Briefing
- Micron's net profit rose nearly 1,400% year-on-year in its fiscal third quarter, the direct result of supply-constrained memory prices amplified by AI infrastructure buildout.
- Quarterly revenue more than quadrupled from $9.3 billion a year earlier to $41.46 billion, a record for the company.
- Gross margin expanded to 84.9% from 39% a year ago, surpassing those of Nvidia and Meta and making Micron the highest-margin company among large-cap technology peers.
- The stock advanced more than 16% in premarket trading following the results, and Micron overtook Meta and Tesla in market value amid relentless AI infrastructure demand.
- Micron also unveiled a slew of new deals alongside the earnings report, which analysts cited as additional reason for optimism about the stock's future momentum.
- The results reignited a broader AI rally, lifting global chip stocks including SanDisk and Western Digital, with analysts saying the earnings justify elevated valuations across the memory sector.
Analysis
- HBM and high-capacity DRAM suppliers face a sustained pricing tailwind through at least calendar Q4 2026, with Micron's above-consensus forward guidance confirming supply has not yet caught up to AI server demand. Each next-generation AI server rack requires multiples more DRAM and HBM than conventional servers; as hyperscaler capex commitments accelerate, the supply-demand gap widens before new capacity from SK Hynix and Samsung can be qualified at scale. This directly pressures cloud operators' server build costs.
- OpenAI's Jalapeño inference chip, built with Broadcom, will reduce OpenAI's per-query DRAM bandwidth consumption per GPU cycle over time as inference workloads migrate to custom silicon optimised for memory efficiency. However, the near-term effect is the opposite: rapid AI infrastructure scaling by OpenAI and peers is the primary driver of Micron's record quarter. Custom silicon reduces memory intensity per chip only at the margin; aggregate memory demand rises as total inference compute capacity expands.
- HIVE's Blackwell GPU deployment for sovereign AI cloud contracts creates incremental DRAM procurement demand, reinforcing the memory supply tightness Micron flagged. Smaller AI infrastructure operators like HIVE typically source DRAM through spot or secondary channels rather than long-term supply agreements, meaning they absorb price increases fully. As the crypto-miner-to-AI-infrastructure pivot scales across multiple companies, aggregate spot DRAM demand from non-hyperscaler buyers adds a structurally stickier bid beneath contract prices.
Historical Context
- 2023
Micron posted a net loss of approximately $5.8bn in fiscal 2023 as post-pandemic inventory corrections cratered memory prices. The 1,400% profit recovery in a single year illustrates the extreme earnings leverage memory manufacturers carry to pricing cycles, amplifying both upside and downside relative to revenue moves.
- 2021-2022
The prior DRAM upcycle saw memory prices peak in mid-2021 before collapsing roughly 50% through 2022 as PC and smartphone demand collapsed post-pandemic. That cycle was demand-driven on the way up; the current cycle is structurally different because AI server DRAM content per unit is 4-8x conventional servers, making the demand floor more durable.
- 2016-2018
Micron and SK Hynix delivered multi-hundred-percent profit gains during the 2016-2018 DRAM supercycle driven by smartphone DRAM content increases and cloud server buildout. That cycle ended when Samsung aggressively added capacity; the current cycle's constraint is HBM, which requires specialised stacking technology limiting how quickly new supply can enter the market.
Connected Events
Nvidia launches $20bn bond deal, its first debt sale since 2021
Nvidia's $20bn bond issuance, earmarked partly for AI infrastructure financing, reflects the same AI server buildout that is driving Micron's record DRAM demand. Both stories point to accelerating data center capital formation with no near-term deceleration signal from the largest infrastructure buyer.
HIVE secures $220M sovereign AI cloud contract with Bell and Cohere
HIVE's 2,304 Blackwell GPU sovereign cloud deployment adds a named, contracted demand anchor for AI compute infrastructure outside hyperscalers, corroborating Micron's guidance that AI-driven memory demand is broadening beyond the largest buyers.
OpenAI unveils Jalapeño, its first custom AI inference chip built with Broadcom
OpenAI's Jalapeño custom inference chip, designed to optimise per-query compute costs, represents a long-term architectural shift that could moderate memory bandwidth intensity per chip, but the current generation still relies on standard memory procurement and does not yet offset aggregate DRAM demand growth.
Further reading
See Indexa more often on Google
Mark Indexa as a preferred source — your Top Stories will surface more Indexa coverage.