They reached the DDR5 crossover six months early and DRAM is finally printing black ink again. HBM demand is growing at 80% while capacity is stuck in 2022. This is a supply squeeze disguised as an earnings report.
Thesis: SK hynix has transitioned from a cyclical memory play to a structural AI infrastructure provider. They own the HBM3 bottleneck. DRAM margins turned profitable this quarter on a 10% ASP jump. Supply is structurally constrained by tech migration rather than capacity expansion. This ensures pricing power through 2024.
Verdict: LONG — Conviction: MEDIUM
Catalyst: HBM3E volume ramp in 2024 and the industry-wide DDR5 crossover in the first half of the year.
Key Risk: NAND inventory normalization takes longer than expected. Continued losses at Solidigm could drain the cash generated by the DRAM recovery.
The Tell: Management admitted that 'investment in general servers is deemed essential for the utilization of AI services.' This reveals that the current cannibalization of general server budgets for AI is a temporary headwind that will likely reverse into a dual-demand tailwind in 2024.
Friction Level: MODERATE_FRICTION — Disagreement over whether HBM leadership can offset the structural rot in the NAND business. One side sees a structural AI demand shift. The other sees a capacity ceiling dictated by TSMC packaging limits and stranded legacy assets in China.
Report not found
The report data is no longer available. Please return to the archive.