FollowingDay 11in the edition since 24 Oct 202610 sourcesupdated 2 Nov 2026
The memory chip 2029 oversupply debate
The gist
Latest: Supply Executives and Huang Counter HBM Oversupply Fears; SK Hynix Posts Record Results
The gap between the oversupply forecast that triggered the selloff and the supply-constrained outlook from memory executives and analysts has direct implications for memory pricing and capital allocation across the AI supply chain. Major customers are locking in multi-year supply agreements years in advance, and pricing for next-generation HBM is being set at multiples of current generation prices.
When stories like this move, I write about them in an email each morning.
One free email each morning, around 8 AM US Eastern.
Check your email.
I have sent you a confirmation link. One click and you are in.
Not in your inbox? Look in spam or promotions. If it's there, mark it "Not spam" so future editions land in your inbox.
The full picture
SK Hynix reported Q3 revenue of $78.9 billion at an 80% operating margin, with Q4 DRAM shipment guidance of roughly 20-25% growth that far exceeded analyst expectations of around 9%, and HBM4 set to surpass HBM3E in both shipments and revenue in Q4 2025. KB Securities projects SK Hynix Q4 operating profit of roughly 104 trillion won at an approximately 83% operating margin, a 441% year-over-year increase. These results follow a sharp memory stock selloff triggered by a South Korean opinion column published October 15, in which two experts predicted the AI memory supercycle could end and the industry could enter a downturn by 2029. Nvidia CEO Jensen Huang pushed back in an October 22 CNBC interview at GTC Berlin, saying a memory surplus by 2029 is unlikely and that 'One of the things that's really great about memory is that you can always use more of it.' UBS characterized the 19% selloff as a buying opportunity, citing strong DDR5 server orders and AI agent-driven demand into 2027. Micron CEO Sanjay Mehrotra said demand for memory will exceed available supply through at least 2028, citing the transition to HBM4 and 4E as an additional headwind on supply growth. Samsung told investors that DRAM and NAND shortages will worsen in 2027 before easing, with tight supply persisting through 2028, and has locked 60-70% of its memory capacity into multi-year data center and AI deals. Samsung is seeking prices in the mid-to-high $4 per gigabit range for HBM4 in 2027 supply talks, more than triple the roughly $1.50 per gigabit it charges for HBM3E. Nvidia is also evaluating at least three downgraded HBM configurations for its Rubin Ultra GPU, including a reduction from a planned 16-stack HBM4E to 12-stack and potentially 8-stack variants, described as a response to the structural HBM supply crunch.
Korean media reported that Nvidia and Google are co-designing custom HBM base dies with memory makers, with Nvidia's NVHBM moving the memory controller into the HBM base die and Samsung considering manufacturing those dies at TSMC.
The custom design commands higher prices and creates multi-year customer lock-in. KB Securities projected SK Hynix will earn ₩104 trillion in Q4 2025 operating profit, which it described as the first quarter above that threshold and a 441% year-over-year increase at an 83% margin.
SK Hynix announced a new Americas headquarters at the president level, effective November 1, to be led by Ahn Hyun, its head of HBM development, with a mandate for direct engagement with major US AI customers including Nvidia.
The unit is paired with a $4 billion HBM packaging plant investment in Indiana. The moves come after a sharp October selloff in memory stocks, triggered by a South Korean opinion column predicting the market could become a 'red ocean' by 2029.
SK Hynix reported Q3 2025 revenue of $78.9 billion with 328% year-over-year growth and 80% operating margins, guided Q4 DRAM shipments 20-25% higher against analyst expectations of roughly 9%, and forecast demand outpacing supply through 2030.
It also disclosed agreements with 14 major customers covering more than 50% of its planned 2027-2028 capacity. Samsung has sought mid-to-high $4 per gigabit for HBM4 in 2027 supply talks, according to Korean media, more than triple its HBM3E rate of roughly $1.50 per gigabit.
Nvidia CEO Jensen Huang said at an October 22 CNBC interview at GTC Berlin that a 2029 memory surplus is unlikely, pushing back on a South Korean opinion column from October 15 predicting a market 'red ocean' that had cut SK Hynix 8.07% and SanDisk 8.51%.
UBS called the 19% sector decline a buying opportunity, citing strong DDR5 orders and AI agent demand into 2027, and projects HBM4 rising from 17% of total HBM demand in 2026 to 39% in 2027.
Bernstein argued long-term contracts covering roughly 40% of DRAM supply dilute reported prices, turning a 30% spot gain into about 18% in published figures, and said a 15-20% quarterly rise ranks in the top 5-10% of the past 15 years, while Western Digital beat estimates and guided $5.0-5.2 billion next quarter.
Component shortages spread beyond memory into multilayer ceramic capacitors, with Samsung Electro-Mechanics lead times reaching 61 weeks for some products as AI servers consume more than 10 times as many MLCCs as standard servers.
On October 23, a South Korean column warning of a memory 'red ocean' by 2029 and CXMT's DRAM announcement sent SanDisk down 8.51%, SK Hynix roughly 8%, and the Philadelphia Semiconductor Index to its longest losing streak since 2008.
SK Hynix also halted a Solidigm U.S. IPO citing volatility. Huang called a 2029 glut unlikely at GTC Berlin, and UBS called the selloff "an opportunity, not a sign the cycle is breaking". Samsung and SK Hynix showed HBM4E products at SEDEX 2026 in Seoul, both targeting above 16 Gbps.
Samsung cuts Q4 mid-range and budget phone production target from 31.4 million to 24.2 million units as AI chip demand raises memory costs.
18 October 2026
Samsung and SK Hynix display competing HBM4E products at SEDEX 2026 in Seoul, both targeting speeds above 16 Gbps
1 October 2026
Micron CEO Sanjay Mehrotra told investors demand for memory will exceed supply through at least 2028, citing HBM generation transitions as an additional headwind on supply growth, and said 2027 memory prices are much higher than 2026 prices.
1 September 2026
Samsung reported to be developing 8-layer HBM4E NVHBM variant for Nvidia as near-term drop-in product
26 August 2026
Nvidia announced NVHBM custom memory solution with 30% higher bandwidth and 15% lower power than commodity HBM4e, for NVLink Fusion partners
Nvidia reported testing downgraded HBM configurations for Rubin Ultra GPU, including 12-stack and 8-stack variants instead of planned 16-stack HBM4E, described as a response to structural HBM supply crunch. Lam Research raised advanced packaging revenue outlook from >40% to >70%.