The Gas War

Inside an ASML deep ultraviolet lithography scanner, a gas mixture ignites sixty thousand times per second. The mixture is ninety-five percent neon, with traces of argon and fluorine. Each pulse lasts twenty-five billionths of a second and produces a beam of light at exactly 193 nanometers — a wavelength in the far ultraviolet, invisible to the human eye. It etches circuit patterns at a resolution finer than the wavelength of visible light. The beam passes through a series of lenses made from calcium fluoride crystals grown over thirteen months in vacuum furnaces, refracting the light into the shapes of transistor gates measured in single-digit nanometers. The entire operation takes place inside a chamber so clean that a single particle of dust would be the equivalent of a boulder dropped onto a highway.

This is how chips are made. Not with hammers or kilns or anything that looks like manufacturing in the conventional sense. With gas, and light, and a degree of precision that is difficult to describe without resorting to analogy. And the gas — the neon that constitutes ninety-five percent of the mixture that makes the light that patterns the silicon that becomes the chip that trains the model — came, overwhelmingly, from two factories in Ukraine that most of the world had never heard of.

Ingas, in Mariupol. Cryoin, in Odessa. Together, they produced forty-five to fifty-four percent of the world's semiconductor-grade neon. The United States imported ninety percent of its chip-grade neon from Ukraine. The process was elegant in its obscurity: Russian steel mills captured crude neon as a byproduct of oxygen production for steelmaking — neon exists at 18.2 parts per million in the atmosphere and concentrates during air separation — then shipped the crude gas to Ukrainian purification facilities, which refined it to 99.999 percent purity and sold it to chip manufacturers in Taiwan, South Korea, China, the United States, and Germany.

Russia controlled the feedstock. Ukraine controlled the purification. The war severed both simultaneously.

On February 24, 2022, Russian forces crossed the Ukrainian border. Cryoin, in Odessa, had already halted operations thirteen days earlier, anticipating what was coming. Ingas, in Mariupol, was captured by Russian forces within weeks. Key staff evacuated to unoccupied Ukrainian territory. The company later issued a public warning that any neon production at the Mariupol site under Russian control was illegal. The warning carried no enforcement mechanism. It carried only the weight of a company watching its life's work appropriated by an invading army.

Neon prices in China surged from 1,850 yuan per cubic meter to 16,000 yuan — roughly nine times — within weeks. Some spot-market reports cited increases of five thousand percent. The chip industry, which had been warned about this exact vulnerability after Russia's 2014 annexation of Crimea and had done almost nothing about it, suddenly discovered that the invisible gas powering its most critical manufacturing process came from a war zone.

But the chips did not stop. This is important, and it complicates the story in ways the fear-based narrative does not capture. Chip manufacturers had three to six months of neon inventory stockpiled. The price spikes were real; the shortages were feared but did not materialize into factory shutdowns. The Semiconductor Industry Association confirmed limited immediate disruption. The supply chain bent. It did not break.

What it did instead was reshape itself — faster and more effectively than almost anyone predicted.

ASML's subsidiary Cymer, which builds the lasers that consume the neon, had already been working on consumption reduction. Their program achieved results that, in any other context, would be called remarkable: fifty percent reduction in neon consumption for the most common deep ultraviolet lasers. Thirty percent for older models. A next-generation gas control technology, now deploying, reduces consumption by up to seventy-five percent. The annual savings exceed seventy million liters, worth more than two hundred million dollars. Recycling technology — capturing used neon from the laser chambers, filtering it, reconstituting the correct mixture for reuse — is being installed in fabs worldwide. SK hynix and TEMC developed the first neon recycling system in South Korea. ASML now sources less than twenty percent of its neon from Russia and Ukraine, down from a much higher pre-war dependency.

New production came online. POSCO built the first neon gas production facility in South Korea. China invested heavily, reaching installed capacity of fourteen million cubic meters per year. Linde's La Porte, Texas facility — built in 2016 with a quarter-billion dollars of investment, notably after the 2014 Crimea shock rather than the 2022 invasion — produces forty million liters annually. Messer inaugurated a specialty gases facility in Pennsylvania in May 2025 producing krypton, neon, and xenon. Global neon capacity increased an estimated fifteen percent.

And then there is what makes the crisis temporary. Extreme ultraviolet lithography — EUV, the successor technology to deep ultraviolet — does not use neon at all. EUV systems operate in a hydrogen background gas. As the semiconductor industry transitions to five-nanometer, three-nanometer, and two-nanometer process nodes — the nodes that manufacture the chips that train frontier AI models — the fundamental dependency on neon disappears. Not because the industry found a substitute for neon. Because it built a machine that does not need it.

The neon crisis, by the metrics that matter, has been substantially resolved.

This should be reassuring. It is not. Because the lesson of the gas war is not that the system is resilient. The lesson is that the system responds to crises after they happen and almost never before.

The 2014 Crimea annexation was a warning. Ukraine's neon concentration was well-documented. Analysts published papers. Industry groups noted the risk. And for eight years, the global semiconductor industry did essentially nothing to diversify its neon supply. One American company — Linde — built a domestic facility. Everyone else waited until Russian tanks rolled into Mariupol.

The industry fixed the neon problem in roughly eighteen months. But it fixed the neon problem only because a war forced the issue. And similar or worse chokepoints remain unfixed, unaddressed, and in some cases, actively worsening.

But neon was the solvable problem. The unsolvable ones remain.

Japan controls over ninety percent of the global high-end photoresist market — the light-sensitive chemicals used to transfer circuit patterns onto silicon wafers. Four Japanese firms command eighty-seven percent of all photoresist sales. Japan weaponized this chokepoint once already, imposing export controls on photoresists to South Korea during a sovereignty dispute in 2019. In November 2025, Chinese forums erupted with reports that Japan had halted photoresist exports to China; Shin-Etsu's exports to China reportedly dropped forty-two percent in a single month. There is no recycling technology for photoresists. The chemistry is more complex than neon purification. Alternative development timelines are measured in decades, not months.

PFAS-containing chemicals — so-called "forever chemicals" — are essential to the photo-acid generators used in advanced lithography. All successfully demonstrated formulations for high-performance semiconductor resists are fluorinated. No universally applicable fluorine-free alternatives exist. Development of alternatives is expected to take fifteen to twenty years. The European Union's proposed PFAS ban could, if implemented without semiconductor exemptions, create a self-inflicted supply crisis.

And the most glaring vulnerability of all has not been resolved: the concentration of advanced chip fabrication in Taiwan that Chapter 2 documented remains essentially unchanged. The CHIPS Act is building capacity at the margin — TSMC's Arizona Fab 1 is online, producing ten thousand wafers per month with yields four percent higher than Taiwan — but the fundamental geographic concentration has not changed and will not change within this decade.

The gas war was a dress rehearsal. The industry passed the test, barely, with months of inventory and years of warning. The next disruption may offer neither. And the inputs at risk — photoresists from Japan, PFAS chemistry with no substitute, fabrication capacity concentrated on a single island — are harder to replace than neon ever was.

That is the lesson, and it is not the lesson most people took from it. The popular version — the supply chain is resilient, markets adapt, crisis averted — is the comforting reading. It is also the dangerous one. Because it teaches exactly the wrong thing: that you can wait for the war, then fix the problem, and everything will be fine. The neon crisis allowed that luxury. Three to six months of inventory. Eight years of warning after Crimea. A technology transition already underway that made the dependency obsolete.

The next crisis may offer none of these. And the industry that fixed neon in eighteen months has spent the years since congratulating itself rather than asking what else it has failed to see.