RockstarMarkets
All news
Markets · Narrative··Updated 23m ago
Part of: Semiconductor Cycle

MU Gross Margins Above 70% as HBM3E Lead Times Stretch Beyond 12 Months Into 2027

HBM3E scarcity, concentrated among just three global suppliers, is delivering extraordinary pricing power to Micron while simultaneously bottlenecking GPU cluster builds for NVDA and AVGO customers. AMAT, LRCX, and KLAC are absorbing record tool orders from memory fabs racing to expand capacity, lifting SOXX equipment

R
Rocky AI · RockstarMarkets desk
Synthesised from 8 wires · 0 mentions in the last 24h
Sentiment
+55
Momentum
70
Mentions · 24h
0
Articles · 24h
10
Affected sectors
Related markets

Key facts

  • Micron gross margins exceeded 70% in late May 2026 amid HBM3E supply constraints
  • HBM3E lead times stretched beyond 12 months into 2027, creating critical AI capex bottleneck
  • Memory-chip scarcity concentrating pricing power among three global suppliers: Micron, Samsung, SK Hynix

What's happening

Memory-chip supply constraints are crystallizing as the primary bottleneck in AI infrastructure capex cycles. Micron's May 2026 disclosures indicated gross margins have expanded beyond 70%, an extraordinary level for commodity DRAM, driven by acute scarcity in HBM3E (high-bandwidth memory) modules essential for GPU clusters. Lead times for HBM3E have extended beyond 12 months, meaning orders placed today are not fulfilled until late 2027. This supply inelasticity is a structural inflection point that could either accelerate capex pull-forward or cause data-center customers to delay projects.

The root cause is capacity. HBM3E manufacturing is concentrated among Samsung, SK Hynix, and Micron, with only a handful of fabs globally capable of the precision required for stacked memory architectures. Unlike GPUs (which NVIDIA has scaled aggressively), memory production cannot be ramped in a matter of quarters. Samsung recently announced expansions, but not until 2027-2028. This means Micron and its peers are harvesting extraordinary margins on existing capacity, and data-center customers (hyperscalers like Meta, Microsoft, and cloud providers) are paying up to secure allocations.

For semiconductor equipment makers (AMAT, LRCX, KLAC), the constraints mean record tool orders from memory manufacturers trying to expand capacity. NVDA and AVGO face a paradox: their GPUs and networking chips are in insatiable demand, but memory scarcity could throttle system builds and push customers toward cheaper, memory-optimized configurations. This favors custom AI chips (like Tesla's Dojo or Google's TPUs) that are optimized for lower memory bandwidth but face longer development timelines.

The margin expansion at Micron is unsustainable beyond 2027 absent structural demand destruction, and the risk is asymmetric: if customers deprioritize capex due to memory constraints, pricing collapses. However, the more likely scenario is that hyperscalers accept higher system costs to proceed with data-center builds, meaning HBM3E supply-driven margins sustain through early 2027. Investors should monitor Micron's inventory levels and Samsung's capacity announcements for signals of relief or continued tightness.

What to watch next

  • 01Micron Q3 2026 earnings and gross margin guidance: June 2026
  • 02Samsung and SK Hynix capacity announcements and fab expansion timelines: June-July 2026
  • 03AMAT, LRCX equipment orders and backlog indicators: quarterly reports through Q3 2026
Mention velocity · last 24 hours
Coverage from these sources
Previously on this story

Related coverage

More about $MU

Topic hub
Semiconductor Cycle: AI Capex, Memory and the SOX Trade

Live coverage of the AI semiconductor cycle — NVDA, AVGO, AMD, ASML, memory demand, capex run rates and overbought signals.