Iran's Hormuz chokehold is rewriting the economics of AI compute

The Strait of Hormuz is not an AI story. It is a shipping lane, a geopolitical flashpoint, a bottleneck for 20% of global oil supply. But the emerging deal that would hand Iran control over inbound traffic through the strait, reported by Reuters and covered by Fortune, is quietly becoming an AI story. The reason is simple: AI compute runs on energy, and energy runs through Hormuz.

The deal taking shape between Iran and Oman would give Tehran authority over ships entering the waterway, with Iran demanding transit fees of 5% to 7% of cargo value. Oman has floated around 3%. The U.S. insists the strait must remain an international waterway with no tolls, but it is not even part of the talks. A U.S. official told the Washington Post that “any temporary routes will be without any impediments,” while acknowledging that Washington is not at the negotiating table. Iran’s Islamic Revolutionary Guard Corps rejected the U.S. account within a day, saying reopening depends on Iran’s conditions, not on an agreement with Oman.

The concession is already made, one source told Reuters. “The concession has already been made regarding some form of control over Hormuz.” Gulf states, according to the Wall Street Journal, now see a temporary deal recognizing Iran’s authority as their best option. Gregory Brew, an Iran expert at the Eurasia Group, posted on X that “U.S. accepting Iranian control of the strait without explicitly acknowledging that control seems a likely endgame here.”

This is the new normal. And it has consequences for every AI lab, every cloud provider, and every startup that depends on GPUs humming in data centers.

The energy-to-compute chain

AI infrastructure is energy-hungry in ways that were unthinkable five years ago. A single training run for a frontier model can consume as much electricity as a small town uses in a year. The largest data centers being planned in the Gulf, in Saudi Arabia and the UAE, are being built precisely because of cheap local energy and strategic location. Those data centers, and the ones in Europe and Asia that draw on Gulf LNG, now face a supply chain where the chokepoint is controlled by a state that has spent five months attacking energy infrastructure.

Iran has already demonstrated the ability to damage tankers and regional oil infrastructure, giving it a de facto hold on the strait regardless of any formal agreement. The threat is not hypothetical. Saudi data shows Iran launched more than 700 missiles and drones toward the kingdom by late March. The Houthis have renewed attacks from Yemen, injuring 11 civilians in the Najran region this week. Iran has warned Gulf neighbors that any new U.S. attack would trigger retaliation against critical energy infrastructure.

For AI builders, this translates into a simple equation: energy prices spike, compute costs spike, training runs get more expensive, and the gap between the labs that can afford frontier models and everyone else widens. The AI economy is not insulated from geopolitics. It is downstream of it.

The hardware angle is worse

The energy story is bad. The hardware story is worse. AI accelerators, advanced networking gear, and the specialized cooling systems that data centers need do not come from the Gulf. They come from Taiwan, South Korea, and Japan, shipped across the Indian Ocean and through the Strait of Hormuz for facilities in Europe and the Middle East. The same waterway that carries oil carries the physical substrate of AI compute.

Iran’s control over inbound traffic means it can inspect, delay, or tax any vessel carrying the components that AI infrastructure depends on. The 5% to 7% fee Iran is demanding on cargo value would apply to a container of GPUs worth millions of dollars. A single shipment of H100-class accelerators could carry a transit tax in the hundreds of thousands of dollars. That cost gets passed down the chain, from hyperscaler to cloud customer to startup renting GPU hours.

The insurance problem compounds it. Al Jazeera’s Mark Pfeifle, a former White House deputy national security adviser, notes that some war-risk policies terminate coverage when vessels pay Hormuz transit charges. If a vessel refuses to pay Tehran, it risks attack. If it pays, its owner faces problems with its insurer or Washington, since U.S. persons and U.S.-owned companies cannot pay Iran for safe passage without running afoul of sanctions. This is a trap with no clean exit. Captains have to reopen Hormuz at sea, even if diplomats reopen it on paper.

The Gulf is building around it

The Gulf states are not waiting for the U.S. to solve this. They are building pipeline capacity to bypass the strait, which will diminish Iran’s leverage over oil markets over time. But pipelines do not carry GPUs. And the security arrangement taking shape is about defense, not commerce.

On August 7, Saudi Arabia, Turkiye, and Pakistan signed the Mecca Joint Defence Agreement in Mecca. An armed attack against one is an attack against all. Egypt is close to joining, having hosted the other three countries’ foreign ministers in Cairo in June. The pact links Saudi Arabia to NATO’s second-largest military and a nuclear-armed Pakistan. A regional official told Pfeifle that Iran will likely test the pact through a proxy, probably the Houthis or Iran-aligned Iraqi militias.

For AI infrastructure in the Gulf, this means one thing: uncertainty. The data centers being built in Saudi Arabia and the UAE are strategic assets, and they are now inside a security arrangement that is untested. A proxy attack on Saudi energy infrastructure could trigger a three-country response, escalating the conflict and further disrupting the supply chains that AI compute depends on. Qatar, which Iran has already attacked and whose energy exports have been disrupted, is weighing whether to join the pact. Joining would give Qatar greater security guarantees but would complicate its diplomatic role as a mediator.

What this means for AI builders

The lesson for anyone building on AI infrastructure is uncomfortable but direct: the compute supply chain is now a geopolitical risk factor, not just a logistics problem. The era when a frontier lab could assume stable energy prices, reliable hardware shipments, and functioning insurance markets is over. Hormuz is the most visible chokepoint, but it is not the only one. The Taiwan Strait, the Suez Canal, and the South China Sea all carry the physical inputs of AI compute.

The response from AI builders should be structural, not reactive. That means diversifying compute across regions that do not share the same chokepoints. It means investing in energy sources that do not depend on Gulf oil and LNG. It means building redundancy into supply chains for hardware, rather than assuming just-in-time delivery will hold. And it means pricing geopolitical risk into the cost of compute, because that risk is now real and measurable.

Some of this is already happening. The push for nuclear-powered data centers, the buildout of compute in less geopolitically exposed regions, and the growing interest in energy-efficient inference all reflect a market adjusting to a world where energy and hardware are not guaranteed. But the adjustment is slow, and the shock is here now.

The deal being negotiated between Iran and Oman will not resolve this. It will formalize a reality that has existed for months: Iran controls the strait, and the world is paying for it. The AI industry, which has spent the last two years assuming that compute is a commodity that can be scaled at will, is about to learn that compute is a function of energy, and energy is a function of politics.

The dinner at the Waldorf Astoria in Doha proceeded without missile alerts this week. That is the new normal. But the ships that carry the components of AI compute still have to pass through waters that Iran controls, and the fees, the insurance headaches, and the risk of attack are not going away. AI builders who ignore this are building on sand, or rather, on a shipping lane that can be closed at any time.