48 Days. That's How Long Before the Helium Runs Out for AI Chips.

A missile-hit helium and LNG hub in Qatar could tighten the physical supply chain behind advanced AI hardware for years—pushing up memory and chip costs, slowing output, and reshaping who can build and deploy compute fastest. The core bottleneck is helium: it’s not just another industrial gas, but a critical input for multiple steps in semiconductor manufacturing, including EUV lithography support and vacuum/etching processes that require helium’s unique properties. With a major Qatar helium plant offline for weeks and additional damage reported, the clock is measured in months for shipping containers and in years for rebuilding complex industrial capacity.

The disruption traces to the Ross Lefon complex, which historically supplies a large share of global helium. The Strait of Hormuz is effectively acting as a choke point for shipments headed to East Asia, where most advanced chipmaking sits. Liquid helium also can’t be stockpiled indefinitely: specialized ISO containers can hold it for roughly 35–48 days before it starts to vaporize, meaning stranded shipments can become unusable if the outage drags on. Adding to the risk, the Ross Lefon facility has reportedly been hit by missiles, and Qatar energy has admitted that about 14% of helium capacity is permanently damaged, with reconstruction timelines that could stretch up to five years.

Helium’s role is tightly coupled to chip fabrication. During plasma etching, helium is blown over wafers to pull heat and keep temperature uniformity; helium’s small atomic size also makes it useful for detecting leaks in vacuum chambers quickly and reliably. As AI chips advance—especially HBM memory used in Nvidia GPUs, AMD accelerators, and Google TPUs—helium consumption rises. A 300 mm EU fab may use between 5 and 20,000 cubic meters of helium per month, and the gas must meet extremely high purity requirements (6N, or 99.9999%). The supply base for that grade is limited, and the transcript emphasizes that there is no practical substitute.

The Qatar shock also links to LNG. Helium is produced as a byproduct of cryogenic LNG processing, so LNG disruptions and helium shortages share the same timeline. That matters for chip fabs because LNG affects energy costs—another major input for semiconductor manufacturing. Higher energy prices in East Asia can translate into higher chip prices and, ultimately, higher inference costs measured in cost per flop.

Beyond costs, the geopolitical angle could tilt the AI race. China is pursuing energy diversification via pipelines such as Power of Siberia 2 and is also building domestic helium capacity in Guangdong with 6N certification. If China secures cheaper, steadier helium and energy, it could lower the cost of its chip fabrication stack and expand deployment of Chinese compute in the late 2020s—creating a structural advantage over South Korea and Taiwan, which lack domestic helium supply and rely heavily on imports.

Even if fabs keep running, the key question becomes capacity and timing: whether production slows, whether scaling is constrained, and how quickly suppliers can qualify new helium sources. The practical forecast offered is that memory costs—especially for HBM and DRAM—stay elevated through mid-2027, while energy-driven cost pressure persists. For consumers and enterprises, the implication is straightforward: delays and price increases are likely, and procurement decisions may need to happen sooner rather than later because the bottleneck may not resolve on a predictable schedule.