The views expressed here are my own and do not reflect those of any client, employer, or affiliated organization.
In writing about the Iran war’s impact on chip input supply chains and, separately, about the hollowing out of the US industrial base, I’ve been circling an input that almost nobody in the AI infrastructure conversation is talking about. It’s worth its own post, because the supply risk is real, the substitution options are essentially nonexistent, and the geography of where it comes from runs directly through the conflict zone I’ve been writing about for the past month.
The input is helium.
Most people know helium as the gas that makes balloons float and voices squeak. Its industrial role is less familiar and considerably more consequential. Helium is non-renewable — it is a byproduct of radioactive decay trapped in natural gas deposits over geological time, and once released into the atmosphere it escapes Earth’s gravity permanently. It is also, in semiconductor fabrication, effectively non-substitutable. Leading-edge fabs use it as a purge gas to prevent atmospheric contamination in process chambers, as a cooling medium in ion implant equipment, in fiber optic production, and as a carrier gas in dozens of specialized steps. Nitrogen and argon work in some applications. In many of the most sensitive ones, they don’t. There is no drop-in alternative at the process level.
The supply picture has been tightening for years. The United States was historically the world’s dominant helium producer, anchored by the Federal Helium Reserve in Texas, which was established during World War II and spent decades as the global backstop. That reserve has been drawn down and its management privatized, removing the strategic buffer it once provided. Production from existing US fields continues but is declining. The other major sources are Qatar, which operates one of the world’s largest helium liquefaction facilities co-located with its LNG infrastructure at Ras Laffan, and Russia, whose Amur Gas Processing Plant in Siberia was intended to emerge as a significant new source. The Amur plant has had a troubled history — fires, construction delays, sanctions complications, and uncertainty about offtake arrangements have left it underperforming its original projections.
That leaves Qatar as the swing producer. And Ras Laffan is no longer a hypothetical risk on a map. It is a facility that was struck, halted, and damaged during an active war.
This is not geographic exposure in the abstract. On March 2, 2026, following initial Iranian drone strikes on the complex, QatarEnergy halted all natural gas production and declared Force Majeure on its supply contracts. On March 18, Iran struck again with maneuverable ballistic missiles capable of evading Patriot air defense systems — a strike that caused fires and significant damage and prompted Qatar to expel Iranian diplomatic staff. Qatar’s energy minister subsequently assessed that 17 percent of the country’s LNG export capacity had been disrupted, with repair timelines of three to five years. Helium production is co-located with LNG processing at Ras Laffan. It does not have a separate facility to fall back on. When LNG production halted, helium production halted with it.
The semiconductor fab dependency makes this timeline deeply uncomfortable. Leading-edge fabs — TSMC, Samsung, Intel — consume significant helium volumes in their process steps. Fab supply buffers are typically measured in weeks to a few months, not years. A three-to-five-year repair horizon at the world’s largest LNG and helium co-production complex is not a supply chain stress test. It is a structural gap. The chips that were supposed to fill the next generation of AI data centers depend on a fabrication process that depends on a gas whose primary swing producer just had its main facility damaged in a missile war.
The downstream effect on data centers is not immediate — it arrives 12 to 18 months later, when chip availability tightens and hardware that was supposed to fill facilities simply isn’t delivered on schedule. The data center debt, already precarious on the terms I wrote about in the first post in this series, gets serviced against hardware that isn’t there.
This is the supply chain fragility argument made concrete. Helium isn’t traded like oil. It moves primarily on long-term bilateral contracts between producers and industrial consumers. The spot market is thin and illiquid. There is no strategic helium reserve with meaningful scale. When a supply disruption hits, it doesn’t get priced in gradually through a visible market signal. It propagates through fab production schedules and chip delivery timelines, largely out of public view, until the shortage shows up downstream in ways that are difficult to attribute cleanly to a single cause.
Mitigation efforts exist. Some leading-edge fabs have invested in helium recovery and recycling systems that reduce consumption per wafer. Exploration is underway in Tanzania, where geological surveys have identified potentially significant deposits, and in Canada. None of these are at a scale or on a timeline that addresses a three-to-five-year disruption at Ras Laffan. Tanzanian deposits require years of development infrastructure. Canadian sources are real but modest.
The AI infrastructure conversation is almost entirely focused on chips, power, and data center debt. Those are legitimate concerns. But the invisible inputs — the gases, the rare earths, the precision materials that don’t make headlines until they’re gone — are where the most acute near-term vulnerabilities sit. Helium is the clearest example: non-renewable, non-substitutable, geographically concentrated, and now physically disrupted at its primary production site by an active military conflict.
The chokepoint doesn’t have to be dramatic to be decisive. It just has to be there. In this case, it was there, and it got hit.
This is the third in a three-part series on AI infrastructure and its supply chain vulnerabilities.