Strategic investment in US indium refining to cut China dependence, HBM4 share pressure as Samsung yields recover…
Four things on Micron on 21 Oct. (1) Micron (release of 20 Oct) is strategically funding Indium Corporation to expand refining and high-purity material production at Rome, NY, with a long-term supply agreement.
Read more
This restores US domestic processing of critical minerals (indium is on China's export-control list) and, alongside the Clay, NY fab (ahead of schedule), reinforces the de-China push in its supply chain. The amount is undisclosed and the financial impact small, but long-term material risk is reduced. (2) The HBM4 competitive landscape is shifting against Micron. Chosun Ilbo reported on 21 Oct that Nvidia's Rubin is partially delayed and that SK Hynix missed HBM4 shipments in Q2 and Q3; Samsung's HBM4 yield is reportedly about 80%. Counterpoint's HBM revenue shares: Hynix fell from 58% in Q1 to 50% in Q2 and Samsung rose from 21% to 33%, implying Micron fell from about 21% to 17%. On its Q2 call, Samsung said Q3 HBM4 revenue would rise 200% quarter on quarter and HBM4 would make up more than 60% of its HBM in H2. Micron's HBM capacity of about 60,000 wafers a month (one-third of each Korean rival's) is a short-term ceiling on its share as Samsung's yield recovers. The Rubin delay pushes out the HBM4 ramp for all three. (3) DRAM contract prices are decelerating. TrendForce (20 Oct, via Commercial Times) has Q4 contract prices up about 20% quarter on quarter, against about 80% in Q2 and about 40% in Q3. The reasons: a high base, weaker end demand and price caps in long-term agreements kicking in. This fits the 10-K disclosure of take-or-pay terms with price floors and ceilings. Long contracts strip out the extreme upside but guarantee volume and a floor, so the profit curve is smoother, with no more doubling in a single quarter. Nvidia is reportedly considering moving Rubin Ultra from 12-high to 8-high HBM: 33% less HBM capacity per GPU and 50% more GPUs per DRAM die, but 8-high is 10-20% more expensive per Gb. Memory is now the bottleneck on GPU shipments, and buyers are sacrificing capacity for volume. Apple and Tesla are reportedly exploring CXMT and YMTC as price leverage. The industry view is that Chinese vendors are still limited in volume, price competitiveness and addressable markets, so the short-term leverage is limited. (4) Taiwan labour update: the Taoyuan union wants 15% of operating profit as a quarterly bonus and strikes on Thursday 22 Oct, the same day as the Taichung union's second mediation. Micron's tech seminar is tonight at about 23:00 SGT, with DeBoer on the DRAM, HBM and NAND roadmap. We will follow up tomorrow.
Key evidence
- Micron's official newsroom (20 Oct): strategic funding for Indium Corporation to expand its refining and high-purity material lines in Rome, NY, restoring US domestic processing of critical minerals. A long-term supply agreement covers high-purity materials used in advanced memory manufacturing
- Context: China has put indium and other critical minerals on its export-control list, and the Clay, NY fab is reportedly running months ahead of schedule. The deal size is undisclosed and the financial impact small, but long-term material risk is reduced
- Chosun Ilbo (21 Oct): Nvidia's Rubin is partially delayed, with mass-production tuning slower than expected. SK Hynix HBM4 shipments were below plan in both Q2 and Q3. Industry estimates put Samsung's HBM4 yield at about 80%
- Counterpoint HBM revenue share: Hynix fell from 58% in Q1 to 50% in Q2, and Samsung rose from 21% to 33%. The residual implies Micron fell from about 21% to 17%
- Samsung Q2 earnings call: Q3 HBM4 revenue to triple QoQ (up 200%-plus), and HBM4 to exceed 60% of its HBM revenue in H2
- Micron's HBM capacity is about 60,000 wafers a month, roughly one-third of each Korean competitor's. The short-term ceiling on its share becomes more visible as Samsung's yield recovers
Show 9 more
- The Rubin delay pushes out the HBM4 ramp for all three suppliers, so on that point the headwind is not specific to Micron
- TrendForce (20 Oct, via Commercial Times): Q4 DRAM contract prices up about 20% QoQ, against about 80% in Q2 and about 40% in Q3
- Reasons for the slowdown: a high base, weaker end-market purchasing power and price caps in long-term agreements kicking in
- This fits the 10-K disclosure of take-or-pay terms with price floors and ceilings: long contracts remove the upper tail but lock in volume and a floor. The result is a smoother profit curve, with no more single-quarter doubling
- Nvidia is reportedly considering moving Rubin Ultra from 12-high to 8-high HBM. HBM capacity per GPU would fall by about 33%, and the same DRAM dies would support about 50% more GPUs. But 8-high costs 10-20% more per Gb (less amortisation of the base logic)
- The signal: memory has become the bottleneck on GPU shipments, and buyers are sacrificing capacity to boost unit shipments
- Apple and Tesla are reportedly exploring CXMT and YMTC as pricing leverage. The industry view: Chinese vendors are limited in output, price competitiveness and addressable markets, so the short-term leverage is limited
- Taoyuan union: wants 15% of operating profit as a quarterly employee bonus, with a one-day strike on Thursday 22 Oct. The Taichung union holds its second mediation the same day
- Micron tech seminar tonight at about 23:00 SGT, with DeBoer presenting the DRAM, HBM and NAND roadmap
Catalysts to watch
- Tonight, about 23:00 SGT: Micron tech seminar, with DeBoer on the DRAM, HBM and NAND roadmap. Follow-up on 22 Oct
- Thursday 22 Oct: one-day strike by the Taoyuan union; second mediation for the Taichung union the same day
- Q4 DRAM contract pricing: TrendForce expects a rise of about 20% QoQ. A downside surprise would validate the price-cap thesis faster
- Quarterly HBM4 share tracking (Counterpoint, TrendForce): the key signal is whether Micron's share keeps compressing below 17%
- Nvidia's HBM stack decision for Rubin and Rubin Ultra (12-high vs 8-high), which materially shifts HBM capacity demand per GPU
Show 3 more
- SK Hynix's next earnings: whether HBM4 shipments miss for a third straight quarter
- Samsung's next earnings: confirmation of the HBM4 ramp guided on the Q2 call
- Any disclosure of progress milestones at the Clay, NY fab
Risks
- HBM4 share compression is the clearest headwind: implied Micron share has fallen from 21% to 17%, Samsung's yield is coming back, and Micron's HBM capacity is capped at about 60,000 wafers a month (one-third of each Korean peer's). Its 'number two' position is no longer protected
- The slowdown in DRAM contract prices (up 20% in Q4 against 80% in Q2) is a structural turning point, not a cyclical blip. Long-term agreements carry price caps, so there will be no more single-quarter doubling, even when demand is strong
- If Nvidia moves Rubin Ultra to 8-high HBM, HBM content per GPU drops by about 33%, a demand-elasticity shock for HBM makers, Micron included
- Apple's and Tesla's flirtation with CXMT and YMTC is noise for now, but a real medium-term risk if Chinese vendors scale
- Taiwan labour action (the 22 Oct strike) could build into an impact on Q1 shipments if mediation fails
Show 2 more
- The Rubin delay cuts both ways: it pushes out the HBM4 ramp for everyone, which weighs on near-term revenue visibility
- The headline 'de-China' story (the indium investment, the Clay fab ahead of schedule) reduces long-term risk but has no near-term financial impact. Do not mistake it for an earnings catalyst
Sources (6)
- Micron official newsroom (20 Oct): strategic investment in Indium Corporation
- Chosun Ilbo (21 Oct): SK Hynix HBM4 shipment misses, Samsung yield recovery
- Counterpoint Research: Q1 and Q2 HBM revenue share data
- Samsung Q2 2026 earnings call: H2 HBM4 ramp guidance
- Commercial Times via TrendForce (20 Oct): Q4 DRAM contract-price forecast
- Micron FY2026 10-K: context on its take-or-pay and price floor and ceiling disclosure
A brief is a lead for our research, not a verified conclusion or advice.