SemiAnalysis 2 items
‹ All posts People /SemiAnalysis /Discussion ARCHIVE
Discussion · Tue, 22 Sept 2026 · 07:26 ET

Is EMIB taking real CoWoS share — or just relocating the HBM queue?

SemiAnalysis ChipBook: Korea→Malaysia HBM +467% YoY, ~$3B/2mo via Penang. Replies split: customers fleeing CoWoS scarcity vs queue transfer / qualify risk.

top-replies SemiAnalysis IntelTSMC Source ↗
Review after: 2026-12-15

Opening

SemiAnalysis has the better signal — Korea→Malaysia HBM trade flows are a packaging-lane print — but the room has not settled that EMIB is durable CoWoS share. @SemiAnalysis_ argues CoWoS still cannot meet demand and Intel EMIB is getting customer response, citing ChipBook August data: +467% YoY HBM exports Korea→Malaysia, ~$3B in two months, attributed primarily to Intel Penang advanced packaging (~50 Malaysia back-end LinkedIn openings). Visible replies largely accept the trade print while splitting on mechanism: overflow/share shift vs leading-edge qualify, and whether CoWoS still owns the scarce queue. Deep tail unread (top-replies).

Sides

FOR — EMIB is winning real packaging share under CoWoS scarcity

  • @SemiAnalysis_ [named] — CoWoS capacity still short; EMIB a compelling alternative; ChipBook HBM Korea→Malaysia +467% YoY / ~$3B in two months → Penang
  • @Rami_Bball_Fan [anon] — when incumbent cannot scale, customers look; Intel finally has a real product
  • @AI0Ncloud [anon] — trade flow shows up before the keynote

AGAINST / CAUTION — CoWoS still scarce; flows may be queue transfer, not durable EMIB win

  • @aeon_jo [pseudo] — CoWoS still the scarce bit; TSMC 5.5-reticle CoWoS already in production; 14-reticle stack on 2028 roadmap; kwpm still opaque
  • @redboynono [pseudo] — packaging-lane signal, not demand soft patch; CoWoS still owns the queue
  • @JohnMillerFPX [anon] — trade flows measure queue transfer, not delivered accelerator capacity
  • @exitgoose [anon] — Taiwan slowdown suggests share to Malaysia, not fresh demand; Penang leading-edge HBM qualify still open
  • @kingli560043 [anon] — if ~50 back-end openings stall while exports stay hot, inventory relocate not durable EMIB share
  • @TechThought_org [pseudo] — Intel foundry credibility skepticism (CPU redo / bailout history)

Room: tier-1 research shop with trade data vs mostly anon/pseudo mechanism quibbles; no OEM BOM primary in the visible sample.

Quotes

Is EMIB really chipping away at CoWoS?

CoWoS has owned advanced packaging for years. TSMC keeps adding capacity aggressively but it still isn't enough to satisfy customer demand.

Intel stepped in with a compelling alternative with EMIB — and customers are responding. (1/3)🧵 — SemiAnalysis (@SemiAnalysis_)

ChipBook August data caught a +467% YoY jump in HBM exports from Korea to Malaysia, ~$3B in just two months.

We attribute this primarily to Intel's advanced packaging capacity in Penang, with LinkedIn showing ~50 back-end job openings in Malaysia. Not an insignificant number. — SemiAnalysis (@SemiAnalysis_)

CoWoS is still the scarce bit. TSMC’s 2026 symposium: 5.5-reticle CoWoS already in production; a 14-reticle stack (~10 compute dies + 20 HBM) is on the 2028 roadmap — reticle size known, capacity in kwpm still opaque. — JosefEmanuel (@aeon_jo)

ChipBook’s +467% YoY Korea→Malaysia HBM export spike (~$3B in two months) while Taiwan inflows slow is a packaging-lane signal, not a demand soft patch.

CoWoS still owns the queue, but EMIB/Penang back-end capacity showing up in trade flows means L0 advanced packaging is — Steven.Sun (@redboynono)

The Taiwan slowdown in the same data suggests this is share moving to Malaysia, not fresh demand. The open question is whether Penang is qualified for leading edge HBM stacks or just absorbing overflow. That decides whether Intel is winning EMIB or just renting floor space. — exitgoose (@exitgoose)

What would settle it

Continued ChipBook (or customs) HBM Korea→Malaysia export run-rate through Q4 2026 / Q1 2027 alongside stable or rising Taiwan HBM inflows — durable dual-path vs one-time divert Knowable — monthly trade prints / ChipBook updates — through 2026-12-15 earliest checkpoint

Intel or customer disclosure that a named AI ASIC / GPU SKU is in volume EMIB/Penang production (not only overflow assembly) Knowable — earnings, foundry customer remarks, or teardown — through 2027-03-31

TSMC CoWoS capacity guide (kwpm or equivalent) that shows the queue clearing without EMIB share language from customers Published — TSMC quarterly / technology symposium updates — through 2026-12-15

The Open/Close  ·  Research commentary, not investment advice. Positions may be held in securities mentioned.