West Texas dust has a way of announcing itself before you even see the land. It coats your boots, sticks to the back of your throat, and turns the sky a dull, bruised orange by evening. Out here, near Abilene, there’s not much romance to the scenery. Flat. Harsh. Unforgiving. And yet, this is where one of the most expensive bets in modern technology is taking shape.
OpenAI CEO Sam Altman calls it Stargate — a sprawling constellation of AI data centers rising from iron-tinged soil, backed by an unlikely alliance of Oracle, Nvidia, SoftBank, and others. On a typical morning, more than 6,000 vehicles stream onto the site. That’s more people building this single campus than OpenAI employs worldwide.
This is what the AI boom looks like when you strip away the demos and valuation decks. Steel. Concrete. Debt. And a power bill big enough to make utilities sweat.
Stargate and the price of intelligence
Altman stood on-site last fall, squinting into the sun, and summed it up plainly. “This is what it takes to deliver AI,” he said. Unlike previous tech revolutions, intelligence at scale isn’t just software. It’s physical.
Each Stargate campus is expected to cost roughly $50 billion. All in, OpenAI’s plans add up to nearly $850 billion — close to half of the $2 trillion global AI infrastructure surge that HSBC now projects over the coming years.
The Abilene facility already has one data center live, another nearing completion. CFO Sarah Friar told CNBC the site could eventually exceed one gigawatt of capacity, enough electricity to power roughly 750,000 homes — about the combined footprint of Seattle and San Francisco.
The timeline is relentless. Compute breaking ground today is meant for 2026. Nvidia’s next-generation accelerators — the Vera Rubin chips — are already spoken for. Beyond that, the planning stretches into 2029.
Altman didn’t sugarcoat the constraint. “We would be way bigger now if we had way more capacity.”
The new geography of AI power
Altman isn’t alone in betting the country.
Across northeast Louisiana, Mark Zuckerberg is turning former soybean fields into Hyperion, Meta’s four-million-square-foot AI campus. When complete, it’s expected to draw more electricity than New Orleans.
Near West Memphis, Arkansas, Alphabet’s Google is erecting what state officials have called the largest private investment in Arkansas history. In South Memphis, Elon Musk converted a shuttered Electrolux factory into Colossus, a supercomputer built in just 122 days. Colossus 2 is already underway, with Musk aiming for one million GPUs and buying an entire power plant to keep the lights on.
Up north, Microsoft is pouring more than $7 billion into a Wisconsin facility it calls the world’s most powerful AI data center. Amazon, meanwhile, has quietly transformed 1,200 acres of Indiana farmland into Project Rainier, an $11 billion training ground built to serve Anthropic.
AWS CEO Matt Garman put it bluntly: cornfields are becoming compute clusters almost overnight.
This isn’t infrastructure as we’ve known it. It’s industrialized belief — the idea that intelligence itself can be manufactured, and that whoever builds the biggest factory wins.
How big is the spending, really?
The numbers are starting to blur into abstraction.
Here’s a snapshot of what hyperscalers are committing:
| Metric | 2025 Estimate | 2026 Projection |
|---|---|---|
| Hyperscaler capex | $443B | $602B |
| Share tied to AI | ~75% | ~80% |
| New debt issuance (YTD) | $121B | Accelerating |
| Potential future borrowing | — | Up to $1.5T |
Source: CreditSights, Bank of America, Morgan Stanley
Meta alone tapped the bond market for $30 billion. Alphabet raised $25 billion. Oracle pulled off an $18 billion deal and now ranks among the largest non-financial debt issuers in the U.S., according to filings tracked by the SEC (https://www.sec.gov).
Wall Street sees more coming. JPMorgan and Morgan Stanley analysts believe AI infrastructure could drive $1.5 trillion in additional tech borrowing. UBS forecasts $900 billion in new issuance in 2026 alone.
For credit investors, it’s unsettling. Citi’s Daniel Sorid warned clients that this kind of transformation demands capital at a scale that makes even seasoned lenders uneasy.
When debt markets start whispering
You can see the nerves showing up in derivatives.
Credit-default swaps tied to Oracle have widened to multi-year highs. A liquid CDS market for Meta — something that barely existed before — sprang to life as investors rushed to hedge. Barclays and Morgan Stanley have advised clients to buy protection.
There’s history here. In the dot-com era, telecom companies borrowed heavily to lay fiber. Demand eventually arrived, but not before restructurings and equity wipeouts littered the path.
The question hovering over AI is familiar: are companies building for demand that’s real, or demand they hope will materialize?
OpenAI’s web of partnerships
OpenAI sits at the center of this storm.
In just a few months, it announced partnerships totaling $1.4 trillion in headline commitments. Nvidia took an equity stake tied to future chip supply. AMD followed with a deal involving Instinct GPUs. Broadcom agreed to co-design custom chips. Amazon Web Services signed OpenAI as a cloud customer, loosening Microsoft’s once-exclusive grip.
Critics call it circular. Nvidia finances demand for Nvidia chips. Oracle builds data centers for customers tied back to Oracle. Revenue, capacity, and capital loop through the same small circle.
Even Nvidia has cautioned investors there’s “no assurance” some of these agreements will finalize as expected, per its public filings (https://investor.nvidia.com).
Oracle’s Clay Magouyrk sees it differently. From his view on the ground, demand is broad, contracted, and growing. “I don’t worry about a bubble,” he said. “I see committed demand.”
Power, not money, is the choke point
Ironically, money may not be the binding constraint.
Power is.
Data centers need always-on electricity, something renewables alone can’t yet guarantee. That’s why companies are looking at gas, nuclear, and hybrid solutions. Utilities and regulators suddenly find themselves holding the keys.
OpenAI has lobbied Washington to expand CHIPS Act tax credits to cover AI data centers (https://www.commerce.gov/chips). The idea of government backstops briefly surfaced — and just as quickly sparked backlash. Altman publicly insisted OpenAI doesn’t want guarantees.
Still, policy matters. Permitting, transmission lines, and substations move slower than venture capital. The Energy Information Administration has warned that data center demand could materially reshape regional grids this decade (https://www.eia.gov).
Where the payoff has to show up
Training models grabs headlines. Inference pays the bills.
Every chatbot response, every automated workflow, every AI agent running in the background adds recurring compute demand. Unlike training runs, inference never ends.
That’s why executives increasingly compare AI to a utility. Always on. Always scaling. Always consuming power.
Anthropic’s Daniela Amodei says even insiders have been surprised. Every year, they expect the scaling curve to flatten. Every year, it doesn’t. Anthropic’s revenue has reportedly grown tenfold year-over-year for three straight years, pushing its valuation into the hundreds of billions.
The reckoning everyone sees coming
Dario Amodei believes we’re nearing a point where AI systems rival top human experts across disciplines. He also worries about what that means for entry-level jobs in law, consulting, and finance.
That fear — and promise — is what’s driving the spending binge.
Skeptics warn the outcome could look familiar: overcapacity, tightening credit, forced sales. Optimists see electrification-level transformation.
Sam Altman falls firmly in the latter camp. People will overinvest, he admits. Others will underinvest and regret it. Long term, he’s convinced the value will dwarf the losses.
Out in West Texas, the debate feels abstract. Trucks keep rolling. Transformers hum. Dust hangs in the air. And across the heartland, the factories of artificial intelligence are rising — powered by debt, conviction, and an unshakable belief that intelligence itself is the ultimate scarce resource.
FAQs
What is OpenAI’s Stargate project?
Stargate is a network of massive AI data centers OpenAI is developing with partners like Oracle, Nvidia, and SoftBank to support large-scale model training and inference.
Why are AI data centers being built in rural areas?
Cheap land, easier permitting, proximity to power infrastructure, and supportive local governments make rural regions attractive.
How much power do these AI centers consume?
Single campuses can exceed one gigawatt — enough to power hundreds of thousands of homes.
Is the AI infrastructure boom a bubble?
Opinions are split. Some see overbuilding funded by debt; others believe demand will grow into the capacity.
What’s the biggest constraint on AI expansion?
Power availability, not capital, is increasingly seen as the limiting factor.
