SK hynix Reports HBM4 Yield and Revenue
SK hynix Reports HBM4 Yield and Revenue
Conclusion: The reported HBM4 yield above 70% is a meaningful signal that SK hynix's HBM4 line has moved past early-stage instability. It is not yet proof of production readiness at scale, and it does not clearly separate SK hynix from Samsung. The yield figure is unofficial, its measurement basis is unknown, and the key open items are 16-layer qualification, Nvidia's Rubin spec, and whether the yield holds across batches. The 2.67% drop is not explained by this news: the yield reports date from November 11–13, and I found no source tying the decline to them.
Data as of: SKHY regular-session close on Thursday, 2026-11-19 16:00 ET (2026-11-19 21:00 UTC). Price , change from a previous close of . Quote timestamp . The ticker resolves to SK hynix on NASDAQ.
Why the yield matters
- Yield is the real production bottleneck. HBM4 is harder to make than earlier generations because stacking defects scrap the whole stack, so early yield largely determines cost and output. Chosun Biz reports that moving from 60% to 70% yields about 16.7% more good units from the same input (Chosun Biz, Nov 12). That arithmetic is correct, but the 60% starting point is also unverified.
- Production is already real. SK hynix said on its Q3 call that HBM4 moved into full-scale mass production and expanded shipments from Q3, and that DRAM and HBM volumes are sold out through 2029 (Seoul Economic Daily, Nov 17). Revenue from Nvidia was in the first nine months, roughly double the prior year's , so the milestone is already showing up in revenue.
- Company-level figures are more conservative. Korea Standard reports that SK hynix said earlier in November that its cumulative HBM4 yield was in the mid-60s% and the best in the industry (Korea Standard). The 70%+ figure appears to be a recent batch or stage-specific number. On November 19, SK hynix declined to confirm specific yield data, and an industry analyst cited by 10jqka said the 70% is a channel estimate that likely reflects best-batch or single-stage numbers rather than steady-state line yield (10jqka/财闻, Nov 19).
Competitive read
- Samsung is the comparison point. Reports from September put its HBM4 yield near 80%, but those figures use different bases and timing, so they are not directly comparable (Chosun Biz). Samsung also began HBM4 deliveries in February and is pursuing hybrid copper bonding for 16-layer stacks.
- SK hynix's competitive edge is Nvidia qualification and volume. Its customer base is also broadening: Nvidia's share of revenue fell to in the first nine months, versus 26.98% for all of last year. Those share figures come from the same Seoul Economic Daily source.
- Spending is rising sharply. Capex was in the first nine months, with about ₩40 trillion expected for the year, and R&D is on pace for about ₩12 trillion.
Remaining hurdles
- Rubin spec. Nvidia's Vera Rubin baseline is 12-layer, 36GB HBM4. Korean suppliers are qualifying 16-layer, 48GB stacks, with qualification closing on December 31, 2026 (East Asia Brief, Oct 11). A 70% yield on 12-layer parts says little about 16-layer economics.
- Thermal and packaging limits. The 16-layer stack must fit under a 775-micron JEDEC height limit, which requires thinning dies below 50 microns. Heat concentrates in the center of the stack. Both suppliers have deferred hybrid bonding to HBM4E or HBM5.
- Batch consistency. A single batch yield is not steady-state yield. The key evidence will be whether the company or customers confirm yield improvement over several quarters.
- Pricing. Yield gains matter only if they convert into margin. Watch whether pricing holds as Samsung scales.
What to watch
- Any official SK hynix disclosure of HBM4 yield or a consistent basis for it, such as on the Q4 call.
- Nvidia's 16-layer qualification outcome around the December 31, 2026 deadline, and whether the Rubin baseline changes.
- Samsung's HBM4 share and yield claims on a comparable basis.
- Quarterly revenue from Nvidia and other top customers, and whether capex stays near ₩40 trillion.
- Whether the November 19 decline reflects company-specific or sector news. I did not find a source that explains it.
Sources
- Driven quote data: SKHY (NASDAQ), close on 2026-11-19 ()
- Chosun Biz, "SK hynix tops 70% HBM4 yield," Nov 12, 2026: https://biz.chosun.com/en/en-it/2026/11/12/7XTICBS4V5BJPD4IYCNFKP7DWE/
- Seoul Economic Daily, "SK hynix Nvidia Sales Double to 35 Trillion Won on HBM4 Ramp," Nov 17, 2026: https://en.sedaily.com/business/2026/11/17/sk-hynix-nvidia-sales-double-to-35-trillion-won-on-hbm4-ramp
- East Asia Brief, "SK hynix and Samsung push 16-layer HBM4 as Nvidia caps Rubin at 12 dies," Oct 11, 2026: https://eastasiabrief.com/semiconductors/sk-hynix-samsung-push-16-layer-hbm4-nvidia-caps-647
- Korea Standard, Nov 13, 2026: https://www.koreastandard.news/article/sk%ED%95%98%EC%9D%B4%EB%8B%89%EC%8A%A4-6%EC%84%B8%EB%8C%80-hbm4-%EC%88%98%EC%9C%A8-70-%EB%8F%8C%ED%8C%8C-%EC%96%91%EC%82%B0-mweiutjx
- 10jqka/财闻, Nov 19, 2026: https://news.10jqka.com.cn/20261119/c682013062.shtml
Limitations: The yield figures are unofficial and the company has not confirmed them. The Samsung yield figure is unverified. I did not check primary SK hynix filings for the Q3 call wording, so those quotes come from press reports.