Silicon Analysts Weekly · Issue #5 · week of 2026-06-292026-07-06

CoWoS-L hits 130k wpm; Nvidia Kyber delayed 12+ months to 2028

Silicon Analysts Weekly for the week of 2026-06-29 to 2026-07-06: CoWoS-L hits 130k wpm; Nvidia Kyber delayed 12+ months to 2028. A sourced, provenance-marked briefing on AI-chip supply-chain and pricing movements — HBM spot prices, wafer pricing, foundry and CoWoS capacity, per-chip bill-of-materials cost, and HBM qualification status — derived from public data and human-reviewed. Part of the Silicon Analysts semiconductor data layer.

The take

CoWoS-L capacity is tracking 130k wafers/month in 2026 versus 105k at end-2025, with JPMorgan projecting 141k in 2027 — the tightest advanced-packaging constraint in the supply chain. Nvidia's Kyber NVL144 rack slip of 12-plus months to 2028 on circuit board challenges removes a near-term CoWoS demand pull-forward but opens space for competing architectures. Combined, the $518B Korea memory buildout and Micron's Japan expansion signal suppliers are betting capacity ahead of demand.

What moved this week

TSMC N16/N12/N28/N40 utilization — flat — 72% vs sub-80% (source)
TSMC N7/N6 utilization — down — 65% vs below 70% (source)
TSMC N5/N4P/N4NP utilization — up — 97% vs 2026-H1 100% booked for H1 2026 (source)
TSMC N3E/N3P/N3B utilization — flat — ~100% vs ~98–100% (fully booked, new project kickoffs suspended) (source)
TSMC CoWoS CoWoS ramp — up — 2026 CAGR of over 60% until at least 2026; more than quadruple from 2023 levels by end of 2026 vs 2026 56% YoY growth in 2026 (source)
TSMC CoWoS CoWoS-L CoWoS capacity — up — 2026 130k wafers per month vs 105k wafers per month by end-2026, rising to 141k wafers per month in 2027 (per JPMorgan) (source)

Supply-chain & pricing signals

Nvidia Kyber rack system delayed 12+ months to 2028 on circuit board manufacturing — Nvidia's next-generation NVL144 rack architecture for 2027 Rubin Ultra chips pushed to 2028 due to unresolved circuit board manufacturing challenges. This creates a technical opening for AMD and Google's custom chips to capture high-end AI infrastructure deals before Nvidia can scale. (CNBC)
Samsung, SK Hynix announce $518B fab buildout in South Korea — Combined 800 trillion won investment in Gwangju and surrounding areas significantly expands HBM and memory production capacity, reducing supply constraints for AI/ML training workloads and supporting margin recovery. Government co-investment of 86+ trillion won signals strong policy backing, enabling sustained capex cycles through 2030s. (Reuters)
SK Hynix $29B US listing signals HBM market strength but raises valuation concerns — SK Hynix's record $29B US IPO validates surging AI-driven HBM demand but Fortune warns the market shows 'extreme speculation' levels. The offering pressure-tests investor appetite for memory chip supply amid questions about demand sustainability from Meta and Apple supply chain shifts. (Bloomberg)
Bernstein raises ASML price target 30% on AI-driven logic and DRAM capacity expansion — ASML's stock up 123% YoY as lithography demand accelerates from unprecedented AI-driven semiconductor capacity buildouts. Analyst upgrade indicates sustained equipment spending cycles for advanced logic and memory manufacturing will persist through 2026-2027. (CNBC)
Micron gains HBM share to 21% as Japan backs $9.3B Hiroshima expansion — Tokyo's ¥1.5 trillion backing of Micron's Japan plant expansion accelerates capacity growth in HBM and DRAM outside Korea, fragmenting supplier concentration. This geographic diversification addresses geopolitical risk but intensifies competition against SK Hynix and Samsung in memory market. (Thestreet)
Anthropic leases 400MW TeraWulf data center in Kentucky for $19B over 20 years — Massive AI infrastructure commitment signals sustained power demand surge and validates data center economics for AI workloads. This locks in 400MW capacity through 2047, creating sustained demand for high-efficiency power delivery and cooling semiconductors used in data centers. (CNBC)

Supply-chain & pricing signals

Nvidia Kyber NVL144 delayed 12+ months to 2028 — circuit board manufacturing challenges push the Rubin Ultra rack architecture out of 2027, creating a gap in high-end AI infrastructure rollout. (CNBC)
Samsung & SK Hynix announce combined $518B Korea fab buildout — 800 trillion won investment in Gwangju and surrounding areas, with 86+ trillion won in government co-investment, targets HBM and DRAM capacity through the 2030s. (Reuters)
Micron reaches 21% HBM share; Japan backs $9.3B Hiroshima expansion — Tokyo's ¥1.5 trillion commitment accelerates capacity outside Korea and intensifies three-way supplier competition. (TheStreet)
SK Hynix $29B US listing proceeds — record offering validates AI-driven HBM demand; analyst commentary flags elevated speculation risk around demand sustainability. (Bloomberg)
Bernstein raises ASML price target 30%; stock up 123% YoY — upgrade cites sustained lithography demand from AI-driven logic and DRAM capacity expansion through 2026–2027. (CNBC)

HBM watch

Samsung × NVIDIA (HBM3E): Qualified Volume shipping — Jensen Huang confirmed all three Korean HBM suppliers passed NVIDIA qualification. (TheLec · timeline)
SK Hynix × NVIDIA (HBM3E): now Volume shipping — included in Jensen Huang's confirmation of all three Korean supplier qualifications. (TheLec · timeline)

Foundry & capacity

TSMC N5/N4P/N4NP utilization — up to ~97%; H1 2026 fully booked.
TSMC N3E/N3P/N3B utilization — flat at ~100%; new project kickoffs suspended.
TSMC CoWoS-L capacity — up to 130k wafers/month in 2026 vs. 105k at end-2025; JPMorgan projects 141k wafers/month in 2027.
TSMC N7/N6 utilization — down to ~65%, remaining below the 70% threshold.

This week's analysis

Terafab, Intel 14A, and the New US Fab Startup Playbook: What the Musk-Intel Alliance Really Signals — Terafab's Intel 14A licensing deal reframes how US fab startups can achieve capital efficiency under the CHIPS Act — and what the exodus of Intel foundry talent means for execution risk. A Silicon Analysts deep-dive.
SK Hynix and Samsung: HBM4 Readiness, 1c DRAM Scaling, and What the 2026 Capacity Race Really Means — SK Hynix is targeting an ~8x ramp in 1c DRAM production while Samsung pursues a ~50% HBM capacity expansion — two very different bets on where AI memory demand goes next. We break down the qualification timelines, capex logic, and supply implications.

AI engines are citing

Hbm Analysis
AMD vs NVIDIA: The AI GPU War in Numbers
NVIDIA GPU Market Share 2024–2026: 87% Peak, What Comes Next

Tool spotlight

Price/Performance Frontier — TFLOPS-per-dollar and TCO across the accelerator landscape.

Sources this week

CNBC
Reuters
Bloomberg
CNBC
Thestreet
CNBC
thelec.net
siliconanalysts.com
trendforce.com
x.com
Get the next issue in your inbox

One short, sourced email a week on the AI-chip supply chain — only when the data moves.

One short email a week — only when the data moves. No marketing, ever. One-click unsubscribe.

Sourced public intelligence, not insider data; status changes are human-reviewed before publication. Full timelines on the HBM Qualification Tracker.