Earnings Call Highlights
Micron's Q4 FY2025 call was unambiguously bullish, with management projecting confidence bordering on triumphalism as they reported record revenue, record HBM quarterly revenue near $2B, and guided to another record quarter at $12.5B with 51.5% gross margins. The dominant theme was AI-driven demand acceleration across both HBM and traditional server DRAM, with Sanjay Mehrotra explicitly linking the uptick in traditional server growth to AI agent workloads—a notable new narrative. Analysts pressed hard on HBM supply, pricing durability, and the HBM4 transition, and management's responses were consistently confident, citing tight DRAM supply as structural rather than cyclical. The tone was that of a company that has successfully repositioned itself as a premium AI infrastructure supplier, with the HBM4 performance claims—2.8 TB/s bandwidth, 11+ Gb/s pin speed—serving as the centerpiece of competitive differentiation.
- Micron explicitly attributed a meaningful portion of the upgrade in calendar 2025 traditional server unit growth (from flat to mid-single-digit) to AI agent workloads: 'We believe this change in outlook is in part related to the growth of AI agents and the traditional server workloads agents initiate as they execute tasks on behalf of users'—a direct signal that agentic AI is now a measurable demand driver for memory infrastructure.
- HBM quarterly revenue reached nearly $2B in Q4, implying an ~$8B annualized run rate, with HBM share now on track to match Micron's overall DRAM market share in CQ3 2025; pricing agreements for the vast majority of HBM3E supply in calendar 2026 are already locked, and HBM4 production shipments are expected to begin in CQ2 2026.
- Micron's HBM4 samples are delivering 2.8 TB/s bandwidth and over 11 Gb/s pin speeds, which management claims outperforms all competing HBM4 products on both performance and power efficiency, enabled by Micron's in-house CMOS-based base logic die—a key differentiator as hyperscalers demand increasingly aggressive specs.
- For HBM4E (the 2027-generation product), Micron is partnering with TSMC for the base logic die and will offer both standard and customized variants, with customized base dies expected to carry higher gross margins than standard HBM4E—signaling a move toward co-design partnerships with leading AI chip customers similar to what NVIDIA and others pursue.
- Micron's data center segment reached 56% of total company revenue at 52% gross margins in FY2025, with the cloud memory business unit alone posting 59% gross margins in Q4; combined HBM, high-capacity DIMMs, and LP server DRAM hit $10B for the full year, a fivefold increase year-over-year, underscoring the scale of the AI infrastructure memory opportunity.