AI-First
What you'll learn:
- 🔑 A strike on Ras Laffan takes 33% of global helium supply offline for five years.
- 🎯 6N helium is irreplaceable for plasma etching, leak detection, and EUV lithography.
- ⚠️ SK Hynix and Samsung imported two-thirds of their helium from Qatar for HBM memory.
- 💡 China is simultaneously securing domestic helium from Guangdong and the Power of Siberia 2 pipeline.
- 🚀 Memory prices stay elevated through mid-2027, with cost per flop and consumer chips rising.
A missile hit Ras Laffan. That sentence, which went almost unnoticed in mainstream media, may be the most important piece of news this year for the AI industry. Because Ras Laffan is the refinery in Qatar that produces one-third of the world's helium. And helium is what makes it possible to manufacture the chips that power all of AI.
We're talking about the helium crisis. Not a party balloon problem. A structural problem that will affect the prices of GPUs, MacBooks, phones, and AI inference costs for years to come.
Helium in fabs: irreplaceable, period
Most people think chips are made with silicon and electricity. That's true, but incomplete. You also need helium, in massive quantities, at 99.9999% purity (6N). Not carnival-grade helium. Ultra-pure helium.
Here's why no substitute exists:
During plasma etching, helium is blown beneath the wafers to control temperature with absolute precision. No other gas conducts heat as well at this scale.
In vacuum chambers, helium is used to detect leaks. Its atomic size is the smallest of all elements, making it detectable before any other gas.
In EUV (extreme ultraviolet lithography) machines, helium maintains a clean and controlled environment.
Helium is among the most significant operational cost items for an advanced fab: between 5,000 and 20,000 cubic meters per month for a 300mm EUV fab. And consumption increases with each generation of chips.
Worse: the most advanced fabs, those producing the HBM memory that powers Nvidia GPUs and Google TPUs, consume the most helium. That's where the shortage hits hardest.
What happened in Qatar: 33% of global supply offline
Ras Laffan is a massive industrial complex in Qatar. It produces liquefied natural gas (LNG) and, as a byproduct of cryogenic distillation, helium. The two are inextricably linked: no LNG, no helium.
The complex was hit by missiles. Qatar Energy confirmed that 14% of its helium capacity is permanently damaged, with reconstruction timelines of 5 years. Five years.
Before this incident, Qatar produced roughly 2.4 billion standard cubic feet per year, or 33% of the 7 billion produced globally. All of that capacity is currently offline.
And that's not all. Qatar had planned to launch Helium 4, which was set to be the largest helium plant ever built with an additional 1.5 billion cubic feet. That project is now in complete limbo.
Helium in a specialized container begins to evaporate after 35 to 48 days. Loaded containers, stuck at sea due to the geopolitical situation, are losing their cargo in real time.
Impact by key player in the manufacturing chain:
Player | Exposure | Detail |
|---|---|---|
SK Hynix (South Korea) | Critical | 2/3 of helium imported from Qatar, #1 HBM memory manufacturer |
Samsung (South Korea) | Critical | Same Qatar dependency, #2 HBM memory manufacturer |
TSMC (Taiwan) | High | Logic chips, depends on LNG for 97% of its energy, 11 days of reserves |
Nvidia / AMD / Google | Indirect | Depend on the fabs above for their GPUs and TPUs |
US data centers | Indirect | End buyers of chips produced in Asia |
LNG: when Asia's energy costs skyrocket
Helium isn't the only problem. Ras Laffan also produced LNG, exported massively to East Asia. That energy is offline for an indefinite period.
TSMC imports 97% of its energy. Taiwan holds 11 days of gas reserves. South Korea imported two-thirds of its helium from Qatar, but also a significant share of its energy.
The effects are already visible: spot LNG prices are climbing, input costs for Asian fabs are rising. These costs will flow through to the cost per chip, then to the cost per flop. That is, into the cost of the AI inference you pay for every month.
Europe is exposed too. Unlike the United States, which has domestic gas production, Europe depends on global markets. The disruption in global LNG supply will push European data center prices higher over a 2 to 3 year horizon.
Geopolitics: why China comes out ahead
While the West takes the hit, China is positioning itself. Two simultaneous moves:
Power of Siberia 2: the Russo-Chinese pipeline (stalled for years over pricing disputes) will likely be revived. China now has an urgent reason to secure a domestic energy supply. If this pipeline materializes, China controls its energy costs while everyone else pays more.
Domestic helium production: China has a plant in Guangdong certified at 6N. It's small (1.2 million cubic meters in a market measured in billions), but it's growing fast, accelerated by the current crisis.
If China secures its helium and energy while Taiwan and South Korea remain exposed, the economics of chip manufacturing shift. China can begin deploying domestic chips (Huawei, CXMT) at lower costs to dominate AI inference by the late 2020s.
This isn't a theory. It's the logic of comparative advantage applied to AI. Whoever controls energy and helium controls the cost of chips. And whoever controls the cost of chips controls AI.
What this means for you, right now
Here's what's going to happen, in order:
Memory prices stay elevated through at least mid-2027. HBM was already sold out. DRAM had already risen 70%. Now, the expected additional production capacity won't arrive.
Cost per flop will rise. Data centers in Asia will pay more for their chips AND their energy. Those costs will be passed on to your AI subscriptions.
Supply lead times are getting longer. If you manage an IT fleet or infrastructure, expect longer timelines on GPU servers.
Laptops and phones will cost more. This isn't limited to hyperscalers. Rising DRAM prices mean more expensive memory in your next MacBook or iPhone.
The advice from the transcript is direct, and I agree: if you need to buy compute, buy it now. Don't wait. With each passing quarter, structural costs will ratchet higher.
The good news: local AI automation tools, like the ones we use with OpenClaw to build AI teams, keep working regardless of these cloud costs. Local agents are a form of hedge against rising inference prices.
My verdict
The AI industry spent 3 years debating valuation bubbles and scaling laws. Those debates are now secondary. We're facing a real physical problem: the material infrastructure of AI depends on a rare gas produced in a handful of places on Earth, and one of those places was just damaged by missiles.
This is not a temporary disruption like the 2019 attacks. It's a restructuring of the global supply chain, measured in years. Fabs will find workarounds, but at higher cost. And that cost will show up in your inference bill, in your next laptop, in the GPU for your next data center.
Buy compute now if you need it. Plan your IT budgets with a structural increase baked in. And pay close attention to what's happening with China and its domestic chip stack. The AI landscape of the 2030s is being shaped, in part, by this helium crisis.