The Strait of Hormuz is 21 miles wide. Three independent chemical supply chains for semiconductor manufacturing pass through it. Not one. Three.
I've mapped the first two before. Helium — Qatar's offline production choking fab cooling and leak testing. LNG and power — Gulf gas disrupting electricity for fabs that consume as much power as small cities.
The third line is the one almost nobody traces. And it may be the one that binds first.
Crude → Naphtha → Propylene → PGME/PGMEA → Photoresist → Every Advanced Chip
The Chain Nobody Traces
Middle East crude gets refined into naphtha. That naphtha gets cracked at high temperatures in Japanese petrochemical plants into propylene. Propylene becomes propylene oxide. Propylene oxide becomes PGME and PGMEA — two solvents that together constitute 90% of all photoresist solvent used in semiconductor manufacturing.
Photoresist is the light-sensitive coating applied to silicon wafers during lithography. Without it, you cannot pattern a chip. At advanced nodes — the sub-5nm processes used by TSMC and Samsung for AI accelerators, smartphone processors, and HBM memory — only EUV photoresist works. And EUV photoresist cannot tolerate impurities, which means it cannot easily switch solvent suppliers.
Japan produces 76% of the world's photoresist. Four companies — JSR, Tokyo Ohka Kogyo, Shin-Etsu Chemical, and Fujifilm — dominate. Japan imports over 40% of its naphtha from the Middle East, and over 90% of its crude oil. When Hormuz constricted in March, Japanese naphtha spot prices nearly doubled: $600/ton to $1,190/ton.
Six of Japan's twelve naphtha cracking centers cut output.
The Notifications
In April, Shin-Etsu Chemical and Tokyo Ohka Kogyo formally notified Samsung Electronics and SK Hynix of impending supply cuts to photoresist and photoresist solvents. This wasn't a warning about hypothetical risk. This was a procurement notification: we cannot guarantee your volumes.
| MATERIAL | ROLE IN CHIPMAKING | HORMUZ EXPOSURE |
|---|---|---|
| PGMEA | Core photoresist solvent (80-90% of formulation) | Naphtha → propylene → PO → PGMEA |
| PGME | Co-solvent in photoresist | Same chain as PGMEA |
| BARC | Bottom anti-reflective coating for EUV | PGME/PGMEA dependent |
| SOH | Spin-on hard mask | PGME/PGMEA dependent |
| HBM adhesive | Temporary bonding for HBM stacking | PGME/PGMEA dependent |
That last row is the quiet devastation. HBM4 — the high-bandwidth memory that NVIDIA's Rubin platform needs — uses temporary bonding adhesives that depend on the same PGME/PGMEA solvents. The photoresist shortage and the memory shortage are the same shortage, traced to the same barrel of naphtha.
The Buffer Clock
Semiconductor manufacturers typically maintain several months of safety inventory for critical materials. The Hormuz disruption began in earnest in early March 2026. U.S. alternative supplies can partially substitute, extending the buffer to roughly six months.
Six months from March = August–September 2026.
We are five weeks from the edge of the buffer. Hormuz traffic has recovered to 50% — but 50% of pre-crisis flow through a still-mined channel, with war-risk insurance at 2% of hull value, does not refill chemical inventories at pre-crisis rates. The math doesn't close.
And switching suppliers isn't an option. Changing the source of raw materials in a photoresist formulation triggers a Process Change Notification — a requalification procedure that takes approximately one year at advanced nodes. South Korea sourcing PGME from China instead of Japan doesn't help Samsung make chips next quarter. It helps them make chips in mid-2027.
The Double Exposure
SK Hynix just made a decision that reveals how tight the market has become. In June, the company slowed its HBM4 production ramp to chase elevated DDR5 margins instead. The stated reasons: revised-down NVIDIA Rubin production forecasts, and the fact that general DRAM operating margins have overtaken HBM premium margins amid severe shortages.
Read that again. Memory is so scarce that commodity DRAM is more profitable than premium HBM. DRAM contract prices rose 90-95% quarter-over-quarter in Q1 2026, then another 58-63% in Q2. SK Hynix signed a three-year DDR5 supply deal with Microsoft, locking in margins.
Meanwhile, NVIDIA is reportedly relaxing its HBM4 specifications — accepting 10Gb/s parts alongside the 11Gb/s it originally demanded — because it cannot get enough supply at any spec level. Samsung's yields on its 1c DRAM node sit around 60%, declining further after back-end processing.
The photoresist shortage feeds directly into this. Less photoresist means fewer wafer starts. Fewer wafer starts in a market where DRAM has already risen 150%+ in two quarters means every additional disruption gets amplified, not absorbed.
Why Three Lines Matter More Than One
Each of these three supply chains — helium, LNG/power, naphtha/photoresist — passes through the Strait of Hormuz independently. They use different ships, different ports, different downstream processors. They fail for different reasons and at different speeds.
But they converge at the same place: the fab floor.
If you estimate that each line has, say, a 30% probability of becoming a binding constraint on semiconductor output over the next six months, the probability that none of them bind is 0.7 × 0.7 × 0.7 = 34%. The probability that at least one binds is 66%.
That's the structural point. Analyzing any one of these chains in isolation understates the risk. The semiconductor industry doesn't have one Hormuz exposure — it has three, and they multiply.
The first line — helium — I mapped in March. The second — power — in April. The third was always there in the data. I should have traced it sooner. The buffer clock is now the story.