Good Thursday,
SK Hynix's exceptional Q1 2026 with 72% operating margin
Earnings season is underway. SK Hynix reported Q1 2026 today and told us memory demand still sits well above available supply:
₩52.6T revenue
72% operating margin
DRAM ASP up mid-60% QoQ
NAND ASP up mid-70% QoQ
Customer demand over the NEXT 3 YEARS already exceeds SK Hynix’s supply capacity by a significant margin. Management explicitly pointed to key equipment procurement, including EUV, as a major capex driver, with long lead times still relevant.
Yongin Phase 1 was pulled forward by 3 months to February 2027, but there are still no new fabs planned outside Yongin. In other words, capacity is moving forward, but not fast enough to erase the imbalance.
There was also an important nuance in NAND: ASP surged while shipments fell about 10% QoQ. That is a pricing power signal under constrained supply, not volume-led growth.
On spot pricing, SK Hynix was clear that recent weakness reflects channel inventory coming into market after the run-up, not an end-demand rollover.
The real bear case to watch is whether inference shifts toward lower-HBM architectures over time. SK Hynix’s view is that those systems hit their own limits and complex workloads still flow back toward HBM-GPU setups. That makes upcoming hyperscaler capex commentary on inference hardware mix especially important.
What I’m watching next:
- LTA confirmation in Q2 or Q3
- ADR SEC review progress
- May spot price direction
ALSO: Intel reports today after the closing bell. A named 18A design win (their latest foundry process technology) or volume commitment reprices the foundry option in the stock.
Our four scenarios for the call:
Bullish: Revenue in-line + named 18A external customer disclosed.
Mixed: Beat + GM above 36%, but no foundry customer news.
Bearish: Miss + GM below 33% + capex rising. Also bearish: Revenue in-line, but foundry language softens to "longer timelines."
See the full pre-call briefing in our terminal here ↓
On Our Radar
Tesla to Prepare First Humanoid Robot Factory in Q2 with 1 Million Unit Annual Capacity
Tesla is betting Fremont's retooled lines can prove out humanoid robot manufacturing at a scale no competitor has approached, and the Q2 preparation timeline makes 2026 a hard accountability moment. Replacing Model S and X production with Optimus lines means Tesla is absorbing real near-term revenue risk to chase a 1-million-unit annual capacity target that has never been demonstrated anywhere in the industry. Watch whether Q2 site prep actually begins on schedule, as any delay resets the entire 10-million-unit Texas roadmap.
Chinese VCs backing DeepSeek at $20B+ means frontier model funding flows domestic instead of to US investors, reducing US leverage over deployment decisions.
Applied Digital's $7.5B hyperscaler lease locks supply off the open market, concentrating GPU hosting among well-capitalized players and squeezing smaller labs.
Regime Snapshot
Compute (CRS): 65, scarcity. More buyers than available capacity. Lead times extending.
Memory (MRS): 89, Shortage. Watch for directional shifts.
Memory tighter than compute this cycle. Watch for HBM allocation pressure on new model deployments.
Narratives Moving Today
Can China build chips without NVIDIA? ▼16 pts this week.
The Daily Chain gives you the headline. The terminal shows you which companies are exposed, how the constraint map is shifting, and what the regime history says about timing.
See you tomorrow,
Teng