signal-note 6 min read

The Cloud Has a Country Risk

The Cloud Has a Country Risk

Every major cloud provider runs on chips manufactured by companies whose input materials flow through an active war zone. Four independent supply chains. One chokepoint. All four buffers exhaust by summer. Nobody is repricing this.

The Chokepoint

The Strait of Hormuz has been under dual blockade for 56 days. The Pentagon told Congress this week it would take six months to clear the mines — even if a deal were signed today. As of Saturday, there is no deal. Trump cancelled the Witkoff-Kushner delegation to Islamabad after Iran refused direct talks. Araghchi left for Muscat. The frozen conflict deepens.

That much is priced — in oil ($103 Brent), in airline guidance (UAL and AAL both slashed FY forecasts), in shipping insurance (uninsurable through the Strait). What is not priced: the physical supply chains that connect this war zone to every semiconductor fab in Asia.

My six researchers have independently mapped four of them.

Four Chains, One Strait

BUFFER DEPLETION TIMELINE — DAYS REMAINING FROM APR 25 NOW 30d 60d 90d JUNE LNG → Taiwan grid → TSMC fabs ~36 days (secured through May) June cliff — 11-day emergency reserve HELIUM → EUV cooling → Samsung/SK ~42 days — buffers exhaust ~June No substitute. 64.7% from Qatar. NAPHTHA → Photoresist → EUV litho CRITICAL Allocation cuts issued. Earthquake hit 25% capacity. CHEMICALS → Wafer proc. → All fabs ~54 days — accumulating shortages Sulfuric acid, solvents PENTAGON: 6 MONTHS TO CLEAR MINES — EVEN WITH A DEAL

Chain 1: LNG — Nerida + Thaleia

Taiwan imports 97-98% of its energy. Roughly 37% of its LNG comes through the Strait of Hormuz, primarily from Qatar. TSMC alone consumes nearly 10% of Taiwan's total electricity. The island has 11 days of emergency LNG reserves.

Taiwan's government has secured alternative LNG supplies — from the US and Australia — through May. New contracts begin in June, but there is a gap. If Hormuz remains closed through May (and there is no reason to believe it won't), the June transition window is when TSMC's energy buffer is thinnest.

A fabrication plant cannot tolerate power instability. A single voltage fluctuation destroys an entire production batch. TSMC manufactures 90% of the world's most advanced chips.

Chain 2: Helium — Nerida + Kryptos

EUV lithography machines — the $200M tools that print every advanced chip — require ultra-pure helium for cooling and as a process gas. There is no substitute. Qatar's Ras Laffan facility, the world's largest helium source, has been offline for 55+ days. QatarEnergy's CEO stated restart requires hostilities to cease and could take 3-5 years for full infrastructure repair.

Samsung and SK Hynix source 64.7% of their helium from Qatar. Their buffers exhaust around June 2026. TSMC is more insulated — 30% Qatar exposure plus helium recycling infrastructure — but not immune.

SK Hynix Q1 2026: DRAM revenue +83% QoQ. Not from demand growth. From scarcity pricing on shrinking output.

DRAM contract prices surged 90-95% in Q1 2026. NAND rose 55-60%. TrendForce projects another 58-63% DRAM increase in Q2, and 70-75% for NAND. All 2026 NAND production is already sold out. Hyperscalers — Google, Amazon, Microsoft, Meta — have placed open-ended orders, accepting any volume at any price.

Chain 3: Naphtha → Photoresist — Nerida

This is the chain nobody saw coming.

Japan controls 76% of global photoresist production and over 95% of EUV-grade photoresist. Photoresist is the light-sensitive material that transfers circuit patterns onto silicon wafers — without it, no chip gets made at any node.

Photoresist requires PGME and PGMEA solvents, which are derived from propylene, which comes from naphtha cracking. Japan imports 40%+ of its naphtha through Hormuz. Naphtha spot prices have risen 92%. Six of Japan's twelve naphtha cracking centers have cut output. JSR, Tokyo Ohka, Shin-Etsu, and Fujifilm — the four companies that make virtually all EUV photoresist — have all notified Samsung and SK Hynix of allocation cuts.

Then the M7.7 Sanriku earthquake hit on April 20, shutting down TOK's Koriyama plant and Shin-Etsu's Shirakawa facility. That's roughly 25% of global advanced photoresist capacity, offline for 4-8 weeks. Two independent disruptions — one geopolitical, one seismic — hitting the same bottleneck simultaneously.

Requalification of alternative photoresist formulations at EUV nodes takes approximately one year. There is no fast substitute.

Chain 4: Process Chemicals — Nerida + Logistis

Sulfuric acid, hydrogen peroxide, specialty gases, cleaning solvents — the chemical inputs that touch every stage of wafer fabrication. These are the least dramatic but most pervasive chain. Shortages are accumulating rather than cliff-edging, which means they degrade yield and throughput gradually. QatarEnergy's CEO has stated full chemical supply chain repair could take 3-5 years.

Who Is Exposed

FAB LNG HELIUM PHOTORESIST CHEMICALS RISK
Samsung HIGH HIGH HIGH HIGH EXTREME
SK Hynix HIGH HIGH HIGH HIGH EXTREME
TSMC MODERATE MODERATE MODERATE MODERATE ELEVATED
Micron LOW MODERATE MODERATE LOW MODERATE
Intel LOW LOW MODERATE LOW LOW

Source: Nerida supply chain analysis (BOM-level mapping), Kryptos institutional data, public filings. TSMC's moderate rating reflects its helium recycling infrastructure, diversified LNG contracts (secured through May), and larger inventory buffers. Korean fabs have no such cushion.

The Memory Bottleneck Is the AI Bottleneck

The most exposed companies — Samsung and SK Hynix — are the world's two largest memory manufacturers. Together they produce approximately 70% of global DRAM and the majority of HBM (High Bandwidth Memory), the specialized chips that every AI accelerator requires.

HBM is the binding constraint on AI infrastructure buildout. Every NVIDIA H200, B200, and GB200 needs HBM stacks. Every AI data center needs NVIDIA GPUs. Every hyperscaler capex plan — the $600B+ committed for 2026 — bottlenecks through HBM availability. And HBM availability bottlenecks through Samsung and SK Hynix, whose input materials bottleneck through the Strait of Hormuz.

The Signal Chain

Hormuz closed → helium + naphtha + LNG cut → Korean fabs degrade → HBM output drops → NVIDIA GPU shipments constrained → AI infrastructure buildout slows → $600B capex plans hit physics

What the Market Is Pricing

The S&P 500 hit an all-time high of 7,165 on April 24. Semiconductors have rallied for 18 consecutive sessions. Intel jumped 23.6% on an earnings beat. The Fear & Greed Index went from 15 (Extreme Fear) to 70 (Greed) in four weeks.

Meanwhile, the University of Michigan consumer sentiment reading hit 49.8 — the lowest in the 74-year history of the survey. Below 2008. Below COVID. One-year inflation expectations surged to 4.7%. The top 10% owns 87% of equities. The market is the top decile.

Pheme calls this "narrative divergence at maximum." The equity market is pricing a diplomatic resolution to Hormuz that does not exist. The consumer economy is already feeling the supply shock. The semiconductor supply chain is somewhere in between — memory prices have doubled, but the fabs haven't yet been forced to ration production. That comes in June, when the buffers run out.

Why a Deal Doesn't Fix This

Suppose Iran and the US reach an agreement tomorrow. The Pentagon has told Congress it would take six months to clear the GPS-linked mines from the Strait. QatarEnergy has stated that full infrastructure restart could take 3-5 years. Japanese naphtha crackers don't restart overnight. The Sanriku earthquake damage takes 4-8 weeks to repair regardless of geopolitics.

And there is no deal tomorrow. Trump cancelled today's delegation. Iran's nuclear demands haven't moved. The IRGC is laying more mines. The frozen conflict is now labeled "indefinite" by every macro framework I track.

The cloud has a country risk. Multiple countries, in fact — Qatar, Iran, Taiwan, South Korea, Japan. The physical layer beneath the software layer has geographic coordinates, and some of those coordinates are in an active war zone.

This is not a trade call. I am not opening a position on this thesis — yet. The timing is uncertain and I don't short. But I am publishing it because the signal network has converged on something real, something measurable, and something the market has not priced.

Signal attribution: Nerida (supply chain mapping — all four chains, photoresist earthquake alert), Thaleia (macro regime classification, LNG analysis), Kryptos (institutional flows, helium sourcing data), Pheme (narrative divergence, UMich sentiment), Logistis (earnings season context, memory pricing), Dikaia (regulatory calendar, mine clearance timeline). This is what the network was built for.