The AI Chip Supply Chain: A Complete Guide

Why the AI chip supply chain is one of the most consequential systems in human history

There are moments in history where a single technology redefines the pace at which humanity advances. The printing press compressed centuries of knowledge diffusion into decades. The steam engine multiplied human labor by orders of magnitude. The internet collapsed geography as a barrier to collaboration. Artificial intelligence is that kind of technology — and the case can be made that it is the most profound of all, because AI accelerates every other technology simultaneously.

AI is already compressing drug discovery timelines from decades to years. It is writing production code, synthesizing research across fields that previously could not communicate, and modeling protein structures that chemists spent careers trying to understand. The pace at which humanity will innovate in the coming decades is being redefined right now — not in a single laboratory, but in the training runs of large language models that will serve as the substrate of the next era of human problem-solving. A disruption to this technology is not merely an economic inconvenience. It is a delay in our combined history.

All of that depends on chips — the GPUs and AI accelerators that power model training and inference. And those chips are manufactured through one of the most complex, geographically dispersed, and brittle supply chains that industrial civilization has ever assembled. As of June 2026, four companies sit at the center of this system: NVIDIA (NVDA: ~$216), whose GPU architecture dominates AI training; ASML (ASML: ~$1,720 — an all-time high reached on June 3), the world's only manufacturer of EUV lithography machines; TSMC (TSM: ~$445, market cap $2.28 trillion), which manufactures over 90% of the world's sub-7nm chips; and SK Hynix (000660.KS: ₩2,160,000), the dominant supplier of HBM memory that makes GPUs fast enough to train the models reshaping the world. These valuations are not speculative — they reflect the market's recognition that these companies hold positions without which AI development cannot proceed.

(Stock prices as of June 2026)

What is the AI chip supply chain and how many tiers does it have?

The AI chip supply chain is the end-to-end network of companies, materials, and processes that converts raw earth elements into the GPUs and accelerators powering AI. AIChipMap maps it across 17 distinct tiers — each a different industrial category with its own concentration patterns and risk profile.

The raw-materials tier includes rare earths (for magnets in lithography machine actuators), silicon purified to parts-per-billion, specialty gases including neon, krypton, and xenon for laser light sources, and specialty chemicals. China dominates rare-earth refining at roughly 85%. Japan controls a large share of specialty chemical production. These concentrations interact: a political decision in Beijing or a natural disaster in Japan can affect chip output in Taiwan within weeks.

Equipment tiers cover lithography (ASML for leading-edge EUV; ASML, Canon, and Nikon for DUV), deposition (Applied Materials, Lam Research), etch (Lam Research, Tokyo Electron), and inspection (KLA). Without this equipment, no fab can operate regardless of capital.

The foundry tier is the most-discussed chokepoint. TSMC manufactures over 90% of the world's sub-7nm chips. Samsung Foundry and GlobalFoundries serve the trailing edge. SMIC, China's leading fab, remains two to three nodes behind the frontier despite massive state investment.

Above the foundry: chip designers (NVIDIA, AMD, Google, Apple), memory makers (SK Hynix, Micron, Samsung), OSAT packaging (ASE Technology, Amkor), server ODMs (Foxconn FII, Quanta Computer, Wiwynn), cloud providers (AWS, Azure, Google Cloud), and the enterprises and researchers running the AI workloads that are reshaping our world. Each tier is indispensable — none can be instantly replicated.

Where are the chokepoints — and why can't they be replaced?

A chokepoint is a supply chain node where concentration is so extreme that disruption cannot be absorbed by rerouting. AIChipMap identifies 13 chokepoints across 17 tiers. Five are existential.

ASML's EUV monopoly: ASML is the world's only EUV lithography supplier. Each machine contains over 100,000 components from 5,000 suppliers across 40 countries, takes 12–18 months to build, costs approximately $200 million, and ships in 40 cargo planes. Canon and Nikon abandoned EUV in the 2000s. There is no alternative. ASML's moat is three decades of accumulated patents, supplier relationships, and manufacturing knowledge that cannot be shortcut. Its stock at $1,720 — an all-time high — reflects the market's view that this monopoly is durable.

TSMC's foundry concentration: TSMC's N3 and N2 nodes have no production alternative for AI-grade chips. Moving an order to Samsung Foundry or Intel Foundry means accepting a process generation penalty — directly translating to worse performance and higher power consumption — which is unacceptable for frontier AI hardware.

SK Hynix's HBM dominance: High Bandwidth Memory uses Chip-on-Wafer (CoW) packaging that took SK Hynix years to develop. At ~₩2,160,000 per share — and with Samsung Memory working to close the gap on HBM4 — SK Hynix currently commands 50–60% of global HBM supply. Capacity expansion takes 2–4 years from investment to production.

Rare-earth concentration in China: China refines approximately 85% of global rare earths. The refining is concentrated not because the ore is rare but because the process is environmentally intensive and was subsidized into dominance over decades.

Specialty gases: Japan (through Tokyo Electron's chemical supply chain) and gas specialists are the primary stable source after Ukraine — which supplied 70% of global neon before Russia's invasion — was taken offline. These gases cannot simply be substituted; their combination of thermal properties, chemical inertness, and atomic size is unique.

How three wars are disrupting the AI chip supply chain in 2026

The AI chip supply chain does not exist in isolation from geopolitics. In 2026, three active conflict zones are reshaping its risk in ways that every procurement decision and investment thesis must account for.

The Iran War (Operation Epic Fury, February 28–May 5, 2026): US-Israel joint strikes on Iran's nuclear and military infrastructure triggered retaliatory missile salvos and, critically for semiconductors, the effective closure of the Strait of Hormuz beginning March 4. Iranian strikes hit Qatar's Ras Laffan Industrial City — one of the world's two plants capable of producing semiconductor-grade helium — knocking offline roughly 30–35% of global supply in days. Helium prices doubled (Fitch Ratings); spot prices surged 40–100% depending on grade. The crisis is compounded by physics: liquefied helium evaporates within 45 days, meaning stranded inventory cannot be stored and shipped later. Roughly one-third of global cryogenic helium ISO containers are stranded in or around Qatar with no viable maritime route to market.

Helium is not optional. It serves four irreplaceable roles in semiconductor manufacturing: cooling EUV lithography machines, detecting microscopic leaks via helium mass spectrometry, acting as an inert carrier gas during thin-film deposition, and cooling wafers during ion implantation. There is no viable substitute for any of these applications.

Bromine is a compounding shock. Israel and Jordan together supply approximately two-thirds of global bromine, used in specialty photoresist chemicals and flame retardants in electronics. South Korea sources 90% of its bromine from Israel. Samsung and SK Hynix — together controlling 80% of HBM and 70% of global DRAM — saw their stocks fall more than 20% in two days as the Iran war began, before partially recovering. Taiwan imports 97% of its energy, with 37% of its power grid running on Middle Eastern LNG; its LNG reserves can sustain the island for only 11 days without imports. TSMC CEO C.C. Wei stated on June 4, 2026 that chip supply will continue to lag AI demand for years — the Iran disruption is one reason why.

Russia-Ukraine (ongoing since 2022): Ukraine was the world's largest neon producer — approximately 70% of global supply — before the invasion. Neon is essential for DUV lithography excimer lasers. Spot prices spiked over 500% in 2022. Japanese and American suppliers have since ramped alternatives, but the dependency remains only partially resolved. Any escalation affecting Ukrainian industrial output will tighten semiconductor supply within a quarter.

Taiwan Strait: Taiwan produces over 90% of sub-7nm chips. A blockade would halt leading-edge chip production for years — no alternative can absorb TSMC's volume at its process nodes. This risk has driven TSMC's $165 billion commitment to expand Arizona operations into six fabs, and the US-Taiwan Pax Silica Declaration of January 2026, under which $250 billion in Taiwanese semiconductor investments will be directed toward allied nations. But even by 2027, the vast majority of TSMC's N3 and N2 capacity will remain on the island. Diversification buys resilience at the margin, not at the core.

What export compliance really looks like — and why it slows everything down

There is a person — often a single person on a small team — whose job is to ensure that every chip, server, and piece of semiconductor equipment that leaves a company's warehouse is legally authorized to reach its destination. This is the export compliance officer, and their work is among the most thankless, invisible, and high-stakes in the technology industry.

Consider a scenario that plays out thousands of times a year. A sales team closes a deal to sell a server loaded with NVIDIA H200 GPUs to a research institution in a country that was not on any restricted list last quarter — but which appeared in a newly published BIS guidance note two weeks ago. The compliance officer must now: (1) classify the product under its Export Control Classification Number (ECCN), which requires understanding the chip's exact performance parameters against BIS thresholds revised six months ago and again four months ago; (2) determine whether a license exception applies — a 19-page analysis of License Exception ENC, STA, and the Validated End-User program; (3) run the buyer, the end institution, and all known ultimate end-users through five separate restricted-party screening databases, each with different update cadences; (4) obtain internal legal sign-off; (5) file the Electronic Export Information in the Automated Export System — a US government portal whose UI last saw major redesign in the Obama era.

If any step surfaces a concern, the shipment stops. A license application to BIS takes 30–90 days. During that window, the customer waits, revenue is deferred, and the sales team fields daily escalation calls. The compliance officer simultaneously handles 40 other open cases at similar stages.

The regulations are written in a cross-referencing maze of definitions, exceptions, and carve-outs designed to be comprehensive rather than legible. For Japanese companies navigating Japan's own export control alignment with US and Dutch policy (effective 2023), and for Korean companies managing the US-Korea semiconductor partnership under the Pax Silica framework, the compliance burden is compounded by multi-jurisdictional requirements that can mean a single shipment triggers review under three regulatory regimes simultaneously.

None of this is the compliance officer's fault. They are doing one of the most important jobs in the technology industry — protecting against genuine proliferation risks — while being chronically understaffed and under-recognized. If you are one of them: the friction you feel is real, the stakes are genuine, and what you do matters.

What happens when a supply chain tier breaks — and why recovery takes years

A disruption at any tier ripples both upstream and downstream, often manifesting in counter-intuitive ways that take months to appear and years to resolve.

Consider a hypothetical total disruption at TSMC. Within days, NVIDIA, AMD, Apple, and every fabless chip designer halt their most advanced product lines. Within weeks, cloud providers pause GPU-dense data center deployments. Within months, AI training costs spike as secondary-market GPU prices surge. Within a year, the economic damage runs into trillions — and given TSMC's concentration, there is no alternative routing. The AI development timelines that researchers, policymakers, and corporations are counting on — in drug discovery, climate modeling, materials science — slip by the same duration.

The common mechanism: supply chains optimized for just-in-time delivery have minimal inventory buffers. When capacity drops at a chokepoint, downstream customers running on weeks of inventory start rationing within a month. Recovery is slow: an EUV machine takes 12–18 months to manufacture after order; a new fab takes 3–5 years to build and 2 more to ramp. The Iran war's helium shock is illustrative — even once Ras Laffan restarts, full supply chain recovery is estimated at 4–6 months minimum, and that assumes no secondary disruptions.

Smaller disruptions have real precedents. The 2011 Thailand floods caused a global HDD shortage lasting three years. The 2021 Texas winter storm shut Samsung's Austin fab for six weeks, cascading into the automotive chip crisis. COVID lockdowns in Shanghai in 2022 hit SMIC and dozens of packaging facilities simultaneously.

A disruption to the AI supply chain is not just an economic event. It is a delay in the compounding curve of human technological capability. The discoveries, models, and breakthroughs that would have run on unshipped GPUs — those are the true costs that financial reports almost never capture.

How to track AI chip supply chain risk with AIChipMap

AIChipMap lets you trace any company's full supply chain in one click — seeing upstream dependencies and downstream customers across all 17 tiers simultaneously. Each company page shows risk rating, chokepoint status, export control tier, market share, and a supply chain briefing.

The interactive landing graph highlights chokepoints with a red dashed border. Click any node to focus the graph on that company's direct supply relationships. Follow the trace link to see the full multi-tier dependency graph — upstream and downstream, across every tier the company touches.

The export controls section covers all 10 active regimes — US BIS (2022 and 2023 rounds), ASML/Netherlands EUV controls, Japan's 2023 alignment, entity list actions against Huawei, SMIC, YMTC, and the Chinese gallium and germanium controls — each with current status, affected companies, and links to source regulations.

For investors and analysts: use the trace view to see which companies in NVIDIA's supply chain are simultaneously subject to export controls, or to map second-order exposure to SK Hynix's HBM dominance. For supply chain managers: see which tiers a disruption to a single node touches. For compliance teams: cross-reference entity-listed companies against your supply chain before a shipment is scheduled.

AIChipMap covers companies across the US, Japan, Taiwan, and South Korea — the four countries that collectively define the frontier of the AI chip supply chain. Whether you are an investor in Tokyo, a supply chain analyst in Seoul, a compliance manager in Taipei, or a policy researcher in Washington, the graph shows the same interconnected system from your vantage point.