What Changed — Semiconductor Data Movements

Recent movements in semiconductor wafer pricing, AI accelerator manufacturing cost, foundry gross margins, foundry capacity, and NRE/mask-set cost, derived daily from an append-only data ledger. Each movement shows direction, magnitude, the two dates compared, and a confidence tier. Public sources only — no proprietary or insider data.

What Changed in Semiconductor Data

Recent movements across wafer pricing, AI accelerator cost, foundry margins, capacity, and NRE — tracked daily from our public-data ledger, every figure with provenance.

Comparing Aug 20, 2026 vs Aug 13, 2026 (7 days of history) ·

Public sources only — no proprietary or insider data

HBM market share4

View dataset →
MetricChangeMove
Micron HBM HBM bit share20.5% 5% -15.5pp
SK Hynix HBM HBM bit share62% 56% -6pp
Samsung HBM HBM bit share17% 21% +4pp
SK Hynix HBM HBM bit share62% 58% -4pp

AI accelerator cost (BOM)20

View dataset →
MetricChangeMove
MI325X HBM cost$2,200 $4,350 +97.7%
H200 HBM cost$1,500 $2,400 +60%
MI325X est. mfg cost$3,800 $5,950 +56.6%
H200 est. mfg cost$4,250 $5,150 +21.2%
MI355X HBM cost$4,350 $4,900 +12.6%
GB200 HBM cost$5,800 $6,500 +12.1%
B200 HBM cost$2,900 $3,250 +12.1%
B100 HBM cost$2,900 $3,250 +12.1%
MI325X gross margin81% 70.3% -10.7pp
MI355X est. mfg cost$8,000 $8,550 +6.9%
B200 est. mfg cost$6,400 $6,750 +5.5%
B100 est. mfg cost$6,500 $6,850 +5.4%
GB200 est. mfg cost$13,500 $14,200 +5.2%
H100 SXM5 est. sell price$28,000 $27,000 -3.6%
H200 gross margin88.8% 86.4% -2.4pp
MI355X gross margin68% 65.8% -2.2pp
B100 gross margin79.7% 78.6% -1.1pp
GB200 gross margin79.2% 78.2% -1pp
B200 gross margin84% 83.1% -0.9pp
H100 SXM5 gross margin88.1% 87.7% -0.4pp

GPU secondary market1

MetricChangeMove
H100 SXM 80GB residual value68.8% 71.3% +2.5pp

GPU cloud rental pricing2

View dataset →
MetricChangeMove
RUNPOD H200 on-demand $/GPU-hr$2.60/GPU-hr $3.59/GPU-hr +38.1%
RUNPOD H100 SXM5 on-demand $/GPU-hr$2.69/GPU-hr $2.99/GPU-hr +11.2%

Fab capacity & utilization1

View dataset →
MetricChangeMove
TSMC CoWoS CoWoS capacity52.5 kwspm 72 kwspm +37.1%

Foundry capacity1

View dataset →
MetricChangeMove
TSMC N3 annual wafer starts2,160,000 wafers_per_year 2,100,000 wafers_per_year -2.8%

Gross margins1

MetricChangeMove
MAINSTREAM CPU gross margin47.2% 54.75% +7.5pp

Chip archetype defaults2

MetricChangeMove
GPU AI hbm cost per gb15 usd_per_gb 17 usd_per_gb +13.3%
GPU AI backend yield92% 95% +3pp

Download the full historical ledger (Pro)

Every figure here is a dated, append-only series. Get the complete history as CSV/JSONL for backtesting and models — join the list and we'll reach out.

Market intelligence behind the moves

All market intelligence →
HighPackagingAug 18, 2026

TSMC CoWoS 5.5x Reticle Hits 98-99% Yield in Volume Production; 14x Roadmap Confirmed to 2029

Yield: CoWoS 5.5x reticle confirmed at 98-99% in volume production (positive); Roadmap: 14x reticle-size CoWoS targeted by 2029; SoIC pitch: 6 microns currently in HVM, targeting 4.5 microns by 2029.

HighPackagingAug 11, 2026

TSMC CoWoS Yield Confirmed at 98-99% in HVM; Nvidia Weighs 8-Stack HBM4E Fallback for Rubin Ultra

CoWoS yield: 98-99% confirmed in HVM (positive); HBM4E 12-stack yield risk flagged — potential stack-count reduction from 12 to 8 layers (est. HBM capacity-per-die reduction of ~33% if downgrade confirmed); TSMC 2026 CapEx raised to $60-64B record (capacity positive).

HighMemoryAug 10, 2026

Samsung HBM4 Hits 80% 'Golden Yield' — Four Months Ahead of Schedule, Supply Tightness Set to Ease

Yield improvement: Samsung HBM4 from <60% (Feb 2026 launch) to ~80% (Aug 2026) — approx. +20 pp in ~6 months; SK Hynix HBM4 yield also reported at ~80%; HBM4E target set at 70% pre-launch (H1 2027). 2027 HBM supply for all three vendors reported as fully booked.

HighMemoryAug 8, 2026

SK Hynix 54 Trillion Won CapEx Locked In; Nvidia Rubin Ultra Downgrading HBM Stack Amid Structural Supply Crunch

CapEx: +54 trillion KRW (~$39B) across Y2 and M17 fabs; HBM capacity additions: Y1 initial production February 2027, M17 December 2028, Y2 June 2029 (plus 12–18 months yield ramp each); Contract DRAM prices: +90–95% QoQ in Q1 2026 (largest single-quarter spike on record per TechTimes); Nvidia Rubin Ultra HBM stack: downgraded from 16-stack HBM4E to 12-stack or 8-stack variants under evaluation; Lam Research advanced packaging revenue outlook raised from >40% to >70% YoY growth in current fiscal year.

Recent corrections

Full changelog →
h200chip_costannotation on frozen history2026-08-16
21752400affected 2026-08-16
b100chip_costannotation on frozen history2026-08-16
29003250affected 2026-08-16
b200chip_costannotation on frozen history2026-08-16
29003250affected 2026-08-16
gb200chip_costannotation on frozen history2026-08-16
58006500affected 2026-08-16
mi325xchip_costannotation on frozen history2026-08-16
38674350affected 2026-08-16

Get alerted when this data moves

Watch the semiconductor data you care about — we'll email you when it moves. Public sources only.

Select none to watch everything.

Methodology

Every day we freeze the canonical semiconductor cost/pricing data into an append-only ledger. This page compares the most recent snapshot against the closest one ~7 days earlier and surfaces the metrics that moved past a per-dataset significance threshold (filtering out rounding noise).

Each movement shows direction, magnitude (percent for value metrics, percentage-points for rate metrics), the two dates compared, and a confidence tier. The trend column sketches each metric's full ledger history (up to ~90 days). The history is young — the ledger began 2026-06-06, and some domains are tracked from later dates — so a 30-day window automatically clamps to the data available and labels the real lookback, and per-row trends can start on different dates. Underlying values are synthesized from public sources only.

Programmatic access: /api/v1/changes and the get_recent_changes MCP tool. For data provenance and terms, see Terms & Data Provenance.

See current wafer pricing →

300mm wafer prices by node and foundry, with per-row provenance and price ranges.