xAI Powers Its Data Center With Jet Engines—Now Emissions Are Challenged

Welcome to today's superpower daily explainer We're cutting straight through the noise to bring you the facts on today's top AI developments I'm thrilled you're here for this November 4th 2026 briefing because we've got a seriously fascinating lineup today touching on physical infrastructure rapid engineering digital rights and enterprise security So here's what's on the docket We're unpacking XAI's jet powered data center OpenAI's rapid hardware prototyping a massive AI fan fiction shutdown and a million dollar LLM heist

Let's get right into it XAI's jet engine power and their new emissions challenge XAI has literally installed eight converted General Electric aircraft turbines to power its Mississippi data center We're talking repurposed equipment that you typically find attached to a Boeing aircraft What's really wild here is the sheer scale These turbines repurposed by a Missouri company called Pro Energy are rated for a maximum generating capacity of up to 50 megawatts each To put that massive number into

perspective Tom's hardware points out that just one of these turbines can generate the electricity needed to power around 15,000 homes It's an aviation derived approach that gives XAI a colossal power source right when data center demand is absolutely surging But we've got to look a bit closer at the numbers here It is absolutely crucial to distinguish that this 50 megawatt figure that represents the equipment's maximum generating capacity It's not the confirmed electrical output currently happening at

the site Basically just because it can generate that much doesn't mean it's constantly running at that absolute limit And you know this distinction becomes super relevant when we look at the regulatory pushback XAI is currently facing Right so the critical limitation here is an active legal challenge from the Southern Environmental Law Center Now XAI does hold state permits for this installation which actually makes this situation different from an earlier lawsuit they had about unpermitted gas generators

But the environmental group is challenging Mississippi's approval anyway They're alleging XAI undercounted projected emissions and lacks guarantees regarding pollutants like formaldehyde carbon monoxide and volatile organic compounds Keep in mind these are currently allegations in a challenge not findings of actual violations The big question to watch moving forward is whether these permitted turbines emissions assurances will actually hold up under legal scrutiny OpenAI's hardware prototype tests and the GPT 6 ASTRA design sprint Two OpenAI employees recently embarked

on a really cool side project to see if the GPT 6 ASTRA model could design a physical hardware prototype meant for talking to Codex their coding agent This perfectly illustrates how AI is stepping out of the chat box and into the physical world The model contributed across the entire hardware stack First appearance ASTRA helped develop the look and form using visual references Second mechanical It designed the physical enclosure and all the internal structures holding it together

Third electronics The model helped pick out the components and lay out the printed circuit board to fit perfectly inside And finally software It actually generated the firmware and UI connecting the physical device to the Codex agent It was a complete end to end design test And this rapid prototyping establishes a totally new for engineering What's so fascinating is the speed of it all The initial AI design sprint took under a week Then a contract manufacturer was

able to produce a physical prototype batch in under a month Also separate from this specific prototype OpenAI actually reports that usage of Codex and chat GPT across their internal hardware teams has grown 23 fold since February measured in tokens So yeah they are aggressively testing these tools to iterate much faster But there is a vital caveat here Can AI models actually account for real world manufacturing limits without human intervention Well right now the answer is no

ASTRA missed production limits manufacturing tolerances and the specific assembly details that really matter to a factory on the floor It took several rounds of human engineering review to revise the design and make it actually manufacturable So while benchmarks like BenchCAD 2.0 might show super high scores for 3D modeling and circuit placement those scores do not measure real world physical manufacturability The human safety net Still entirely necessary Alright let's look at our third story The AI fanfiction

clone shutdown and the fall of AO3 life A fanfiction clone site called AO3 life was just forced to shut down after some massive author backlash over scraped stories To understand the scale of that backlash you just have to look at the volume of complaints The clone site's creator who goes by Ganesh shut the whole operation down on November 2nd after getting more than 2,000 takedown requests in a single day The site had attracted over 126,000 unique

users in 24 hours utilizing 40 virtual machines to scrape existing stories while the main archive of our own site was offline due to a security incident The creator did issue an apology stating it was never my intention to hurt so many authors and I'm truly sorry to you There's no way to show my sorry better than taking this site down Now AO3 life had tried to justify its existence by offering authors 91% of the profits from

artificially narrated audiobooks they were selling But the core lesson here is pretty stark Offering generous compensation is absolutely not the same thing as securing permission The site scraped and monetized works including archive locked stories meant only for logged in without ever asking the writers And honestly here is where the site's logic completely fell apart The most glaring flaw was their takedown process itself To get a stolen story removed AO3 life required authors to verify their identity

by updating their AO3 profiles But remember the entire reason the clone site existed was because the actual archive of our own repository was completely offline So it was literally an impossible task that trapped authors leaving them with absolutely no control over their own writing Finally our last story today million dollar AI heists and uncovering the world of LLM jacking Google's threat intelligence group and Mandiant have detailed these underground routing schemes that are sticking compromised companies with

massive AI bills It's a newly documented phenomenon they're calling LLM jacking And the financial exposure here is staggering In one extreme case investigated by Mandiant a single threat actor generated over 500,000 model calls in under eight hours This saddled a victim company with nearly $1 million in unauthorized corporate charges These underground services like one called Jetknight advertise unlimited access to premium frontier models for just 60 bucks a month to their subscribers And Google assesses that stolen

enterprise credentials are exactly what makes these cheap illicit subscriptions possible in the first place So how does this actually work Well it's a very specific four step playbook First attackers steal cloud or model credentials often from exposed GitHub repositories or service account keys Second and this is the sneaky part they test the access quietly with a single output token just one tiny unit of text which confirms the key works without triggering any volume based security alerts

Third they pool these keys together and resell access via those illicit gateways we mentioned And finally they switch on premium AI models at the victim's expense Astonishingly in some cases the victim companies hadn't even enabled AI services themselves before the attackers arrived and flipped the switch To combat this Google highly recommends a defense in depth approach You've got to tighten credential management and limit each key to strict necessary permissions You also need to monitor for any

unauthorized AI service activations Basically treat a new service turning on as a critical event You should immediately investigate those tiny low volume token requests that attackers use to test the ladders And finally set strict hard spending caps on model consumption to limit your financial exposure if a breach actually does happen Okay let's synthesize this whole briefing with the four essential takeaways you need to remember today First XAI's aviation grade jet turbines give them incredible generating capacity

but they are now battling environmental groups over emissions proving physical infrastructure is never just plug and play Second OpenAI proved an AI model can design a hardware prototype across four disciplines in just a week but it still heavily relies on a human engineering review to actually be manufactured Third the AO3 life shutdown showed us that a promise of future payment definitely does not equal legal permission to scrape someone's work And fourth stolen cloud credentials are now

fueling underground LLM jacking schemes sticking unsuspecting companies with million dollar bills Looking at all four of these stories you kind of see a common thread Technology is moving way faster than the guardrails meant to contain it As AI rapidly accelerates across physical infrastructure mechanical engineering and digital rights how do we enforce boundaries before the damage is already done That's the big question we all need to be asking as we build this new economy Thank you so

much for joining me for this explainer Read more at superpowerdaily com and stay curious