Data center power and cooling: judges, voters and grid queues test AI buildout

A federal judge in Mississippi refused on Friday, October 30, to dismiss the NAACP's Clean Air Act lawsuit against xAI, now owned by SpaceX. The suit concerns the gas turbines powering its Colossus data centers near Memphis. CNBC reported the ruling the next day. The judge's opinion, released late Friday, let the case proceed to the merits.

The order matters beyond one site. It tests a diesel-style argument: that backup or temporary turbines are "mobile sources." That argument struggles in court when the machines are bolted down and run for months. The judge said no. That single word carries weight for every operator who has treated on-site generation as a way around a slow interconnection queue. It also matters for every regulator who has watched the practice spread without a clear ruling on whether it is legal.

Because the plaintiffs have statutory standing under the Clean Air Act, because the combustion turbines are not 'mobile sources' under Mississippi's SIP, and because the citizen suit provision does not violate Article II of the Constitution or the private nondelegation doctrine, the plaintiffs' claims will proceed.

CNBC quoted that language, and it is the judge's, not a company statement. The suit was filed in April 2026 in the U.S. District Court for the Northern District of Mississippi. The Southern Environmental Law Center and Earthjustice filed it on behalf of the NAACP. It alleges xAI and subsidiary MZX Tech installed and operated 27 gas turbines in Southaven between August and December 2025 without an air permit. SpaceXAI's June motion to dismiss made three arguments. It said the turbines were mobile, that the NAACP had no right to sue, and that citizen suits of this kind were unconstitutional. All three arguments failed. CNBC notes the same court had already rejected a Department of Justice request to end the case about two weeks earlier. SpaceX did not immediately respond to a request for comment, CNBC said.

What the turbines are accused of emitting

The emissions numbers in the case are the kind that turn local zoning meetings into national stories. Tom's Hardware, citing a CNBC report, said the facility allegedly emits over 5,000 tons of nitrogen oxides and 47 tons of cancer-causing formaldehyde annually. The SELC puts fine particulate matter at 400 tons. The two outlets frame their figures differently. Tom's Hardware describes the totals as annual facility emissions, while the SELC account refers to what the plant could emit. The discrepancy is worth flagging rather than smoothing over.

Either way, the case has become the clearest legal test of the gas-turbine shortcut. AI operators have used that shortcut to bypass slow grid interconnection. Power is the constraint. NVIDIA's own blog, published October 22, says more than $300 billion in equipment is backlogged. It also says over 200 gigawatts of projects are waiting in U.S. interconnection queues. That queue is why companies build their own plants. It is also why courts and regulators are now being asked whether "temporary" generation is temporary.

The political weather has changed

Data centers are no longer a permitting story confined to utility commissions. TechCrunch reported on October 31 from a Brooklyn midterm event hosted by Newsreel. A reporter moderating a panel on data centers asked the room who was pro-AI data center, and the room booed. The same piece cites a Data Center Watch finding. Almost $200 billion in data center projects were delayed, stalled or blocked in the first half of 2026 because of political and community opposition. States including Pennsylvania, Georgia, Texas, New York and Virginia are weighing more guardrails, TechCrunch reported. Hundreds of new data centers are planned in Virginia alone.

That is the backdrop for OpenAI's reported repositioning. TechCrunch reported on November 1 that Axios had said OpenAI is preparing a public relations push after the midterms. The push includes rebranding data centers as "AI compute campuses." OpenAI's chief global affairs officer, Chris Lehane, told Axios the midterms would be "a bit of a movie trailer for what you're going to see in 2028." TechCrunch quotes Lehane calling the data center debate "good and healthy" and saying communities should hold companies accountable. The rebrand is a communications move, not an engineering one. It lands after Amazon and Microsoft said they would stop using NDAs in local negotiations, per TechCrunch.

Boom's pivot shows where the money went

The most striking evidence that power, not flight, is where capital is flowing comes from Boom Supersonic. Ars Technica reported on October 30 that Boom unveiled a factory in Centennial, Colorado on October 22. The plant is dedicated to producing a 42-megawatt natural gas turbine for AI data centers. The same article quotes United Airlines CEO Scott Kirby, one of Boom's first customers. He told the Office Hours: Business Edition podcast on October 28 that the 2029 Overture timeline is not going to happen. "It's not going to happen in '29, it's going to take longer," Kirby said, according to Ars. He had put the odds of commercial service at "50-50" in March.

Boom's Overture factory at Piedmont Triad International Airport in Greensboro, North Carolina, was completed in 2024. It has mostly sat dormant, Ars reported. That raises questions for state leaders who provided more than $50 million in taxpayer funding. The FAA proposed a rule in June 2026 to overturn the ban on commercial supersonic flights over the U.S. The proposal followed an executive order from President Donald Trump. Certification, not regulation, is the bottleneck Kirby describes.

Cooling and the software layer

The cooling side of the equation is being sold as a systems problem. NVIDIA's October 22 blog describes its DSX platform as a way to plan, simulate and run grid and AI factory as one system, from substation to rack. The post names GE Vernova, Hitachi Energy, Schneider Electric with AVEVA and ETAP, and Siemens Energy as partners. Hitachi Energy's battery storage system qualified as DSX Ready. Phaidra said integrating DSX Flex, DSX Max-Q and DSX Exchange delivered about 10% more compute within the same power envelope. Emerald AI said its Conductor platform, integrated into DSX Flex, was modeled to deliver up to 100 gigawatts on the existing U.S. grid. These are vendor claims published on a vendor blog, and should be read as such.

Microsoft's research arm took a different angle on October 30. It described an Industry Evaluation Data Generation Platform that builds synthetic enterprise environments to test Copilot. The platform draws on public sources such as O*NET occupational data, online job postings and SEC filings. The point is not cooling but confidence. The company says real enterprise data cannot be used for evaluation, so it manufactures stand-ins. It is a reminder that the AI stack's hidden costs include test infrastructure, not just megawatts.

Quantum, malware and the long tail

Two other items show how far the power-and-compute story reaches. HPCwire reported on November 2 that researchers led by Arslan Munir at Florida Atlantic University published a study in Discover Quantum Science. The study evaluated quantum machine learning on lung cancer prediction with deliberately limited data. D-Wave's annealing quantum computers, used in a quantum support vector machine framework, reached 94.31% accuracy in simulation. They reached 91.90% on quantum hardware, according to HPCwire. The dataset began with 309 records and was expanded to 540 synthetic samples, not new patients.

On the security side, Tom's Hardware reported on November 1 that Cisco Talos documented a self-deleting Windows infostealer called VOLTA-AUTODOXER. By default it sends stolen data to Anthropic's Claude and asks it to rank the victim's assets by cash-out value. The malware can switch models if one refuses, or skip the AI step entirely, Talos found. It is a small, concrete example of the same pattern. Compute is being pointed at problems that used to be handled by people, and the guardrails are being written after the fact.

The through-line is not that AI data centers are doomed. It is that the constraints have moved. For two years the story was chips. Now it is a judge in Mississippi, a zoning board in Virginia, a queue of 200 gigawatts and a CEO in Chicago saying 2029 is too soon. The buildout will continue. The terms on which it continues are being renegotiated in courtrooms and at ballot boxes, not in press releases about compute campuses.