The global semiconductor supply chain entered 2026 with one dominant narrative: TSMC's Arizona fabs beginning production, Intel's Foundry division losing $2.5 billion on $4.5 billion in revenue, and Samsung navigating the memory cycle. [1] The Iran war has added a new and less anticipated variable to all three storylines: helium. And helium is not substitutable, not storable at scale beyond existing reserves, and not something the market can produce more of on short notice.

The connection between helium and chip manufacturing is physical. Extreme ultraviolet lithography — the process TSMC, Samsung, and Intel use to produce chips at the 3nm and 2nm nodes that power AI accelerators and advanced processors — requires high-purity helium as a coolant for the light source optical path. Without helium, the EUV machine overheats and cannot produce chips to specification. [2] TSMC's Arizona facilities, which are producing Apple's next-generation processors and NVIDIA's H200 AI accelerator, are among the highest-priority allocations in the US government's helium distribution hierarchy — which is itself not publicly documented but is inferred from supply chain disclosures.

The helium shortage has introduced a cost variable that the semiconductor industry did not model in its 2025 capital expenditure plans. The Bureau of Land Management's Amarillo reserve is contractually allocated through 2027 under existing medical and scientific agreements, leaving the spot market for new purchases effectively uncovered. Spot helium prices at $2,000 per thousand cubic feet represent a 100% increase from early 2026. [3] For a TSMC fab running 24 EUV machines, the helium cost increase alone adds approximately $0.40 per wafer processed — a small number multiplied by millions of wafers that adds meaningful cost to the most expensive manufacturing process humans have ever designed.

Samsung is facing the additional complication of Korean government helium reserve allocation for domestic semiconductor fabs, which has created tension with the company's Arizona and Texas operations. [4] Intel's Foundry division, already operating at a $2.5 billion quarterly loss, cannot absorb additional input cost increases without either raising chip prices — and losing market share to TSMC — or accepting further margin compression. The company has publicly stated it is in discussions with customers about price adjustments for 2027 wafer deliveries. [5]

The AI hardware investment cycle is running headlong into these cost realities. The $2.5 trillion in projected AI spending for 2026, cited by Gartner this week, was modeled on semiconductor supply chain assumptions that no longer hold. [6] The chip shortage the industry feared in 2021 has not materialized in the same form. What has materialized is a cost shortage — the ability to produce AI hardware at a price point that makes the investment math work — that is not the same problem but may be harder to solve.