从石油到带宽：伊朗战争刚刚关闭了互联网的另一个咽喉要道

Key Takeaways

• Both of the world’s internet chokepoints are closed simultaneously for the first time in telecommunications history: Multiple submarine cable systems transit the Persian Gulf and Strait of Hormuz, including FALCON, Gulf Bridge International, and the planned 2Africa Pearls extension. Over a dozen more pass through the Red Sea, carrying the vast majority of Europe-Asia data capacity. Both corridors are now effectively no-go zones.
• AWS data centers have been hit by military drones — a global first: On March 1-2, Iranian drones directly struck two Amazon Web Services facilities in the UAE and damaged a third in Bahrain, marking the first confirmed kinetic military attack on a major hyperscale cloud provider in history. AWS advised customers to migrate workloads out of the Middle East entirely.
• Meta’s 45,000-kilometer undersea cable is stranded: Alcatel Submarine Networks declared force majeure on the 2Africa Pearls extension. The cable-laying ship Ile de Batz is docked off Saudi Arabia, unable to complete connections that were scheduled to go live in 2026.
• Qatar’s helium — essential for chip fabrication — is offline: Qatar supplies 30% of the world’s helium. Ras Laffan is one of the only facilities that produces 6N-grade helium (99.9999% purity) at commercial scale, the grade semiconductor fabs require. It was hit by drones. Spot helium prices have doubled. Samsung and SK Hynix have activated conservation protocols.
• Taiwan has 11 days of natural gas reserves: TSMC manufactures 90% of the world’s most advanced semiconductors. Taiwan imports 97% of its energy. Roughly a third of its LNG comes from Qatar. The Strait of Hormuz is not just an oil chokepoint. It is the chokepoint for the chokepoint of the global AI supply chain.

The Chokepoint You Didn’t Know About

Here is the mental model most people carry: the Strait of Hormuz is where the oil flows. Close it, and you get an energy crisis. That mental model is correct. It is also incomplete.

Multiple submarine cable systems transit the Persian Gulf, connecting the Gulf Cooperation Council states to the global network through the Strait of Hormuz and the Gulf of Oman. TeleGeography identifies four active systems passing through the Strait itself — FALCON, Gulf Bridge International, AAE-1, and Tata-TGN Gulf — with the 2Africa Pearls extension under construction when the war began. But the Persian Gulf cables are only half the story. Over a dozen more systems run through the Red Sea — including SEA-ME-WE 3, SEA-ME-WE 4, SEA-ME-WE 5, FLAG Europe-Asia, and the Europe India Gateway — carrying over 90% of all Europe-to-Asia data capacity.

Both chokepoints are now closed.

Doug Madory, director of internet analysis at Kentik, the firm that monitors global routing infrastructure, put it plainly: “Closing both choke points simultaneously would be a globally disruptive event. I’m not aware of that ever happening.”

It is happening now. And the consequences extend far beyond latency.

The First Kinetic Attack on a Hyperscaler

On March 1 and 2, 2026, Iranian IRGC drones directly struck two Amazon Web Services data center facilities in the UAE’s ME-CENTRAL-1 region, while a third facility in Bahrain’s ME-SOUTH-1 region was damaged by a nearby drone explosion. The IRGC explicitly claimed responsibility, citing the facilities’ role in “supporting the enemy’s military and intelligence activities.”

This is the first confirmed kinetic military attack on a major hyperscale cloud provider’s infrastructure in history.

The strikes caused structural damage, disrupted power delivery systems, and triggered fire suppression that caused additional water damage to server equipment. AWS issued guidance advising customers to back up data, migrate workloads to other regions, and redirect traffic away from both Bahrain and the UAE.

The downstream disruptions were immediate. Abu Dhabi Commercial Bank, Emirates NBD, First Abu Dhabi Bank, payments platforms Hubpay and Alaan, Snowflake’s data cloud, and Careem’s ride-hailing service all experienced outages. Banking, payments, delivery apps, and enterprise software across the region went dark in sequence, as each service discovered that “multi-region redundancy” meant redundancy across two regions that were both under fire.

Sam Winter-Levy of the Carnegie Endowment for International Peace noted the vulnerability: “If you knock out some of the chillers you can take them fully offline.” Chris McGuire, formerly of the National Security Council, went further, suggesting that the Middle East data center model may now require missile defense systems on data center campuses.

A theoretical scenario became a concrete precedent. Data centers are now confirmed targets in modern armed conflicts. The question is no longer whether a hyperscaler facility can be hit. It is who builds the next one in a war zone.

The 45,000-Kilometer Cable That Can’t Connect

Meta’s 2Africa submarine cable system was designed to be transformative: 45,000 kilometers of fiber optic cable reaching 3 billion people across Africa, Europe, and Asia. The Pearls extension — the segment that would connect the cable through the Persian Gulf to Oman, UAE, Qatar, Saudi Arabia, Bahrain, Kuwait, Iraq, Pakistan, and India — was scheduled to go live in 2026.

Alcatel Submarine Networks, the contractor responsible for laying the cable, declared force majeure. Its cable-laying ship, the French-flagged Ile de Batz, is now stranded off Dammam, Saudi Arabia, unable to complete the job. Much of the cable has already been laid on the Persian Gulf seabed, but it still needs to be connected to onshore landing stations before any segment of the Pearls route can enter service.

The timeline for completion is now undefined. Cable-laying ships cannot operate in an active war zone. Insurance underwriters — the same ones whose premium hikes effectively closed the Strait to oil tankers — will not cover commercial cable operations in the Persian Gulf or Gulf of Oman. e-Marine, the UAE-based firm that serves as the maintenance authority for Gulf cable repairs, operates five vessels. Only one is currently inside the Gulf. The others are in the Red Sea and Indian Ocean, where they are also unable to operate freely.

The precedent from the 2024 Red Sea cable cuts is not encouraging. When three cables — SEACOM/TGN-EA, EIG, and AAE-1 — were severed in the Red Sea in February 2024, likely by the dragging anchor of a Houthi-struck cargo ship, the AAE-1 repair alone took nearly five months. That was a single chokepoint, with relatively limited military activity. Both chokepoints are now closed, with an active war in progress. Repair timelines are not weeks. They are quarters, possibly years.

The +200 Millisecond Tax

When submarine cables through Hormuz and the Red Sea go down, internet traffic does not stop. It reroutes. The question is through what, and at what cost.

The rerouting paths are primarily trans-Pacific: traffic that would normally travel the short route from Asia through the Gulf and Red Sea to Europe instead circumnavigates through undersea cables crossing the Pacific Ocean, then traverses North America, then crosses the Atlantic. Round-trip latency on the direct India-to-Europe route, typically 130-140 milliseconds, balloons to over 250 milliseconds on the Pacific reroute as traffic circumnavigates the globe.

For a consumer streaming video, 100-plus additional milliseconds of latency is an annoyance. For a high-frequency trading desk, it is the difference between profit and loss. For a distributed AI training run pulling data from nodes across multiple continents, it is a throughput collapse. For the emerging class of real-time AI applications — autonomous vehicles coordinating through cloud inference, remote surgical systems, financial AI agents executing multi-step transactions — 200 milliseconds might as well be an eternity.

The bandwidth is the deeper problem. The February 2024 Red Sea cable cuts alone disrupted 25% of traffic between Asia, Europe, and the Middle East, and that was three cables in one chokepoint. The current simultaneous closure of both chokepoints affects cable systems carrying the vast majority of Europe-Asia data capacity. The surviving trans-Pacific and overland routes do not have the spare capacity to absorb that volume without degradation.

Gulf states are now racing to build overland data cables to Europe as a war backup. The Arctic route project Polar Connect — connecting Europe, North America, and East Asia at 4,000 meters depth — is being accelerated, with the EU allocating funding through the Connecting Europe Facility. But these are multi-year infrastructure projects. They solve the 2030 problem. They do not solve the March 2026 problem.

The $650 Billion Buildout Meets the Drone

The timing of this conflict could not be worse for the global AI infrastructure buildout.

Hyperscaler capital expenditures are projected to rise from roughly $380 billion in 2025 to approximately$650 billion in 2026, a 71% year-over-year increase, as Amazon, Microsoft, Alphabet, and Meta race to build the compute capacity that the AI boom demands. A significant share of that investment was targeted at the Middle East.

The numbers were staggering. Trump’s May 2025 regional tour generated $2.2 trillion in investment pledges. <!-- Source: CNBC: How Iran war could impact hyperscalers' Middle East AI buildout --> Stargate UAE — the$30 billion, 5-gigawatt AI campus in Abu Dhabi backed by OpenAI, G42, NVIDIA, Oracle, Cisco, and SoftBank — had 5,000 workers on site and steelworks weighing 1.5 times the Eiffel Tower, with its first 200-megawatt phase due in Q3 2026. Amazon committed $5 billion for an AI hub in Riyadh. Saudi Arabia’s PIF-backed venture HUMAIN purchased 18,000 NVIDIA GB300 chips to build 500 megawatts of data center capacity, with its CEO stating the goal was to “build in 2026 the capacity equivalent to what Saudi has built in the last 20 years, in one year.”

The US had approved the sale of computing power equivalent to 70,000 advanced AI chips split between the UAE and Saudi Arabia, with the UAE separately licensed to purchase up to 500,000 Blackwell GPUs annually.

Then Iranian drones hit the data centers.

Kristian Alexander of the Rabdan Security and Defence Institute captured the shift: “A theoretical scenario has become a concrete precedent.” Data centers are now demonstrated military targets. The insurance calculus, the geopolitical risk premium, and the basic question of whether you build a $30 billion AI campus in a region where drones can reach it — all of these are recalculated overnight.

The hyperscalers are unlikely to abandon existing builds. But the next wave of capacity decisions — the ones being made in boardrooms right now — will tilt toward Northern Europe, India, and Southeast Asia. India is already positioning itself as a major data center hub alternative, with over $160 billion in capacity either underway or announced. The Gulf’s pitch — cheap energy, strategic location, sovereign wealth fund capital — collides with a new reality: cheap energy does not matter if a Shahed drone can shut your cooling system off.

The Helium Clock

The AI supply chain crisis does not stop at cables and data centers. It reaches into the molecular inputs of semiconductor fabrication.

Qatar supplies approximately 30% of the world’s helium. Ras Laffan, the country’s massive LNG and helium export hub, is one of only a handful of facilities on Earth that can consistently produce 6N-grade helium — 99.9999% purity — at the commercial scale that semiconductor fabs require. Helium at this grade is essential for cooling wafers during the etching process in chip fabrication. There is no viable substitute.

On March 2, Iranian drones struck infrastructure at Ras Laffan. QatarGas halted all production of LNG and associated products the same day, then declared force majeure on March 4. Follow-up Iranian strikes on March 26-27 caused what QatarGas described as “extensive” damage that will take years to repair, cutting annual helium exports by 14%.

The downstream effects are cascading through the chip supply chain. South Korea, home to Samsung and SK Hynix, imports roughly 65% of its helium from Qatar. Both companies have activated helium conservation protocols, reducing non-critical usage and prioritizing their highest-value production lines. Seoul has flagged 14 semiconductor supply chain materials for war-related vulnerability monitoring.

Spot helium prices have doubled since the crisis began. Approximately 200 specialized helium transport containers — each worth roughly $1 million — are stuck in the Middle East with 35-to-48-day storage windows before the helium escapes through diffusion. Phil Kornbluth, a helium production specialist, estimated that even in a best-case scenario, recovery would take “six weeks or something like that,” though he added this was “highly unlikely” given the extent of the current damage.

The helium crisis creates a direct supply chain link between the Iran war and the AI boom. Every advanced chip in a data center, every GPU in an AI training cluster, every processor in a smartphone, required helium during its fabrication. The war is not just disrupting the cables that carry AI traffic. It is disrupting the production of the silicon that makes AI possible.

The 11-Day Clock in Taiwan

The supply chain vulnerability converges at the most critical node in the global semiconductor ecosystem: Taiwan.

Taiwan imports 97% of its energy. The Middle East supplies roughly 37% of the fuel powering its electric grid. Approximately 28-34% of Taiwan’s LNG imports come from Qatar specifically.

Taiwan has 11 days of natural gas reserves.

Eleven days. Not months. Not quarters. Eleven days of strategic LNG reserves stand between the world’s semiconductor supply and a power crisis on the island that fabricates 90% of the most advanced chips on Earth.

TSMC consumed 27.5 billion kilowatt-hours of electricity in 2024, accounting for roughly 9-10% of Taiwan’s total electricity consumption. Its energy profile is 93% purchased electricity and 6.9% natural gas. If Qatari LNG shipments remain disrupted and the 11-day buffer is not bridged by alternative suppliers, power curbs could interrupt production at precisely the time TSMC is supplying the highest-value AI chips to every major hyperscaler on the planet.

TSMC has stated that supply disruptions are “not expected to significantly impact current operations.” This is a carefully hedged sentence. It says “current.” It does not say “sustained.” Wood Mackenzie’s base case assumes disruptions lasting approximately two months — mid-March to mid-May — with Qatari production gradually ramping back by end of May 2026. If that timeline slips, the hedge becomes a liability.

The Strait of Hormuz is not just an oil chokepoint. It is the chokepoint for the chokepoint of the global AI supply chain. Close Hormuz, and you threaten the LNG that powers the fabs that make the chips that run the data centers that train the models. The dependency chain is five links long, and the first link just broke.

The Steelman: Why the Internet Won’t “Go Down”

The honest counterargument is that the internet was designed for exactly this kind of disruption. And it is not wrong.

The ARPANET architecture that underpins the modern internet was explicitly built to survive the destruction of nodes and links. Traffic reroutes. Packets find alternate paths. AWS’s guidance to migrate workloads to other regions is itself evidence that the system’s redundancy mechanisms work. International transit traffic has successfully rerouted through Pacific cables with 100-200 milliseconds of additional latency but negligible packet loss.

The internet will not “go down.” Consumer email will still work. Video calls will still connect. The apocalyptic framing that the war has “severed the world’s digital arteries” overstates the immediate impact on end users.

But the steelman misses the distinction between “the internet works” and “the AI infrastructure buildout proceeds on schedule.” These are different claims. The internet is resilient. The AI supply chain is not.

An AI training run that requires low-latency, high-bandwidth connections between distributed GPU clusters across continents cannot tolerate a 200-millisecond latency penalty. A semiconductor fab that requires 6N-grade helium cannot substitute a different noble gas. A $30 billion data center campus cannot relocate after construction has begun. A cable-laying ship cannot operate in a war zone.

The internet is a distributed network built for resilience. The AI supply chain is a concentrated, sequential, just-in-time system built for efficiency. The war is stress-testing the latter, not the former. And the latter is failing.

The Convergence

Step back and look at what the Iran war has done to the technology supply chain in 29 days:

It closed both submarine cable chokepoints simultaneously. It demonstrated that hyperscale data centers are viable military targets. It stranded the largest undersea cable project in history. It took 30% of the world’s helium supply offline. It doubled helium spot prices. It put Taiwan’s 11-day LNG reserves on a countdown. It forced the rerouting of the majority of Europe-Asia data traffic through longer, slower, more congested paths. It triggered an insurance crisis that prevents cable repair ships from entering the war zone. And it injected a geopolitical risk premium into every future decision about where to build AI infrastructure.

Each of these is a separate crisis. Together, they reveal a structural dependency that the technology industry has spent years ignoring: the digital economy runs on physical infrastructure, and physical infrastructure occupies geography, and geography is subject to war.

The technology industry’s implicit assumption has been that its supply chains are somehow exempt from geopolitics. That assumption just collided with Iranian drones at three different GPS coordinates, and the debris is still being catalogued.

The Strait of Hormuz carries 20% of the world’s oil. The cables running through it and the Red Sea carry over 90% of Europe-Asia data capacity. The barrels got all the headlines. The bandwidth is the story that matters next.