Party’s Over
Another cruel reminder of how taken for granted and fragile the AI supply chain really is
The views in this article do not represent views of any past, present, or future employer
We are living in a complicated world right now. I am not going to open this with a joke about birthday balloons because the AP already did that this week, so I will spare you that one, but the helium situation is getting dangerous and I am not sure if people in the technology investment world have connected it to how serious it is.
I am not here to repeat what the news already covered but you know what happened in Qatar. Drones, missiles, chaos. Ras Laffan, roughly 30% of the world’s helium, declared force majeure. The helium spot prices doubled and QatarEnergy says the damage will take years to repair, not weeks, and that timeline changes who gets hit and how badly. I just want to write now about what comes after and who gets the worst part.
Helium is used across every semiconductor fab on Earth: cooling wafers, detecting microscopic leaks in vacuum systems, purging the optical path inside EUV lithography scanners. Every chip on this planet, from the most advanced AI processor to the controller in your car that reminds you to lock the window. Worst part? Semiconductors are not even alone in this fight because every MRI machine in every hospital or medical center needs liquid helium for its superconducting magnets. When supply tightens, AI buildout and cancer detection compete for the same finite gas, and somebody has to decide who gets it. I do not want to be the one making these allocation decisions.
This gas is not like oil, you cannot pump more somewhere else and solve the problem. Once you release helium it rises through the atmosphere and leaves the planet permanently. What we extract is the accumulation of billions of years of radioactive decay trapped in geological formations, and you cannot synthesize it or produce it any other way.
We are building a $600 billion AI infrastructure on top of this.
I have had helium on my chokepoints list for a couple of months now. This week it lit up.
Where are we?
The Strait of Hormuz is effectively blockaded. The ships are anchored outside, which means that even undamaged Qatar capacity cannot export. The fact that Hormuz is closed stops almost 30% of world helium regardless of what happens at any facility. The NVIDIA chips for your future datacenter are being sieged right now.
Three companies control the vast majority of global helium distribution. Linde (NASDAQ: LIN) handles roughly 25%, the largest share on Earth. Air Liquide (Euronext: AI) sources approximately 50% of its helium from Ras Laffan and tell me if timing is not a bitch because a few weeks ago at the end of January, Air Liquide signed a brand new long-term supply agreement with QatarEnergy. Less than two months later the first drone hit the facility. There is also Air Products (NYSE: APD) that operates QatarEnergy’s Helium 2 plant directly, roughly 25% of global supply.
It is all offline, and so will be everything downstream of it if things do not change. Right now, in this exact moment you are reading this, there are cryogenic containers that cannot be moved to their destination, with a tight timeline until the helium leaves Earth forever. I guess physics does not get a table in political negotiations.
Now follow it into the fabs and this is where I want to maybe give a glance to your broker app, because helium is not just one input, it is woven through every critical process at every step.
Lam Research (NASDAQ: LRCX) makes the plasma etch tools where helium does its most dramatic work. Plasma hits the wafer of your chips and generates extreme heat, and the helium fills the gaps conducting heat away six times faster than nitrogen, which is the only gas anyone has ever seriously proposed as an alternative. So this gas is the one stopping thermal damage on advanced AI chips because this etching happens hundreds of times per wafer where each lot at a leading-edge node is worth hundreds of thousands of dollars.
Applied Materials (NASDAQ: AMAT) needs helium across deposition and etch, that is the most obvious part, but also for something less visible that I just understood this week: the gas delivery systems and valve assemblies feeding every process chamber. Helium leak testing verifies every connection point of the valves. The only thing smaller than helium to do this is hydrogen, and guess what is a key characteristic of hydrogen? It is flammable. 1 + 1.
ASML (NASDAQ: ASML), your Dutch AI investment, the flagship of the Eindhoven engineering (recently hinted by a massive layoff thanks to McKinsey consultants btw). Their EUV scanners, each over $200 million btw, use helium to cool the CO₂ drive laser running at over 20 kilowatts. Helium purges the optical path because EUV photons get absorbed by virtually every other gas, meaning without helium the light does not reach the wafer at all, so no NVIDIA, AMD, Intel chips. No Texas Instruments, no Tesla, no nothing. Sounds scary right? Well it is.
Tokyo Electron (TYO: 8035) has similar dependencies across its etch and coating platforms.
At the end of the chain: TSMC (NYSE: TSM) burns roughly 500,000 cubic feet of helium per fab per year making the logic inside every GPU. Samsung (KRX: 005930), freshly qualified for Vera Rubin HBM4, needs helium from etching through chemical mechanical polishing where helium-assisted thermal control keeps planarization within tolerance, and of course I would not let my favorite company on the planet out, SK Hynix (KRX: 000660), the dominant HBM supplier, the company I have been writing about for instance here or here, imported almost 65% of its helium from Qatar in 2025.
So: missile in Qatar, delayed wafer at SK Hynix, missing HBM stack on an NVIDIA GPU, server that does not ship, AI model that does not train on time, and AI assistant that cannot do inference. Do your own numbers, how tied is your daily workflow to this?
Six weeks. That is the buffer before allocation tightening according to Fitch. Kornbluth Helium Consulting says best-case restart is six weeks and right now that looks highly unlikely. DRAM and HBM prices nearly doubled Q1 2026. No fab has announced curtailments.
Not yet.
Who did the homework
When the cuts come, Samsung and SK Hynix will prioritize because they make 70% of the world’s DRAM and 80% of HBM. HBM for NVIDIA stays, the margins are enormous. Commodity DRAM for your laptop gets pushed and the chip in your car gets pushed further.
But have you thought about what it does to the cost of intelligence? Every AI server has HBM. HBM gets more expensive, servers cost more, the $650 billion hyperscaler capex number buys less, and the cost per token goes up. I have been thinking about token economics for weeks and now there is a new variable in the model, a noble gas that probably was never supposed to be in anyone’s financial projections.
One more layer because this is worth thinking about: if fabs prioritize leading-edge wafers like Blackwells, Rubin or Groq, everything below gets rationed. Automotive. Industrial controllers. Medical devices. The chips that are invisible until they are missing. You lived through this in 2021 with a particular flu that happened, maybe you remember it.
Korea’s Ministry of Trade elevated its helium response to vice-minister level because they seem to be the most affected by the Qatar situation, but there was someone that unlike Korea did its homework: Japan.
Iwatani Corporation (TYO: 8088) controls roughly 50% of domestic helium distribution. They doubled inventory in recent years, they have been running dual sourcing from the U.S. and Qatar. Stock up more than 10% since the crisis began.
Japan’s METI pre-positioned 20 billion yen across five helium assurance plans involving Iwatani, Air Water (TYO: 4088) through subsidiary Nippon Helium, Nippon Sanso Holdings (TYO: 4091), Japan Air Liquide, and JFE/Tokyo Gas Chemical. Over 70% of Japanese fabs already run recycling at 80-90% recovery rates. Koike Sanso Kogyo (TYO: 6137) builds the equipment with 99%+ capture and according to Japanese industry reporting, Nippon Sanso is renaming its helium division to “Global Helium Strategy Division” in April. Maybe nothing.
Two countries. Both import 100% of their helium. Korea has no visible backup plan and the other had been preparing for years.
I will do a futurology exercise here. If you want to know where the canary will sing first: check ASML utilization in Q2 earnings calls. If things do not improve, fabs will keep the highest-margin wafers running, so leading-edge for Apple and NVIDIA will remain and everything else gets pushed. If utilization dips, now you know why.
Who will fix this mess?
I sometimes think we are living in a bad simulation and the one running it is unwell. Every previous helium crisis had a safety net. The United States Federal Helium Reserve: a geological formation near Amarillo with decades of accumulated buffer. In June 2024 the government sold the whole thing to Messer, a privately held German industrial gas company, for $460 million. Privatized, before the worst helium crisis in history. Satire genre should be renamed simply news. But this does not mean that helium disappears, just a totally different dynamic and pricing power.
There is no strategic helium reserve anywhere in the world anymore. Oil? Yes. Grain? Yes. The gas without which no advanced semiconductor can be manufactured? Sold for less than a quarter of what NVIDIA paid for Groq. If you are not questioning right now why you invested in AI and not in staples, start from the beginning of this piece.
New supply is coming but geology does not care about socioeconomic timelines.
Renergen (JSE/ASX) in South Africa is producing from deposits with concentrations up to 12%, which is rare in a way that matters. Pulsar Helium (TSXV: PLSR) hit 15% in Minnesota. The biggest U.S. project, Blue Spruce/Dry Piney at $737 million, just got approved. First production: two years away and the crisis is measured in weeks.
But not everything is lost, and there is a part of this story that gives me hope and a part that does not.
The part where I stay positive: recycling works. Evonik (Frankfurt: EVK) developed SEPURAN Noble, a membrane capturing up to 90% of helium from fab exhaust, over a thousand installations globally. They manufacture these membranes in Schörfling, Austria, a few hours from where I am writing this, and I did not know that until I started pulling on this thread. Evonik partnered with Linde to build the first commercial membrane-PSA helium plant in Saskatchewan, 95% recovery from feed gas at just 1-2% concentration. And their membranes are the ones Air Water uses in Japanese fabs, the same Air Water in METI’s assurance plan. Austrian-made membranes, installed in Japanese fabs, funded by Japanese government R&D policies, years before any of this happened. That is what preparation looks like.
Chart Industries (NYSE: GTLS) builds the cryogenic equipment alongside those membranes. And there is a company in Korea called TEMC (티이엠씨) (KOSDAQ: 425040) that you will probably not find on any English report or article. They purify helium from neon byproduct, supply Samsung and SK Hynix directly, and are expanding to Japan and Singapore. The market figured it out in Korean before anyone wrote about it in western media. If you are only reading Western sources you missed the signal, which is exactly why I built chipchain and I wrote Information Arbitrage, a tool that finds these kinds of hidden things in Chinese, Korean and Japanese alongside English ones. The semiconductor supply chain does not operate in English, and neither do the signals that tell you where it is breaking.
Now where I am less optimistic: Even with recycling, leak detection helium is unrecoverable because the whole point is releasing it to find the leak. Think about it, a 95% recovery rate still means 5% permanent loss on volumes that compound every day, so recycling slows the bleeding but cannot stop it, moreover the worst part is that we are not just maintaining existing capacity, we are building dozens of new fabs across the U.S., Europe, Japan, and India under the CHIPS Act, the European Chips Act, and Japanese semiconductor subsidies. Every single one will need hundreds of millions of cubic feet of helium over its lifetime. Fabs are coming online faster than helium plants.
The $600 billion AI buildout assumes the supply chain underneath it is solved. It is a web of invisible dependencies and geological constraints that no amount of capital can accelerate, and we just watched a $500 drone prove how fragile it is. A gas that makes balloons float is what stands between the most important technology buildout in human history and a catastrophic cascade, because it took billions of years to fill those reservoirs and it takes one drone to shut them down.
AI scaling is no longer limited by compute design but invisible, non-scalable resources. The helium case is one of chokepoints on the investmap and I wish it were the last.
Which one is next?
This is not financial advice. It is a supply chain traced from a crater in Qatar to a missing server in a data center, with stops in an Austrian membrane factory and a Korean company you need Hangul to google. If you take financial advice from me, I have a helium balloon to sell you. Do your own research. I don't pitch stocks.
Sources
Iran war halts Qatar helium output, threatening global tech supply chains
Korea chipmakers have helium stockpiles 'for six months'
One-third of world's Helium supply gone: How it affects chips to AI chatbots
Messer Completes Acquisition of Federal Helium System from BLM
Middle East Turmoil Disrupts Chip Supply Chain
Why Iwatani Is Up 12.7% After Qatar Helium Disruptions Test Its Supply Resilience
Thanks for the fantastic article! Do you think there will be lags on IREN’s data centers buildouts?
Thanks for the excellent article! Do you have any view on Helix Exploration company?