Stargate is OpenAI’s gargantuan AI infrastructure program, a whopping 500 billion dollar plan with Oracle and SoftBank to build roughly 10 gigawatts of hyperscale data center capacity for training and running its models.
I walked into “AI Infrastructure: Building Stargate & Beyond” hoping SXSW might, at minimum, stage a grown-up conversation about what this AI hyperscaling infrastructures are doing to power grids, water, and local communities. What I got instead was a one‑hour corporate promo reel: three OpenAI infrastructure execs plus a friendly tech-transactions lawyer moderator, no critic in sight, no contrarian interviewer, no third party… just a smooth choreography of talking points about how they’re heroically building the compute factories of the future.
I’ve been going to SXSW for over two decades, and this was the worst session I’ve ever sat through… by a long shot. The sheer entitlement on stage was suffocating: executives floating miles above reality, utterly disconnected from the places they’re about to chew up, with no meaningful ethical or value grounding in sight. It’s the only session so far that has left me genuinely furious, not just sceptical or disappointed but viscerally angry at the contempt they showed for limits, for communities, for the basics of “should we” rather than “can we.” Honestly, the only image that fits is medieval: they should be marched through the streets in metaphorical tar and feathers while the crowd chants “shame, shame, shame”; because that’s the level of public accountability this kind of arrogance deserves.
The cast: Kaylen Bushell, “Compute Delivery Lead” at OpenAI, the person who makes these megasites physically happen. Matthew Castle, now Head of Infrastructure Business Development at OpenAI, the guy who’s “basically seen and bought every GPU that OpenAI has purchased” since launch. Peter Hoeschele, VP Strategy and Operations; Infrastructure, who proudly says he founded their industrial compute group, the Stargate machine room; and Justin Haan, partner at Morrison Foerster, playing moderator and deal lawyer, the person who negotiates the contracts that turn land, water, power, and tax breaks into data centers.
No ERCOT (Electric Reliability Council of Texas)representative. No governor’s office, no water district, no environmental advocate. No union organizer, o one from a community that will live next to these things and watch their water tables and power bills bend around OpenAI’s appetite. Just them. Just “we,” “our AI,” “our community plans,” as if they already own the future by default in an astonishing display of Machiavellic tech utopia.
“The world needs all this compute”: the religion of scale
At one point, Hoeschele says, more or less with a straight face: “I don’t think OpenAI needs all this compute, I think the world needs all this compute… I fundamentally believe more computation is good for the world.” He cites the “bitter lesson” in AI: scale is everything, throw more compute at the problem, get better results. That’s the whole ethical framework: because their scaling curves go up, the rest of us are morally obligated to step aside.
They brag about one‑gigawatt data centers, one facility that’s about 1.1% of the entire U.S. grid’s peak capacity. This is not a cute “server shed,” it’s a regio-changing energy regime. The International Energy Agency estimates that global data center electricity use will roughly double by 2030, reaching nearly 945 TWh per year, basically another Japan bolted onto the grid just to keep servers humming and models “thinking.” They don’t present this with any warning; they present it like a startup pitch deck. In this panel’s logic, there is no off‑ramp. More data, more training, more “test‑time compute,” more users, bigger flywheel, repeat. They talk about seeing a “stunning” feedback loop: as they throw more compute at the system, the models get better, people use them more, the revenue goes up, therefore build more compute. It’s a powerhungry addiction dressed up as a vision.
Texas as a “welcoming” resource colony
And of course, they’re doing this in Texas. According to them, Texas is “one of the best places to build right now” because the state is “very welcoming” to data centers, with no red tape, ERCOT is accommodating, there’s “lots of open space,” and there’s more available power than elsewhere. They say this like it’s a charming hospitality note, not an indictment of how easy it is to externalize costs onto Texans.
Let’s translate. “Welcoming” means: a deregulated grid that already failed spectacularly in 2021, a political class obsessed with attracting corporate megaprojects, and a patchwork of water planning that doesn’t have a coherent strategy for the kind of demand AI is bringing. Texas already hosts hundreds of data centers-over 460 in 2025, with many more in the pipeline- and new analysis suggests that data centers could use up to 2.7% of the state’s total water by 2030, about as much as 1.3 million households.
Meanwhile, OpenAI’s Stargate complex and its sibling sites are being sited in rural zones with limited political leverage. They talk about “hundreds of acres” like it’s some awe-inspiring techno-landscape, not farmland and fragile ecosystems being permanently converted into heat engines.
Cooling an AI furnace in a drying state
You know what they absolutely do not center? The physics: they talk about “chillers,” “yards” full of electrical gear, and tweaking inlet water temperatures for hotter chips. They admit the buildings are mind‑bogglingly large and the cooling is finicky, but they never truly name the absurdity: we’re putting massive heat-spewing, water‑dependent factories in a state whose summers are getting longer and hotter, where drought is a recurring fact of life.
A single data center can easily gobble hundreds of thousands of gallons of water per day for cooling alone; larger or multi‑building campuses can hit several million gallons daily. That’s on the order of a town of tens of thousands of people. When you add in the water used to cool the power plants that feed these centers, U.S. data centers in 2023 drove an estimated 211 billion gallons of water use, well over a gallon of water per kWh, before even counting on-site consumption.
Texas watchdogs and researchers are already warning that the AI‑driven data center boom could significantly strain regional water supplies, particularly in West Texas and other arid areas chosen for “cheap land” and “available power.” That’s the backdrop. On stage, the story is: we’re tweaking inlet temperatures and “innovating on cooling.” They never ask the obvious question: should we even be doing this here?
PR fig leaves over systemic harm
The panel’s rhetorical trick is to throw out soft-focus human anecdotes to distract from the planetary ones. There’s the cute story about how their model helped someone diagnose a sick dog in Australia. Lots of “we’re helping people be more productive,” “artists and musicians in the audience,” and “community plans” for every site where they’ll hand out some philanthropic crumbs.
What you don’t hear: anything about AI‑mediated suicides and self-harm, about people nudged by chatbots in their most vulnerable moments. Nothing about how much of this compute goes into adtech, surveillance, and automating away tedious, low‑paid human jobs, not just curing cancer. No mention that independent analyses show companies are consistently under‑reporting the true water and energy costs of these facilities, especially once you factor in upstream power generation.
They keep saying “we” as if “we the people” asked for gigawatt‑scale compute factories in dry regions so that a handful of companies could race for AGI bragging rights. “We” did not. They did.
No brakes, no map, but full throttle
What really rankles me is the total absence of humility or restraint. They describe this as a “historic reinvestment” in American infrastructure, driven by AI demand signals, as though the only question is how fast to lay more concrete and string more transmission lines. The possibility that we might want (global, regional, local) democratic control over whether this build‑out happens at all-or how much, or where-is treated as unthinkable.
Globally, data center electricity demand is on track to double in just a few years, with AI a major driver. Texas planners already expect data centers, crypto mining, and heavy industry to nearly double peak power demand by 2030, and AI infrastructure is a huge part of that. Yet the panel shrugs at the idea that their sites would pull down 1% of U.S. grid capacity each, as if that’s just the natural order of things.
And what about coordination? Nothing. No serious mention of national caps, regional planning, global agreements on AI infrastructure, or even genuine community veto power. Instead, we’re getting scattered mega‑projects dropped wherever the political resistance is weakest. The same mentality is playing out in orbit, with tens of thousands of satellites already launched and projections of a million (!) more to be launched by xAI, turning the night sky into an LED mesh and making astronomy harder…all in the name of “connectivity” and “innovation.” It’s the same pattern: colonize the commons first, ask questions never.
Greed before good
Sitting through that session, what struck me most wasn’t just the content; it was the posture. The anthropomorphic buzz about “our models thinking,” the casual bravado about “putting the pedal to the metal,” the complete absence of grief or even discomfort about what this scale of infrastructure actually costs in water, energy, land, and human attention. The arrogance of assuming that a theoretical AGI payoff in some hazy future justifies ripping up the present. OpenAI, xAI, Google, Anthropic and the rest are engaged in a full‑tilt arms race to build ever‑larger data centers, soak ever‑more power from already stressed grids, and drink ever‑more water from aquifers and rivers, all on the assumption that “it will all be worth it in the end.” Worth it for whom? Decided by whom?
Because it sure as hell wasn’t decided by the people in the communities where these gigawatt furnaces are landing. It wasn’t decided in transparent public processes that weighed water security, climate goals, biodiversity, mental health, or the right to see the night sky. It was decided in boardrooms and law firm conference rooms, the same kind where panelists like these spend their days, and then packaged for SXSW as inevitable progress.
So here’s the question I walked out with, the one their panel never dared to ask: why are we letting a small cluster of megarich tech bros and their corporate shells unilaterally decide what is “good for us, the people”, commandeering land, water, power, and even our sky in the name of an uncertain AGI future, with almost no real oversight, no global coordination, no democratic consent, and almost no regard for nature, ethics, or the very real possibility that this is not a path to salvation, but a carefully branded recipe for disaster? Argh.
José Andrés’: the antidote
Luckily, after this display of cold, media‑trained greed, José Andrés’ induction into the SXSW Hall of Fame felt like the perfect antidote. José Andrés is the founder of World Central Kitchen, the disaster‑relief nonprofit that rushes into hurricanes, earthquakes, wars, and refugee crises to cook hot meals on the ground with local communities.
What a man, what a courage: he offered a raw, heartfelt plea about a world where cities are under fire, drones and bombs tear through civilians, nurses and children become collateral damage, and yet we somehow find the political will to spend 2.5 trillion dollars a year on war while claiming we “can’t afford” to end hunger. He reminded us that we could eradicate global hunger for around 80 billion dollars annually, less than 2% of the rich world’s defense budget, if we chose to, and insisted that we must speak louder, even scream if we have to, against leaders who stoke our worst demons instead of our best.
His call was disarmingly simple and utterly devastating in the shadow of Stargate: stop pouring genius and capital into ever more sophisticated ways to build weapons and fortify walls, and start pouring it into feeding people and proving, with action not slogans, that we actually care.
Not all heroes wear capes…
