SK Hynix starts mass producing its new 12-layer HBM3E chips with 36GB per chip capacity, up 50% from its previous gen, slated for delivery by the end of 2024
Anniek Bao / CNBC :
Context & Ripple Effects
SK Hynix’s production milestone follows its disclosure that HBM manufacturing capacity was nearly fully booked through 2025, making execution and deliveries as important as a new capacity specification. The company is also pursuing the next transition through a TSMC partnership on HBM4 and advanced packaging.
The 36GB, 12-layer format puts SK Hynix in the same product class Samsung had already outlined with its 36GB 12-layer HBM3E launch, turning the contest from product announcements into volume qualification and supply fulfillment.
First-order effects
- SK Hynix can begin supplying a higher-capacity HBM3E option by year-end, giving its customers a denser memory component for AI-oriented systems.
- The start of mass production strengthens SK Hynix’s ability to convert constrained HBM capacity into shipments, rather than merely reserving future output.
Second-order effects
- Samsung faces more pressure to turn its competing 36GB 12-layer HBM3E roadmap into qualified, volume supply; feature parity alone will not settle the market.
- Demand shifts further toward advanced stacking and packaging capacity, reinforcing the supply-chain importance of the production bottlenecks identified in SK Hynix’s booked-out capacity.
Third-order effects
- HBM competition is likely to be defined increasingly by dependable scale, customer qualification, and packaging partnerships rather than by conventional memory-cycle dynamics alone.
- If 12-layer HBM becomes a standard AI-system building block, suppliers with credible paths from HBM3E into HBM4 may gain disproportionate influence over AI hardware availability.
The trend: AI infrastructure demand is separating high-bandwidth memory into a capacity-constrained, technology-led segment where manufacturing execution and packaging ecosystems determine competitive position.