Does CoWoS 4× capacity expansion translate to a BESI order-backlog multi-year growth cycle?

Surfaced by /prospect-chains (move: second-order beneficiary off hbm-cowos-as-binding-bottleneck). The forcing function — CoWoS sold-out through 2027 and TSMC expanding 4× — is confirmed and cited; the beneficiary leg (BESI revenue/margin compounding over 2026–2027) is the un-verified gap this hypothesis exists to research.

The chain

TSMC's CoWoS advanced-packaging capacity is 50%+ consumed by Nvidia alone; the 15–20% persistent supply-demand gap is forcing TSMC to expand CoWoS from ~35K WPM (late 2024) to ~130K WPM (end 2026) — nearly 4×. Layer 3 street path: 140k → 220k wpm; ASE takes 2026 overflow; Amkor Arizona is 2028. Expansion is absorbed by demand rather than easing the constraint through 2027+. (From hbm-cowos-as-binding-bottleneck · 2026-06-01-autoresearch-hbm-cowos-bottleneck-2026-corroboration · amkor-cowos-tsmc-overflow-osat)
BESI (BE Semiconductor Industries) is a dominant hybrid-bonding and CoWoS-like (flip-chip / TCB / oS) tool supplier. Q1 2026 orders more than doubled YoY to €269.7M. ⚠ The May claim that "each incremental CoWoS tool line requires BESI hybrid bonders" is too tight — see 2026-08-17 update. (From 2026-05-19-autoresearch-cowos-packaging-capacity-may-2026-update · 2026-08-17-autoresearch-besi-hybrid-bonding-backlog)
→ besi commands a structural orders backlog through 2026 and into the HBM4/HBM4E hybrid bonding wave (2027–2030). (⚠ HBM hybrid bonding is still evaluation, not 2026 volume. €400M backlog was an analyst calc, not a BESI disclosure.)

Why it matters

The hbm-cowos-as-binding-bottleneck mechanism names TSMC and HBM suppliers (SK Hynix, Samsung, Micron) as beneficiaries, but the equipment supplier enabling TSMC's 4× expansion is omitted from the chain. BESI is the purest picks-and-shovels play on the CoWoS expansion cycle itself — it captures revenue whether Nvidia, AMD, Google TPU, or Broadcom wins the packaging allocation. The ~6 month tool lead time (100 nm hybrid, Q2-26) means current orders lock in near-term revenue, and the HBM4 hybrid bonding wave (2027+) would be a second independent growth leg if it graduates from evaluation.

Tradeable: BESI (Euronext Amsterdam: BESI, US OTC: BESIY). Primary listing is Euronext Amsterdam; US investors access via BESIY ADR (thinner liquidity). Twelve Data free plan cannot price BESI.

Why it may not work

Identity gap (narrowed, not closed): CoWoS 4× shows up in BESI as CoWoS-like + hybrid overlay, not as hybrid-bonder units per incremental wpm. Units undisclosed.
Already priced in: Q1 2026 orders >2× YoY was public in April; Q2 confirmed continuation. Entry vs. fundamental thesis is a separate question.
TSMC tooling internalization: Process capacity is being internalized (Layer 3). Tools are being bought from BESI (AP6 ~60, AP7 filling). Those are different claims — the in-house-bonder risk did not fire on this pass.
ASE overflow ≠ BESI HB: 2026 CoW residual looks ase; fetched equipment talks name Korean TCB/dicing suppliers.
HBM HB delay: JEDEC height relief + cost/yield can keep TCB as the HBM volume process through 2026–27.
Cycle length: Blickman will not call a super-cycle; typical up-cycle 6–8 quarters.
Euronext liquidity: BESIY OTC has thin US liquidity.

Update (2026-08-17) — orders compounded; the identity is wrong. Do not graduate

From 2026-08-17-autoresearch-besi-hybrid-bonding-backlog. Status hypothesis → narrowed.

Q2-26 beat and kept booking. Revenue €249.9M (+35.2% QoQ vs +30–40% guide); orders €292.9M (+8.6% QoQ, +128.8% YoY); LTM orders record €987.6M; H1 orders €562.6M (+116.5%). Q3 guided +10–15% sequential. Hybrid-bonding customers 15 → 21. This is not a one-quarter surge that then digested.
€400M backlog was never official. van Putten (MS) calculated ~€400M at 2026-03-31; BESI does not disclose backlog or units. Do not treat the May entity/question figure as first-party.
CoWoS-like ≠ hybrid bonding. Blickman: CoWoS-like is already "a significant part of the orders" (Q4/Q1/Q2) and "increasing penetration of CoWoS, CoPoS" means flip-chip and TCB and hybrid. Q2 HB orders were CPO / logic / (probably) wearables — not HBM.
TSMC is buying BESI tools. AP6 ~60 bonders; AP7 (TSMC Chiayi) filling with first systems; published capacity ~double AP6 / "anticipated 100… may be significantly higher." Blickman sees no SoIC-for-CoWoS slowdown. Applied 9% is a partner (integrated D2W line), not a substitute.
ASE 2026 overflow is a different tool basket. TrendForce/ETNews CoW POs: Korean dicing/bonding suppliers. Do not equate Layer 3's ASE residual with a BESI hybrid-bonding book.
HBM hybrid bonding still unverified. Evaluations at 1 of 3 memory makers, another starting; cost/yield; >3 years for Taiwan logic. 2027–2030 remains Blickman's horizon language. Gate: 2026-10-22 Q3 print.
Cycle caution is first-party. Blickman: 6–8 quarter typical up-cycle; "this time is different" has been heard before.

2026 answer: CoWoS 4× does show up in BESI's order book as a multi-quarter growth cycle (Q4→Q1→Q2, Q3 guided up). It does not translate 1:1 into a hybrid-bonding backlog identity, an official €400M backlog, or a booked HBM4 wave. TSMC in-house tools did not fire. Do not graduate.

Open questions remaining

Hybrid-bonder units shipped/booked in 2026 (BESI refuses to disclose). Without units there is no bridge from 140k→220k wpm to BESI.
What share of Q2/H1 orders is CoWoS-like flip-chip/TCB/oS vs hybrid bonding vs photonics vs mobile?
Does the possible H2 AP7 second round land, and at what tool count vs the "100, maybe significantly higher" Q1 color?
HBM hybrid-bonding volume: which of the three memory makers, which generation (HBM3 / 4E / 5), and is it 2026 eval or 2027 production? Gate: 2026-10-22.
Any first-party ASE/Amkor PO that names BESI (vs Hanmi/Avaco/Wonik) for 2026 CoW lines?
Already priced in? This pass does not re-underwrite BESI.AS / BESIY multiples.

What evidence would graduate this to an active thesis

BESI discloses hybrid-bonder units, or a first-party TSMC/OSAT split that maps CoWoS wpm to BESI tools
OR: a memory-maker first-party statement that HBM4E/HBM5 hybrid bonding is in production with BESI as POR
Remaining: official backlog (BESI still will not print one)
Remaining: ASE/Amkor 2026 CoW lines actually book BESI, not only Korean TCB

Does CoWoS 4× capacity expansion translate to a BESI order-backlog multi-year growth cycle?

Does CoWoS 4× capacity expansion translate to a BESI order-backlog multi-year growth cycle?

The chain

Why it matters

Why it may not work

Update (2026-08-17) — orders compounded; the identity is wrong. Do not graduate

Open questions remaining

What evidence would graduate this to an active thesis

Sources

Related