半導体サプライチェーンの隠れた脆弱性は「シリコン」ではない。それは「ヘリウム」だ。

台湾の TSMC 工場で現在、エッチング光源（EUV 装置）が単位数ナノメートルの世界でトランジスタを蝋刻み、加工されています。この技術を実現するために必要な光源の冷却には、地球上の要素でこれまでに酷い冷却効果を示し安定動作を保証しうる唯一の元素であるヘリウムが用いられています。そして、世界中のヘリウム供給量は枯渇しつつあります。 決してゆっくりと、決して理論的なレベルではありません。今まさに。 世界のヘリウム供給は、少なくとも 3 つの方向から同時に圧縮され、あらゆるスマートフォンやサーバー、自動車、AI アクセラレーターの基盤となる半導体産業は、まさにその被害範囲内にあります。 これは表紙のニュースになるようなサプライチェーンの物語ではありません。劇的な輸出禁止や単一の出来事があるわけではありません。これは、複数の地域にまたがり、何十年かの政策の無視の結果として蓄積され続けてきた、ゆっくりとした複合的な失敗に過ぎません。あなたが何かを作成する場合、それがチップに依存しているなら（つまり、すべてです）、何が起きているのかを理解する必要があります。 多くの人はヘリウムを派手な気球や気高い声と結び付けますが、これは世界のヘリウム消費の約 8% に過ぎません。残りの 92% は、多くのエンジニアが思いつきもしないものへ向けられます：MRI 装置、光ファイバの製造、ロケット推進、そして、特に重要なのは、半導体製造です。 現代のファブでは、ヘリウムは何が別の元素にも替えることができない役割を果たしています。シリコンウェハ加工時に必要な不活性大気が、原子レベルでの酸化や汚染を防ぐためです。3nm ノードにおける単一の欠陥は、ウェハ全体を損なうことができます。ヘリウムの化学的惰性はその場を埋め替えないようにしています。 しかし、より大きな物語は EUV 光蝋刻み技術にあります。TSMC、サムスン、インテルが最先端生産のために依存している ASML の装置は、極紫外光源を冷却するためにヘリウムを使用しています。これらの装置は、それぞれ 1 億 5,000 万ドルを超えて高く、安定した高純度のヘリウム供給なしには動作できません。代わりはありません。アルゴンや窒素に代えることはできません。ヘリウムの熱伝導率は空気よりも 6 倍高いです。他の何もその仕事をしません。 14 年以上のシステム建設と運用の歴史を踏まえると、私が常に目睹するのは、誰の監視もしない最も危険な単一故障点は、誰も監視していないというパターンです。半導体製造におけるヘリウムはまさにそのように見えない依存関係です。 現在の危機には、業界ですでに名前があります：ヘリウム不足 4.0 フィル・カーンブルースは、世界的ヘリウム市場の専門家であり、世界的ヘリウム市場の専門家であり、世界的ヘリウム市場の専門家であり、世界的ヘリウム市場の専門家です。 ここで何が間違いましたか。 ロシアのアムール工場は届きませんでした。ロシア東部のアムールガスパrocessing プラントにある巨大な新しいヘリウム生産プラントが、世界全体のヘリウム需要の約 3 つに供されるはずでした。それは供給量の締め付けに対する業界の回答でした。その後、ウクライナ戦争が起き、制裁が続き、プラントは火災を含む複数の技術的遅れを被りました。供給を緩和するはずだったヘリウムが、単純に到着しませんでした。 米国の戦略的貯備は売却されました。テキサス州アマリロ近くの連邦ヘリウム貯備庫は、冷戦期に戦略的在庫として設立されました。何十年にもわたって、これは世界の供給変動に対する緩衝材として機能しました。米国土地管理局は、このプロセスは 2023 年に完了する予定でしたが、その緩衝材は消去されました。 カタールは依然として集中リスクです。カタールは世界最大のヘリウム生産国の 1 つです。2017 年に国が外交的封鎖に襲われた際、世界のヘリウム供給の約 30% 相当のプラントが一時的に停止せざるを得なくなり、ガスワールドは報告しました。封鎖は最終的に解決されましたが、その事実は残酷な真実を暴きました：ヘリウム供給は、地政学的に不安定な地域の一部に集中しています。 私は、この同じパターンを、地政学リスクがクラウドインフラとどのように交差するかについて調べたときに、この同じパターンを見ました。

The Global Chip Supply Chain's Hidden Weakness Isn't Silicon. It's Helium. Somewhere in a TSMC fab in Taiwan right now, an EUV lithography machine is etching transistors at scales measured in single-digit nanometers. The light source that makes this possible runs at temperatures so extreme that only one element on Earth can cool it reliably: helium. And the world is running out of it. Not slowly. Not theoretically. Right now. The global helium supply is getting squeezed from at least three directions at once, and the semiconductor industry — which underpins every phone, server, car, and AI accelerator you care about — sits directly in the blast radius. This isn't the kind of supply chain story that makes front-page news. There's no dramatic embargo, no single event to point to. It's a slow, compounding failure across multiple geographies and decades of policy neglect. If you're building anything that depends on chips (so, everything), you need to understand what's happening. Most people associate helium with party balloons and squeaky voices. That's roughly 8% of global helium consumption. The other 92% goes to stuff most engineers never think about: MRI machines, fiber optic production, rocket propulsion, and critically, semiconductor manufacturing. In a modern fab, helium plays several roles that no other element can fill. It creates the inert atmosphere required during silicon wafer processing, preventing oxidation and contamination at the atomic level. A single defect at the 3nm node can ruin an entire wafer. Helium's chemical inertness makes it irreplaceable here. But the bigger story is EUV lithography. The ASML machines that TSMC, Samsung, and Intel all depend on for cutting-edge production use helium to cool their extreme ultraviolet light sources. These machines cost upwards of $150 million each. They cannot operate without a steady, high-purity helium supply. There is no substitute. You can't swap in argon or nitrogen. Helium's thermal conductivity is six times higher than air. Nothing else does this job. After 14+ years of building and operating systems, the pattern I keep seeing is the same: the most dangerous single points of failure are the ones nobody's monitoring. Helium in semiconductor manufacturing is exactly that kind of invisible dependency. The current crisis actually has a name in the industry: Helium Shortage 4.0. Phil Kornbluth, President of Kornbluth Helium Consulting and one of the most cited experts on global helium markets, has been tracking these cycles for decades. The short version: multiple unrelated events collided at once. Here's what went wrong. Russia's Amur plant never delivered. A massive new helium production facility at the Amur Gas Processing Plant in eastern Russia was supposed to supply nearly a third of global helium demand. It was the industry's answer to tightening supply. Then the war in Ukraine happened, sanctions followed, and the plant suffered multiple technical setbacks including fires. The helium that was supposed to relieve global markets simply never arrived. The U.S. sold off its strategic reserve. The Federal Helium Reserve near Amarillo, Texas, was established during the Cold War as a strategic stockpile. For decades, it acted as a buffer for global supply fluctuations. The U.S. Bureau of Land Management has been systematically selling it off, with the process expected to wrap up by 2023. That buffer is gone. Qatar remains a concentration risk. Qatar is one of the world's largest helium producers. When a diplomatic blockade hit the country in 2017, plants accounting for roughly 30% of global helium supply were forced to shut down temporarily, as reported by Gasworld. The blockade was eventually resolved, but the incident exposed a brutal truth: helium supply is concentrated in a handful of geopolitically volatile regions. I wrote about this same pattern when looking at how geopolitical risk intersects with cloud infrastructure. Critical infrastructure concentrated in a few locations, with the industry pretending diversification exists when it doesn't. Different resource, same structural failure. The tech industry has spent the last several years obsessing over chip supply chain risks. CHIPS Act funding. TSMC building fabs in Arizona. Intel's comeback strategy. Samsung expanding in Texas. Everybody's talking about where chips get made. Almost nobody is talking about what chips get made with. Helium isn't like other industrial gases. You can't manufacture it. It's extracted as a byproduct of natural gas production, and once it escapes into the atmosphere, it's gone. Not "gone" in a recyclable sense. Gone gone. Helium is light enough to reach escape velocity from Earth's gravity. Every liter that leaks from a pipeline or floats away in a party balloon is helium the planet will never get back. Global production sits at roughly 6 billion cubic feet per year. Semiconductor manufacturing's share has been growing steadily as the industry pushes to more advanced nodes requiring more EUV lithography steps. A single leading-edge chip can require over a dozen EUV layers, each one consuming helium for cooling. As TSMC ramps 3nm and pushes toward 2nm, and as Intel tries to catch up with its own EUV-heavy process nodes, demand is only accelerating. Meanwhile, supply is fracturing. The math does not work. The semiconductor industry has built a $600 billion annual revenue machine on the assumption that helium will always be available and affordable. That assumption is breaking down right now, and almost no one outside the gas industry is paying attention. I've debugged production outages where the root cause turned out to be some upstream dependency nobody thought to monitor. Helium is that dependency for the entire chip industry. Except the "system" is the global economy and the "debug time" is measured in years, not hours. To be fair, some companies are responding. Helium recycling systems are becoming more common in fabs, capturing and purifying helium for reuse rather than venting it. ASML has been working on reducing helium consumption in its EUV tools. Some research labs and MRI manufacturers have invested in closed-loop cooling systems that cut waste dramatically. But recycling doesn't solve the fundamental problem. It slows the bleed. New sources are being explored — Tanzania has promising deposits, and there are efforts to extract helium from non-traditional natural gas fields — but these projects are years from meaningful production. The deeper issue is structural. Helium has historically been so cheap and so invisible that nobody built the kind of strategic thinking around it that, say, lithium or rare earth elements now command. There's no OPEC for helium. No international body coordinating supply. No futures market providing price signals to drive investment. It's a commodity managed with the seriousness of an afterthought, despite being essential to the most strategically important manufacturing process on Earth. This is the AWS us-east-1 outage pattern all over again: a single process, assumed to be reliable, turns out to be a critical failure mode nobody architected around. Here's what I think actually happens. The chip industry probably won't face a dramatic, sudden helium shutoff. That's the sensationalized version. Supply chain crises don't work that way. What's more likely is a sustained price squeeze that disproportionately hits smaller players. TSMC, Samsung, and Intel can afford to pay 3x or 5x more for helium. They'll lock in long-term contracts, invest in recycling infrastructure, absorb the cost. Smaller foundries, MEMS manufacturers, research institutions, and the entire medical imaging industry (which depends on helium for MRI superconducting magnets) will get priced out. Helium prices have already more than doubled in recent years. If supply keeps tightening without new sources coming online, those costs push chip manufacturing expenses higher across the board. And those costs flow downstream. To every device, every data center, every AI training run. For engineers and engineering leaders, the takeaway isn't to panic. It's to recognize that the geopolitical forces reshaping infrastructure decisions don't stop at data centers and cloud regions. They reach all the way down to the noble gases cooling the machines that make your silicon. The helium crisis is a preview of something the tech industry needs to internalize: the physical supply chains underpinning digital infrastructure are fragile, concentrated, and poorly understood by the people who depend on them most. We spent 2020-2022 learning this lesson with chips themselves. We're about to learn it again with the materials that make chips possible. Helium today. Neon (another critical fab gas, mostly sourced from Ukraine before the war) yesterday. Ultra-pure water, specialty photoresists, and a dozen other invisible inputs tomorrow. My prediction: within 18 months, at least one major fab will publicly cite helium supply constraints as a factor in production guidance. When that happens, this conversation shifts overnight from "what's helium got to do with chips?" to "how did we miss this?" The answer is always the same. We were looking at the wrong layer of the stack. Originally published on kunalganglani.com