in the way of reasonable discourse. Pointing to a post on the Thread service that argued that data centers are for quote military use, the surveillance state, land grabs through eminent domain, and to speed up Armageddon for the white Christian nationalist agenda. Oh, and some folks believe that it's a cover for hoarding our water to supply the billionaire bunkers, a post which by the way had 1.4 thousand likes.

19:21Taylor said, I'm worried vast swaths of our population are experiencing data center psychosis. Someone responded, I work at a data center with a pretty small number of servers that support scientific research. It is vastly different than the AI hyperscale data centers, I had someone on here say they would rather die than do my job. Continuing her thought on a podcast, Taylor said, I went on a podcast recently because I'm in this community group in LA, and they found out that there's a data center in downtown LA. It's not a hyperscaler, but the comments are like, this is why I've been feeling off, or this is why my dog has cancer.

19:53That's not to say that there aren't serious environmental concerns with Colossus in some of these. Construction can disrupt the water table. Okay, but let's have an actual conversation about it. Everybody has lost the plot. There's no nuance. And if you talk to a lot of these people, they can't even explain what a data center is. A lot of the outrage about data centers is coming from people that live in Brooklyn. You're feeding people slop and churn, and you're not moving these issues forward. And we're not talking about sustainable ways to build it. Now, reinforcing the point that Taylor is not some rabid pro data center activist in sheep's clothing, in another tweet she said, absolutely no one wants a new hyperscale AI data center in their backyard.

20:27Even if you don't think a moratorium is the best policy to enact change, literally everyone hates the data centers because they are the most physical manifestation of big tech people are encountering. I do not think we should encourage people's delusions about data centers. That's just going to backfire. A lot of people believe straight-up conspiracy theories, which makes it easier for tech companies to write critics off. There are plenty of legit reasons to not like DCs. And indeed, the politics of data centers are getting weirder and weirder. There is, of course, a dimension of it on the left that views it as one of Trump's great corrupt boondoggles.

20:59With a headline that is a major contender for the juiciest left meat for the progressive crowd, the New York Times recently published a piece called Trump's Vision for AI Dominance Comes with Major Air Pollution. And yet this is not a left-right issue. In fact, Molly Taft recently wrote and wired a piece called How Data Centers Broke American Politics, with arguably one of the most insane subheaders I've ever seen, what the Unabomber, Steve Bannon's tech guy, and Bernie Sanders taught me about the great data center backlash of 2026. And the point here is, of course, that the left and right are coming together to make very weird bedfellows around data centers.