The Pin: How the Strait of Hormuz Could Burst the AI Bubble
The war in Iran, the closure of Hormuz, and the connection between natural gas powering your AI tools.
In my previous piece, The Subprime GenAI Crisis, I laid out the financial fragility of the AI industry. The losses, the unsustainable burn rates, the circular funding. What I didn’t cover was the physical vulnerability. The thing that actually keeps the lights on.
This article connects two stories that the media is covering separately but almost nobody is linking together. The first: the US-Israeli war in Iran and the closure of the Strait of Hormuz. The second: the AI industry’s near-total dependence on natural gas to power data centres. When you draw a line between them, the picture is not pretty.
If you are a business leader, board member, or risk professional who has spent the last 18 months embedding AI into your operations because everyone told you to, this matters. A lot.
The Strait
Let’s start with the geography.
The Strait of Hormuz is a narrow channel between Iran and Oman. At its narrowest, it’s 21 nautical miles wide. Around 20% of the world’s oil and a similar share of its liquefied natural gas (LNG) flows through it every year.
On 28 February 2026, the US and Israel launched coordinated airstrikes on Iran under Operation Epic Fury. Iran retaliated with missile and drone attacks across the Gulf. Within hours, Iran’s Revolutionary Guard Corps declared the Strait closed, warning that any vessel attempting to pass would be set “ablaze.”
By early March, tanker traffic through the Strait had dropped to zero. Protection and indemnity insurance was pulled. Major shipping companies including Maersk suspended all crossings. Over 150 ships sat anchored outside the strait with nowhere to go.
The result was immediate. Crude oil jumped from around $67 per barrel to nearly $100 in under a week. The US national average for petrol rose 14% in seven days. European natural gas futures surged over 40% on the first Monday of the conflict.
Daniel Yergin, vice chair of S&P Global and one of the world’s foremost energy historians, called it a “nightmare scenario” and warned that the world is looking at the biggest disruption in oil production in history. Nobel-winning economist Paul Krugman wrote that the continued closure of the Strait represents a shock to world oil supplies bigger than the oil crises of the 1970s.
Why It Won’t End Quickly
As Ed Zitron lays out in his excellent newsletter piece on the situation, there are essentially four ways this ends: Iran reopens the strait voluntarily, the security situation improves enough for ships to pass, the Iranian government is overthrown, or both sides reach a deal.
None of those looks imminent.
Trump has demanded unconditional surrender. Iran has chosen Khamenei’s son, reportedly a hardliner, as the new Supreme Leader. Iran’s Shahed drones are cheap, mass-produced, and effective enough that even Ukraine, with years of experience, still cannot fully counter them. Robin Brooks of the Brookings Institution put it bluntly on why this matters: all Iran needs is to sneak through a couple of drones to blow up one ship and the situation escalates from serious to catastrophic.
Iran are prepared for this conflict. Some academics have warned that the nation has prepared for a US attack for over two decades. Surrendering to any US demands is unlikely.
Krugman also flags a critical non-linearity: a two-week closure is far more than twice as damaging as a one-week closure. As producers shut down wells, restarting them could take weeks or months, regardless of when a ceasefire is reached. Iraq has already cut output by 60%. Kuwait and the UAE have followed. Qatar shut down its Ras Laffan LNG facility entirely after Iranian drone strikes hit its infrastructure.
That last point is the one that matters most for this article.
The Fuel Behind AI
Here is what few are connecting.
The AI data centre boom runs on natural gas. Not in theory. Not as a backup. As the primary fuel source, right now, today.
A February 2026 report from Cleanview found that nearly 75% of planned on-site power equipment for data centres is natural gas. They identified 46 data centres building their own power generation, with a combined capacity of 56 gigawatts. Despite press releases touting renewables and nuclear, the equipment actually being installed is, in Cleanview’s words, “almost entirely gas-powered.”
The reason is simple. The grid cannot keep up. Getting connected to a local utility can take up to five years. Nuclear is 7 to 12 years out. Solar and wind cannot provide the firm, uninterrupted baseload power AI workloads demand. Natural gas can be deployed now. So that is what they are building.
The examples are everywhere:
Stargate Abilene (OpenAI/Oracle): Powered by over 1 GW of natural gas turbines, largely off-grid.
Stargate Shackelford County (OpenAI/Oracle): Entirely off-grid, running on 700 MW of natural gas generators via Voltagrid.
xAI Colossus (Elon Musk, Memphis): Powered by natural gas generators, with a private gas plant planned in Mississippi.
Bloom Energy fuel cells: Natural gas-powered fuel cells deployed for Oracle, AEP, and Brookfield data centres. Stock up over 400% in the past year.
Permian Basin builds: Multiple off-grid natural gas data centres, including CloudBurst/Energy Transfer (1.2 GW) and the Nvidia-backed Poolside/CoreWeave Horizon campus (2 GW).
The Global Energy Monitor found that proposals for new natural gas facilities in the US tripled in 2025 compared to the year before. The US is now planning over 250 gigawatts of new natural gas energy capacity. A large portion of that is being driven by data centres.
Natural gas is not a bridge fuel for AI. It is the foundation.
The Collision
So what happens when the primary fuel source for the AI boom is caught in the crosshairs of a global energy crisis? Perhaps we’re looking at the perfect storm. Perhaps this is the final act that will burst the inevitable bubble.
The direct impact on US natural gas prices has been relatively contained so far. On the first Monday of the conflict, US prices rose about 5%, compared to the 40-45% surge in Europe. That’s because the US is a net exporter of gas, not an importer. But as the Financial Times reports, US LNG producers are now racing to redirect cargoes to take advantage of skyrocketing European and Asian prices, where LNG spot prices have nearly doubled.
This is the mechanism that will hit American data centres. When US gas producers can sell to Europe at 50% higher prices, they will. Domestic supply gets tighter. Domestic prices rise. Scott Shelton of TP ICAP told the Financial Times
Whatever we can put on a boat we are going to send.
Saul Kavonic of MST Marquee was even clearer:
Nothing can make up for the loss of Qatari LNG.
Rapidan Energy, one of the most respected energy advisory firms, warned CNBC that LNG could be hit harder than crude oil in the long run, because LNG production is more concentrated and harder to ramp up. Qatar’s Ras Laffan, which produces roughly a fifth of global LNG, is a single facility. It is now offline. QatarEnergy has delayed expansion plans until 2027.
Meanwhile, Counterpoint Research told CNBC that electricity accounts for roughly half of a data centre’s operating expenses. If energy costs spike and stay elevated, data centre operators may be forced to cut capital spending and reduce demand for AI infrastructure.
Brad Gastwirth of Circular Technology put it plainly:
For the technology sector the immediate risk is not a direct interruption in semiconductor production but a broader inflationary impact through energy and transportation costs.
The maths is not complicated. If the fuel behind AI gets significantly more expensive, everything downstream gets more expensive. Training runs. Inference. API calls. The subscriptions your company is paying for ChatGPT, Claude, Copilot. All of it.
Investor patience on profitable AI companies must run dry at some point. And conflict is an honest excuse to lean on cash, precious metals, and a sit and wait strategy. How will OpenAI and Anthropic function when investors simply go quiet but energy prices rise?
The Gap in Reporting
This brings us to a broader point about how this information reaches you.
In The Black Swan, Nassim Taleb makes the observation that journalists reporting in lockstep naturally causes opinion to converge and dimensionality to shrink. Everyone ends up using the same items as causes, telling the same story, landing on the same conclusions.
That is exactly what is happening here. Energy journalists are covering the Hormuz crisis. Tech journalists are covering AI data centre power. Few, if any, are drawing a line between the two. The result is that business leaders, risk professionals, and boards are making decisions about AI adoption without understanding that the energy supply powering these tools is under significant, sustained threat.
The media has spent two years telling you to adopt AI or get left behind. That same media is not yet telling you that the cost of running AI is about to get materially more expensive, with no clear timeline for resolution. At what point does the cost to use ChatGPT outweigh the return?
What To Watch
If you are responsible for technology strategy, operational resilience, or enterprise risk, here are the signals to monitor:
US natural gas spot prices. The Henry Hub benchmark is currently around $2.96 per MMBtu. If it pushes past $4 and sustains there, data centre operating costs will feel it.
LNG export volumes from the US. As producers redirect gas to higher-priced markets in Europe and Asia, domestic supply will tighten.
AI service pricing changes. Watch for adjustments to API pricing, subscription tiers, or usage caps from major providers. These will be the first visible signs of cost pressure.
Data centre construction delays. Higher energy costs, higher materials costs, higher labour costs. Stargate is already behind schedule. This will get worse.
Fed interest rate signals. If energy-driven inflation forces rates up, the entire debt structure underpinning AI infrastructure becomes more fragile.
What To Do Now
The advice here is not to rip out your AI tools tomorrow. It is to stress-test your assumptions.
If your organisation has made AI central to its operations, ask the uncomfortable question:
what happens if the cost of those tools doubles? What happens if API rate limits tighten? What happens if providers raise prices to cover their own rising costs, or worse, if some of these companies go under because their already-unsustainable economics get hit by an energy shock they never planned for?
Build optionality. Maintain manual or lower-tech fallbacks for critical workflows. Diversify your provider stack. And above all, stop making long-term strategic commitments based on the assumption that AI will remain cheap and abundant.
The media-driven FOMO that pushed everyone to embed AI into everything did not account for a war in the Middle East closing a 21-mile-wide strait. But the physical world does not care about hype cycles. Oil tankers are sitting still. Gas prices are climbing. The natural gas turbines running AI’s biggest data centres are about to get a lot more expensive to feed.
The bubble was already fragile. This might be the pin.
Ollie Law is Co-Founder and Managing Director of Fixinc, a resilience advisory firm based in Oceania and ASEAN, and Editor-in-Chief of Unbreakable Ventures and the UV podcast. He writes about risk, technology, and business.