Update: Test Tools +31% vs Packaging +9.6% — Mid-Year Cut Explains CoWoS Rationing

What changed since the Q3 CoWoS tracker

In our Q3 CoWoS vintage we treated early-August packaging trackers as the market’s binding print: year-end CoWoS capacity toward ~140,000 wpm, a supply-demand gap reprinted near 20%, and NVIDIA reservations of 800–850k wafers locking more than half of 2026 CoWoS. That post answered the capacity vs demand question — did packaging ease when capacity rose, or did demand re-accelerate faster? Equipment dollars were carried as background: SEMI Mid-Year still at $165.9B (+23%), DRAM tools +39%, Q1 billings $36.55B as the pace check.

This August cut flips the lens. The newest official OEM stack markets still under-read against that rationing story is SEMI’s Mid-Year Total Equipment Forecast — specifically the back-end asymmetry. Semiconductor test equipment is projected at $15.3 billion in 2026 (+31.0% YoY), a strong upward revision from the prior year-end path. Assembly and packaging equipment rises only 9.6% to $6.7 billion — broadly in line with the prior forecast, and the slowest major segment YoY in the Mid-Year cut. That ~21-point growth gap between test tools and packaging tools sits underneath a CoWoS shortage that the Q3 trackers still print near 20%.

The dashboard is built as that vintage delta: back-end tool dollars versus the CoWoS gap path, Mid-Year segment levels and YoY ranking, bottleneck tightness Δ from Q3 → this Mid-Year reading, plus a Q1→FY pace ladder with the Taiwan/Korea AI-install signature. For the July equipment-cycle map see research; for the earlier HBM/CoWoS frame see AI GPU packaging & memory; for the Q1 billings stamp see the August WWSEMS update.

The headline table: Q3 tracker → Aug Mid-Year back-end cut

The non-obvious cut is not “equipment stopped arriving.” Front-end and DRAM tool dollars still boom. The August news is that official packaging-tool growth is muted relative to test and to the shortage the capacity trackers describe — so a $165.9B equipment year can coexist with CoWoS rationing for another year.

The back-end paradox: test surges, packaging tools lag

SEMI’s Mid-Year OEM perspective (July 14, 2026) is explicit on the split. After surging 55.3% in 2025, test equipment is projected to rise another 31% to $15.3B in 2026 — SEMI’s own language flags a strong upward revision from the 2025 year-end forecast, driven by device complexity and AI/HBM burn-in intensity. Assembly and packaging equipment, after +20.8% in 2025, is only expected to rise 9.6% to $6.7B, “broadly in line with the prior forecast.” By 2028 the path still extends to $20.8B test and $8.6B packaging — growth continues, but packaging tools remain a thin slice of the total $229.5B equipment market.

That is the paradox the dashboard’s composed panel is built to show. Test dollars (area) sit high; packaging dollars (bars) sit low; the CoWoS gap line still prints near 20% across research → mid-print → Q3 → this Mid-Year reading. Markets that only watch WFE headlines will read a broad tool boom. Markets that watch which tool dollars grow will see why advanced packaging capacity can remain rationed even while global billings set records: the segment that buys CoWoS-class tools is not growing at the same rate as the segments that buy front-end and test tools.

Yield and OSAT outsourcing from the Q3 notes still matter on the capacity side — they change effective output per installed tool. They do not rewrite the OEM dollar path. If packaging tool spend only grows ~10% while GPU demand and NVIDIA lock-in keep the gap near 20%, scarcity is a feature of the capital plan, not a temporary miss.

Upstream dollars still boom — DRAM +39%, WFE +23%

None of the back-end asymmetry revises the headline Mid-Year totals. Total semiconductor manufacturing equipment is still $165.9B in 2026 (+23.2% from $135.1B in 2025), with wafer fab equipment at $143.9B (+23.1%). Inside that path, DRAM equipment jumps 39% to $38.8B — still the clearest AI-memory signature in the official OEM print. Toggle the equipment panel: 2025→2026 grouped bars show the level jump; the YoY ranking panel puts DRAM first, test second, total/WFE next, and packaging last among the charted segments.

Q1’s $36.55B global billings (+14% YoY) remain the only hard 2026 quarterly stamp against that path — about 22% of the full-year Mid-Year total. A simple 2×Q1 H1 run-rate of ~$73.1B is ~44% of the FY path — an editorial pace check, not an official Q2 WWSEMS print (still the rewrite the series is waiting on). Taiwan ($8.77B, +24% YoY) and Korea ($8.93B, +16%) still carry the AI install signature; China stays the largest absolute market at $10.99B but cooled 16% QoQ. Treat ×2 and ×4 annualizations as geometry, not forecasts — H2 AI tool intensity is typically stronger than H1.

Bottlenecks: CoWoS stays maxed; test eases; packaging-tools tighten

Editorial tightness scores (1–10 composites, not SEMI metrics) move for the layers the Mid-Year back-end cut actually informs:

• CoWoS packaging: 10 → 10. Gap and NVIDIA concentration dominate; packaging-tool lag reinforces.
• Packaging equipment $: 7 → 8. Official +9.6% is the slowest major YoYscarcity of tools that buy capacity rises as a distinct layer.
• Final test / burn-in: 7 → 6. Official test path +31% to $15.3B eases tool scarcity even if burn-in intensity stays high.
• HBM / advanced DRAM: 9 → 9. Korea’s tool dollars help; HBM remains allocation-gated.
• EUV, OSAT, leading-edge logic: broadly unchanged from the Q3 reading.

The dual-vintage scatter plots tightness against lead weeks with bubble size for supplier concentration. CoWoS still sits in the upper-right. The new geometry is the packaging-tools point drifting tighter while test drifts easier — the Mid-Year dollar split made visible.

What would rewrite this vintage

1. WWSEMS Q2 2026 billings confirm or break the Mid-Year $165.9B path after Q1’s $36.55B stampespecially whether packaging/test mix inside billings matches the OEM forecast asymmetry.
2. SEMI’s next OEM forecast (year-end cycle) lifts assembly & packaging growth materially above ~10%, or cuts the test revision.
3. CoWoS demand trackers revise NVIDIA / ASIC reservations below the 800–850k bandcollapsing the ~20% gap even if packaging tools stay slow.
4. HBM supply from Korea eases allocation faster than packaging, flipping which layer binds first.
5. OSAT process qualification fails to absorb outsourced CoWoS front-end stepstightening TSMC-only scarcity again.

Until those resolve, the shareable framing is narrow: official test tools boom (+31%) while packaging tools crawl (+9.6%), and the CoWoS gap still prints ~20%.

Caveats and methodology

• Official vs tracker. Equipment dollars are SEMI Mid-Year OEM forecast (July 14, 2026) and WWSEMS Q1 billings. CoWoS capacity, gap, and NVIDIA reservation figures are carried from the Q3 tracker synthesis (TrendForce / institutional / TSMC packaging notes) and may revise.
• Segment baselines. 2025 test (~$11.7B) and packaging (~$6.1B) levels are implied from Mid-Year 2026 levels and stated YoY rates; treat as rounded OEM-path geometry.
• H1 run-rate. 2×Q1 is editorial, not an official Q2 WWSEMS release.
• Bottleneck scores are editorial composites for visualizationdirectional, not survey microdata.
• Packaging tools ≠ CoWoS POR. Assembly & packaging equipment is a broad SEMI segment; it is not identical to TSMC CoWoS tool bills alone.
• This post is explanatory data journalism, not investment advice.

Primary sources: SEMI, Mid-Year Total Semiconductor Equipment Forecast – OEM Perspective (July 14, 2026); SEMI, Q1 2026 Equipment Billings $36.55B (June 4, 2026); prior theme posts Q3 CoWoS tracker and August WWSEMS Q1 update.

The shareable takeaway

Versus the Q3 CoWoS tracker, the August Mid-Year back-end cut shows test equipment at $15.3B (+31%) while assembly & packaging tools only rise 9.6% to $6.7B — a ~21-point YoY asymmetry under a CoWoS gap that still prints ~20% with NVIDIA locking >50%. SEMI’s total path stays $165.9B with DRAM tools +39%; Q1’s $36.55B remains the pace check. Upstream dollars fund the stack; slow packaging-tool growth helps explain why CoWoS still gates the ship.