Everyone figured the AI gold rush would chug along forever, datacenters sprouting like mushrooms after rain, fueled by venture bucks and Nvidia chips. Hyperscalers like Microsoft and Google promised returns that justified the trillions in capex. But here’s the wrench: Donald Trump’s push to reopen the Strait of Hormuz amid the Iran war is jacking up energy costs worldwide, and it’s hitting AI’s soft underbelly hardest.
Look, power prices were already a sore spot—those massive GPU farms suck electricity like black holes. Now, with oil imports choking and global south economies rationing fuel, the US can’t hide. Gasoline’s spiking, sure, but datacenter operators? They’re sweating bullets over cashflow.
Why Energy Costs Are AI’s Kryptonite
It’s simple math, really. Training one big model rivals a small city’s power draw. And Sam Altman—OpenAI’s chipper CEO—tried brushing it off back in February, right before his company’s IPO dreams.
“People talk about how much energy it takes to train an AI model – but it also takes a lot of energy to train a human,” he said. “It takes about 20 years of life – and all the food you consume during that time – before you become smart.”
Cute analogy, Sam. But humans don’t rack up $400 billion in capex on $60 billion revenue. That’s the AI industry’s dirty secret, straight from a Quinn Emanuel report: revenues peanuts, spending insane. We’re talking off-balance-sheet vehicles, asset-backed securities—echoes of 2008 subprime madness.
The Bank of England nailed it in their latest risk survey. Pre-war, investors were already dumping AI stocks over debt worries. Now?
“The conflict could increase these concerns, particularly given the energy-intensive nature of the supply chain for key components and the operation of datacentres.”
Spot on. Higher power bills don’t just nibble margins; they explode them.
And get this—my unique take, after two decades chasing Valley hype: this isn’t just a hiccup. It’s the spark for an AI shakeout worse than dot-com 2.0. Remember 2001? Pets.com folded when ad dollars dried. Here, startups like Anthropic or xAI, burning cash on uneconomic models, face extinction if power hits $0.20/kWh sustained. Only the hyperscalers—Amazon, Meta—with moats and cash piles—survive. The rest? Roadkill.
How Bad Could This Get for Datacenters?
Picture it: Indonesia’s work-from-home Fridays, Egypt’s curfews, Philippines energy emergency. Oil importers scramble. The US, oil-rich, still feels the pinch—analysts say prices stick high for months, Strait or no.
Datacenters? They’re power hogs. A single hyperscale facility gulps 100 megawatts, enough for 80,000 homes. Multiply by hundreds planned. WTO’s Robert Staiger warned me last month: prolonged highs “crimp” AI investment. And 70% of last year’s US investment growth? AI gear.
Companies like CoreWeave borrow big from opaque private credit outfits. Regulators—Bank of England included—fret over the shadows. Special purpose vehicles own datacenters, pool debts, slice ‘em up for pensions. Sounds diversified? Nah, it’s concentrated risk, just like CDOs pre-crash. One default ripples.
But wait—supply chains fracture too. Chips from Taiwan, rare earths tangled in geopolitics. Trump’s war amplifies it all.
Short-term, cashflow crunches force delays. Long-term? VCs pull back, IPOs flop. Altman’s “human training” quip? It’ll haunt him when blackouts hit training runs.
Is AI’s Business Model Even Viable?
Strip the buzz. AI revenues? $60bn last year. Capex? $400bn. That’s a 6.6x burn rate. Sustainable on cheap power, maybe. Not when Iran’s messing with Hormuz.
Ed Zitron’s been screaming this—real builds lag announcements. Promises of exaflop clusters? Vaporware in dusty fields. Lenders hide in private markets, opacity reigns. Bank of England flags it: fragilities everywhere.
WTO outlook underscores: war weighs on growth, spikes inflation, tightens conditions. AI, energy-thirsty and debt-heavy, buckles first.
Here’s the cynicism: who’s really winning? Nvidia sells picks-and-shovels, sure. But hyperscalers? Betting farm on unproven returns. Consumers get what—fancier chatbots? While bills soar.
My prediction: by Q3, we’ll see datacenter project cancellations, startup funding winter. The boom creaks, then snaps.
Old-timers like me saw telecom bubble burst on overbuilds. Same script. Energy was the missing villain then; it’s center stage now.
🧬 Related Insights
- Read more: Citrini’s 2028 Nightmare: When AI Ghosts Haunt the Economy
- Read more: GPT-5’s Router: OpenAI’s Sneaky Path to Ads in Your Chat
Frequently Asked Questions
What does the Iran war mean for AI energy costs?
It spikes global oil and power prices via Strait of Hormuz risks, hammering datacenter ops for months.
Will higher energy prices kill the AI boom?
Not overnight, but they’ll expose debt fragility, forcing cuts and shaking out weak players.
How much power do AI datacenters really use?
One big facility: 100MW+. The sector’s exploding demand rivals small nations amid rising costs.
