Charted: China + Taiwan + Korea Clear 79% of Chip-Tool Billings — Taiwan Gates the AI Stack

Our July research map answered where the equipment cycle and stack bottlenecks bind when tool dollars boom. The concentration lens then measured how thick the vendor tip is at each layer — EUV at 100%, leading-edge foundry near 90%, CoWoS near 78%, median top-1 around 72%. The Q3 CoWoS tracker and August Mid-Year back-end cut moved the vintage: capacity toward ~140k wpm, a gap that still reprints near 20%, and assembly & packaging tools growing only +9.6% while test equipment prints +31%. This post asks the desk question those vintages imply but do not draw as a map: where does activity, risk, and capacity sit geographically — and do tool-install geography, wafer-seat geography, and design-HQ geography agree?

The interactive dashboard above is built as a geography lens. Toggle Equipment, Stack geo, Stages, and Meters. On equipment flip share versus billings dollars; on stack geo filter by stage and tip (≥70%). The punchline is deliberately multi-map. On equipment billings, China leads at about 36.5% of SEMI’s 2025 WWSEMS print, and China + Taiwan + Korea clear ~79%. On leading-edge foundry, Taiwan alone is ~90%. On CoWoS-class packaging, Taiwan is still ~78%. On HBM, Korea’s two memory majors clear ~88%. On AI GPU design / revenue HQ, the United States is ~88%. Tool dollars, wafer seats, packaging gates, and design HQs are four different maps.

The headline ladder: regional and country shares that disagree

Read the table as a family of maps, not one slogan. Equipment geography tells you which regions absorb tool dollars. Foundry and CoWoS geography tell you which island gates the AI wafer and package. HBM geography tells you which memory corridor binds GPU ship schedules. Design-HQ geography tells you which corporate domicile books the accelerator revenue. Averaging these rows into “the chip stack is Chinese” or “the chip stack is Taiwanese” or “the chip stack is American” is a category error — all three can be true on different ledgers.

Equipment: China leads dollars; Taiwan prints the AI-install signature

Open Equipment. Against a $135.1B 2025 WWSEMS perimeter, China rolls to about $49.3B (~36.5%), Taiwan $31.5B (~23.3%), Korea $25.8B (~19.1%), N. America $10.9B (~8.1%), Japan $9.5B (~7.0%), Europe only $2.9B (~2.1%), and RoW the residual. Flip the metric to Billings $B and the same ladder appears in absolute dollars — useful when desks need a spend envelope rather than a share slide.

The equipment pie is therefore China-weighted without being a monopoly. A system where one country holds a little over a third of global tool billings still leaves Taiwan + Korea with a combined tip larger than China alone on AI-relevant advanced installs. That is the geographic twin of the concentration scoreboard: ASML remains the sole EUV vendor; rolling China, Taiwan, and Korea onto the same install sheet makes East Asia the thickest region without claiming that “most leading-edge wafers are Chinese.”

The stacked path from 2022→2025 is an editorial East Asia trio share track, not a new SEMI microdata extract. China’s share drifts from the high-20s toward ~36.5% as Taiwan and Korea hold the low-20s / high-teens. Treat the path as a shape check against the live 2025 snapshot — useful for seeing that China’s install lead thickened without flipping Taiwan’s AI-gate role.

Q1 2026 billings at $36.55B (+14%) remain the pace check against SEMI’s Mid-Year $165.9B path. Taiwan’s +24% and Korea’s +16% in that quarterly print are the AI-install signature the research map already flagged — geography of growth, not just geography of level.

Stack geo: Taiwan dual-gates the midstream; Korea owns HBM

Switch to Stack geo. Eight layers mapped to country tips rather than vendor names alone. EUV is Netherlands 100% (ASML). Leading-edge foundry is Taiwan ~90% (TSMC) with Korea ~8% and the US a thin residual. CoWoS-class AI packaging is Taiwan ~78%, with US OSAT / Amkor geography and Korea filling the next seats. HBM rolls to Korea ~88% (SK Hynix + Samsung) versus Micron’s US ~12%. AI GPU design HQ is United States ~88%. EDA is US-heavy (~72% for the Synopsys–Cadence tip). 300 mm wafers tip toward Japan ~56%. WFE OEM HQs split across US / NL / Japan — a reminder that who builds tools is not where tools install.

Toggle Stage to Midstream and the geographic choke sharpens: foundry LE, CoWoS, and HBM are almost entirely an East Asia story. Toggle Top-1 ≥70% to hide the softer tips and keep only the extreme gates — EUV, foundry LE, CoWoS, GPU HQ, and the Korea HBM roll-up. The share × risk scatter puts those same layers in the upper-right: high country tip, high editorial risk score. That is the geography version of the vendor concentration ladder — same bottlenecks, different axis.

For the packaging gate that sits beside HBM, keep the theme’s GPU packaging & memory bottleneck frame and the Q3 CoWoS tracker open: capacity can rise toward ~140k wpm while the supply-demand gap reprints near 20% when demand re-accelerates and NVIDIA reservations lock more than half of the pool — a buyer concentration sitting on top of a Taiwan capacity geography.

Stages: midstream East Asia ~92%; upstream and downstream disagree

Open Stages. The stage flip is the post’s operational summary. Upstream (tools, materials, EDA) still has heavy OEM / IP weight in the United States, Netherlands, and Japan — East Asia’s stage share is only mid-20s even though East Asia absorbs most installs. Midstream (foundry, HBM, CoWoS) is the geographic choke: East Asia clears roughly ~92% of the composite tip, with Taiwan dual-gating foundry LE and CoWoS while Korea owns HBM. Downstream (AI GPU design / revenue) flips again: United States ~88%, East Asia a thin residual.

That midstream concentration is why “friend-shoring the stack” and “diversifying tool installs” are not the same policy sentence. CHIPS-era dollars can thicken N. America equipment share (still only ~8% of 2025 billings) without moving Taiwan’s leading-edge or CoWoS tip in a single vintage. The August Mid-Year update already showed assembly & packaging tools growing only +9.6% — slow packaging-tool growth helps explain why the CoWoS geography stays sticky even when total equipment rides a $165.9B path.

Country risk seats make the same point as a scoreboard. Taiwan gates two critical layers at extreme tips. Netherlands gates EUV at 100% while Europe’s install share is only ~2% — IP geography ≠ fab geography. Korea gates HBM. The United States gates GPU design and most commercial EDA, with thin advanced-wafer seats. China leads equipment dollars without leading the EUV / foundry LE / CoWoS / HBM tips that bind AI ship schedules.

Meters: five geography prints, one category error to avoid

Toggle Meters. Six geography meters sit side by side: China equipment tip ~36.5%, East Asia trio ~78.9%, Taiwan foundry LE ~90%, Taiwan CoWoS ~78%, Korea HBM ~88%, US GPU HQ ~88%. Desks that average these into a single “semiconductor geography is concentrated in X” sentence will mis-price export-control exposure, earthquake / strait risk, memory-cycle shocks, and design-margin competition in the same week.

The August 608 concentration cut and Q3 concentration refresh already asked how thick the vendor tip is. Geography does not replace those vintages; it asks which map a desk is actually using when it says “the supply chain is concentrated.” Vendor concentration and geographic concentration often rhyme — TSMC is Taiwan for leading-edge; ASML is the Netherlands for EUV — but they diverge on China (install leader, not LE gate) and on GPU design (US HQ, Asia wafers).

Caveats and methodology

• Equipment region shares are SEMI 2025 WWSEMS billingsdisclosed regional totals. Q1 2026 billings are a pace check, not a rewrite of the 2025 level map.
• Stack-layer country seats are editorial geography carries from theme concentration bands (foundry LE, CoWoS, HBM, EUV, GPU HQ). They are not a single SEMI census table; confidence is flagged per layer in the data module.
• HBM “Korea ~88%” rolls SK Hynix + Samsung as a country tip; vendor concentration still shows SK Hynix alone near the low-50s in recent theme prints.
• Risk scores (1–10) are desk composites of single-country tip × replaceabilitynot insurance ratings or official stress tests.
• Stage East Asia shares average the midstream layers’ geographic tips; they are composites for visualization, not audited regional GDP of chips.
• 2022–2025 equipment path is an editorial shape check, not a new panel dataset. Do not splice it into WWSEMS point-in-time rankings as if they shared a methodology.
• Design HQ ≠ wafer location. US GPU revenue geography does not mean US wafer capacity; Taiwan foundry geography does not mean Taiwan books the accelerator ASP.

What to watch next

Watch three geography meters, not one. First, whether CHIPS-era and Japan/Europe fab programs move N. America / Europe equipment share without denting Taiwan’s foundry LE and CoWoS tips — install diversification versus gate diversification. Second, whether Korea’s HBM tip softens as Micron ramps, or whether the country tip stays near ~80%+ even if vendor shares reshuffle. Third, whether Mid-Year packaging-tool growth stays in the high single digits while CoWoS demand reprints a ~20% gap — the flow story that keeps Taiwan’s packaging geography binding. For vendor top-1 / HHI keep the concentration series open; for capacity versus demand keep the Q3 CoWoS tracker open; for the equipment-cycle map keep research open.

Primary sources: SEMI Mid-Year OEM Forecast (July 2026) and WWSEMS 2025 billings; TrendForce / company disclosures for HBM and CoWoS location bands; theme posts research, concentration, Q3 concentration, August 608 concentration, Q3 CoWoS tracker, August Mid-Year update, and packaging & memory.