Charted: Semiconductor Equipment Hits $166B in 2026 — While CoWoS Still Gates AI GPUs
SEMI puts total equipment at $165.9B in 2026 (+23%) after Taiwan’s +90% 2025 surge. DRAM tools jump 39%, but the binding AI constraint remains packaging and HBM — not another front-end wafer line.
The chip stack has two clocks
Semiconductor equipment dollars and AI GPU shipments are both booming — but they are not the same clock. SEMI’s Mid-Year OEM Forecast (July 2026) puts total manufacturing equipment sales at $165.9 billion in 2026, up 23.2% from the record $135.1 billion billed in 2025. Wafer fab equipment (WFE) alone is projected at $143.9 billion. By 2028, SEMI sees the total equipment market near $229.5 billion.
That is the upstream investment clock: etch, deposition, lithography, test, and assembly tools ordered years before wafers leave the fab. The downstream shipping clock is still paced by HBM allocations and CoWoS slots. Our earlier HBM & CoWoS bottleneck map showed packaging sold out into 2026; this research post zooms out to the full equipment cycle and regional concentration that fund — and still fail to fully relieve — those bottlenecks.
The dashboard above toggles five panels: the equipment cycle, regional billings, segment growth, a multi-layer bottleneck scatter, and CoWoS capacity vs demand.
$135B became $166B — five years of growth, not a one-year spike
SEMI’s April 2026 billings print confirmed $135.1 billion of global equipment sales in 2025, up 15% from $117.1 billion in 2024. That was already a record. The July mid-year forecast then lifted the 2026 outlook to $165.9 billion — a further +23.2% — and extended the runway through 2028 at $229.5 billion, implying five consecutive growth years as AI demand reshapes capital plans.
| Year | Total equipment (SEMI) | WFE | Status |
|---|---|---|---|
| 2024 | $117.1B | ~$95B (est.) | Actual |
| 2025 | $135.1B | $116.9B | Actual |
| 2026 | $165.9B | $143.9B | Forecast |
| 2027 | ~$197B | ~$175B | Forecast path |
| 2028 | $229.5B | ~$200B | Forecast |
WFE growth is not evenly distributed across process modules. SEMI explicitly ties the upward revision to advanced memory — especially HBM-related DRAM — and leading-edge logic. DRAM equipment sales are projected to rise 39% to $38.8 billion in 2026, then keep climbing toward $56.9 billion by 2028. NAND tools grow 30.7% to $13.9 billion in 2026 on 3D layer migration. Front-end tools fund the wafers; they do not automatically create CoWoS interposers or HBM stacks.
Taiwan’s +90% is the AI signature in the regional ledger
The 2025 regional billings table is the cleanest public fingerprint of where AI capacity is being installed:
| Region | 2025 billings | YoY |
|---|---|---|
| China | $49.3B | −0.5% |
| Taiwan | $31.5B | +90% |
| Korea | $25.8B | +26% |
| N. America | $10.9B | −20% |
| Japan | $9.5B | +22% |
| Europe | $2.9B | −41% |
| Rest of World | $5.2B | +25% |
China remains the largest single market near its prior record, but growth has stalled. Taiwan nearly doubled as TSMC ramped 2nm and CoWoS. Korea advanced on HBM/DRAM programs at Samsung and SK hynix. North America and Europe cooled after earlier CHIPS-era and automotive cycles. East Asia (China + Taiwan + Korea + Japan) still accounts for roughly four-fifths of 2025 equipment dollars — the geographic concentration that makes export controls, earthquake risk, and single-island packaging capacity macro variables rather than supply-chain footnotes.
Toggle the dashboard’s Regions panel between 2024, 2025, and a directional 2026 share view scaled to SEMI’s $165.9B total. Treat 2026 regional bars as illustrative, not an official SEMI regional release.
Back-end intensity: test +55%, assembly +21%
SEMI’s 2025 billings detail matters for the AI thesis. Test equipment surged 55% as AI devices and HBM raised performance and burn-in intensity. Assembly and packaging equipment rose 21% as advanced packaging adoption accelerated. Those are the tool categories closest to the shipping constraint: more EUV scanners help logic density; more handlers, probe cards, and bonding tools help when every Blackwell-class accelerator needs a package that actually exists.
This is why equipment-cycle bullishness and packaging tightness can coexist. Tool billings are orders and installations. GPU revenue is finished modules. Between them sit substrates, HBM stacks, CoWoS interposers, and final test capacity — layers that are smaller in dollar terms than WFE but higher in tightness.
Where the stack still binds
The bottleneck panel scores seven layers on tightness (1–10), lead time (weeks), and supplier concentration:
- CoWoS packagingtightness 10; ~52-week lead times; TSMC still the production-proven gate for most AI GPUs.
- HBM / advanced DRAMtightness 9; SK hynix near 52% share; sold-out narratives into 2026.
- EUV lithographytightness 8; ASML remains the sole HVM EUV supplier as High-NA ramps.
- Final test / burn-intightness 8; consistent with SEMI’s +55% test billings print.
- OSAT advanced package and ABF/substratestightness ~7; overflow and materials buffers, not full substitutes for CoWoS.
- Leading-edge logic fabtightness ~6; still concentrated, but less binding than packaging for 2025–26 AI shipments.
CoWoS capacity trackers (aligned with our packaging research) show roughly 37.5k → 75k → 125k wafers per month from end-2024 through the 2026 target, while annualized demand estimates still run ahead of installed capacity. Dollars authorized for front-end tools do not erase that gap on the same schedule — a theme that also shows up when hyperscaler AI capex guidance races ahead of energized megawatts.
How to read equipment headlines without mixing scopes
Use this checklist when a slide claims “the chip shortage is over” or “semicap is the AI trade”:
- Equipment vs accelerators. $166B is OEM tool salesnot NVIDIA revenue, not CoWoS slots.
- WFE vs back-end. WFE is ~87% of the 2026 total in SEMI’s split; the binding AI constraint often sits in the other ~13% plus memory specialty tools.
- Region vs company. Taiwan’s +90% is a TSMC-centric AI build signature; China’s flat $49B is a different industrial policy story.
- Forecast vs billings. 2025 figures are WWSEMS billings; 2026–28 are OEM forecast perspectives and can revise.
- Tightness vs spend. Rising DRAM equipment (+39%) is the industry’s attempt to relieve HBM; it is not proof the bottleneck has cleared.
Caveats and methodology
- SEMI totals are OEM semiconductor manufacturing equipment sales (billings / forecasts). They exclude used tools and many materials.
- 2024 WFE (~$95B) in the cycle chart is an estimate consistent with the 2025 WFE print; treat it as softer than SEMI’s disclosed 2025–28 WFE path.
- 2026 regional bars are editorial shares scaled to the $165.9B total for interactionnot a SEMI regional forecast table.
- Test / assembly index series use SEMI’s 2025 growth rates as anchors; forward years are illustrative continuations for the dashboard.
- Bottleneck scores are editorial composites for visualization, informed by public lead-time and concentration reportingnot SEMI statistics.
- CoWoS capacity/demand figures are industry-tracker midpoints (TrendForce / TSMC disclosure synthesis) and may revise.
- This post is explanatory data journalism, not investment advice.
Primary sources: SEMI, Mid-Year Total Semiconductor Equipment Forecast – OEM Perspective (July 14, 2026); SEMI, 2025 Equipment Billings $135.1B (April 7, 2026); TrendForce / TSMC packaging disclosures for CoWoS & HBM context.