June 27, 2026 · 14 min read
Micron Technology has gone from a money-losing cyclical chipmaker to one of the most critical enablers of the AI revolution. With HBM3E memory sold out through 2026, DRAM prices at multi-year highs, and $13.6B in CHIPS Act support building US fabs, MU trades at just ~12x forward earnings — making it arguably the cheapest pure-play on AI infrastructure in the market. We break down the thesis in full.
High Bandwidth Memory (HBM) is a specialized type of DRAM that stacks multiple memory dies vertically, connected by thousands of tiny through-silicon vias (TSVs). The result is memory that delivers 5-10x the bandwidth of standard DDR5 DRAM while consuming 30-40% less power per bit. For AI workloads — training massive language models, running inference at scale — bandwidth is the bottleneck, not compute. That makes HBM the most critical component in every NVIDIA H100, H200, B100, and B200 GPU.
Every NVIDIA B200 GPU requires 36GB of HBM3E memory arranged in an 8-stack configuration around the processor die. A single DGX B200 server with 8 GPUs contains 288GB of HBM — roughly $8,000-$10,000 worth of memory alone. With NVIDIA, AMD, and custom ASIC makers shipping hundreds of thousands of AI accelerators per quarter, the total addressable market for HBM has exploded from under $2B in 2022 to an estimated $25-30B in 2026.
Micron was the last of the three major memory makers to enter HBM production but has caught up rapidly. The company began volume shipments of its HBM3E 8-high stack in late 2024, and in early 2026 qualified its industry-leading HBM3E 12-high stack — which offers 50% more capacity per stack than the 8-high version. Micron's HBM3E has been validated by both NVIDIA and AMD for their latest GPU platforms, and the company has confirmed that its entire HBM production for 2026 and most of 2027 is already sold out.
Micron's income statement has undergone a dramatic transformation. After posting negative gross margins during the 2023 memory downcycle, the company has rebounded to ~45% gross margins — driven by HBM mix shift, disciplined DRAM pricing, and improved NAND cost structures. Operating margins have recovered to approximately 32%, reflecting both top-line leverage and cost discipline.
DRAM remains Micron's dominant revenue driver at approximately 73% of total sales, followed by NAND flash at 25%. Within the DRAM category, the highest-growth segment is HBM, which contributes over $8B in revenue and is expected to more than double year-over-year. Standard server DRAM for cloud and enterprise also benefits from strong data center demand and favorable pricing.
Data center and AI workloads now account for 55% of Micron's total revenue — up from roughly 30% just two years ago. Mobile (smartphones using LPDDR5X) contributes 20%, client PCs 15%, and the fast-growing automotive and industrial segment rounds out the remaining 10%. The data center shift is structural: as AI model sizes grow exponentially, memory requirements scale proportionally.
Memory semiconductors have historically been one of the most brutally cyclical industries in technology. Prices swing wildly based on supply and demand, and the big three producers — Samsung, SK Hynix, and Micron — have repeatedly over-invested at the top and suffered massive losses at the bottom. But this cycle has several structural differences that suggest a more durable upcycle:
The HBM market is a three-player oligopoly. SK Hynix holds the leading position with roughly 50% market share, having been the first to mass-produce HBM3E for NVIDIA's H200 and B100 GPUs. Micron and Samsung split the remaining 50%, though Micron is widely regarded as having the superior technology compared to Samsung's struggling HBM3E 8-high yields.
Micron's 12-high HBM3E stack delivers 36GB per stack with 1.2 TB/s bandwidth — representing a meaningful performance and density advantage over 8-high stacks. The company was the second manufacturer (after SK Hynix) to achieve volume production of 12-high stacks, and Micron claims its product offers 20% better power efficiency than the competing offerings. This efficiency advantage matters enormously in data center deployments where power is the primary constraint.
Looking ahead, Micron has confirmed development of HBM4, the next-generation standard expected to arrive in late 2027. HBM4 will feature a new logic-base die architecture that integrates more functionality directly into the memory stack, enabling over 2 TB/s bandwidth per stack. Micron is investing heavily in its Hiroshima, Japan facility to serve as the primary HBM4 development and early production site, with plans to bring production back to the US for future generations.
Micron's financial trajectory over the past three years tells the story of a company riding the AI wave from the bottom of a brutal memory downcycle to record revenue and profitability. The swing from -$5.8B in net income in FY2024 to an expected +$11B in FY2026 is one of the most dramatic turnarounds in semiconductor history.
The capex number deserves attention: Micron is spending over $14B in FY2026, one of its largest investment years ever. But critically, this spending is directed toward high-return projects — HBM capacity expansion and the Idaho mega-fab — rather than commodity capacity that would depress pricing. Management has guided that HBM-related capex generates returns well above the company's cost of capital.
Micron is one of the largest beneficiaries of the CHIPS and Science Act, securing $6.1 billion in direct grants and up to $7.5 billion in federal loans — totaling $13.6B in government support. This funding is earmarked for two transformative manufacturing projects that will reshape Micron's production footprint and reduce its dependence on overseas fabs.
Micron is building a state-of-the-art DRAM fabrication facility in Boise, Idaho, with construction well underway and initial production expected to begin in late 2027. When fully operational, the Idaho fab will be one of the most advanced memory manufacturing facilities in the world, capable of producing leading-edge DRAM and HBM on the latest EUV lithography nodes. The total investment is expected to exceed $15B over the next decade.
In addition to Idaho, Micron has committed to building a massive memory fabrication complex in Clay, New York, with total planned investment of up to $100B over 20+ years. The first phase, supported by CHIPS Act funding, will focus on advanced DRAM production. The New York site positions Micron as a strategic supplier for US government and defense applications that require domestically sourced semiconductors.
The memory semiconductor industry is a three-player oligopoly, making peer comparison essential for understanding Micron's relative positioning. Each competitor brings different strengths: SK Hynix leads in HBM market share, Samsung has the most diversified semiconductor portfolio, and Micron offers the most compelling valuation with strong technology execution.
For a deeper comparison of how Micron stacks up against AMD in the broader semiconductor landscape, see our detailed MU vs AMD comparison.
Despite the compelling growth story, Micron carries meaningful risks that investors should weigh carefully. Memory stocks are not for the faint of heart — even in structural upcycles, the stocks can sell off sharply on the first sign of demand softening or pricing pressure.
Memory is cyclical by nature. While structural AI demand provides a floor, a global recession or hyperscaler capex slowdown could trigger a demand correction. Micron's revenue fell 49% in FY2024 during the last downturn — a reminder that the amplitude of memory cycles can be severe. Investors betting on Micron are implicitly betting that the AI demand pillar is strong enough to prevent another downcycle of similar magnitude.
While DRAM pricing remains robust, the NAND flash market is more competitive and prone to oversupply. NAND accounts for 25% of Micron's revenue, and a sharp NAND price decline could drag on overall margins even if DRAM/HBM remain strong. Chinese NAND producers like YMTC continue to gain share in the lower-end of the market, adding supply that major players cannot easily match on cost.
China represented approximately 25% of Micron's revenue before the Chinese government's Cyberspace Administration effectively banned Micron products from critical infrastructure in May 2023. While Micron has partially offset this loss through growth in other markets, any further geopolitical escalation could hurt revenue, and a potential relaxation of restrictions remains uncertain.
SK Hynix holds a dominant 50% share of the HBM market and has deeper relationships with NVIDIA dating back to the earliest HBM generations. If SK Hynix accelerates its HBM4 roadmap or captures an outsized share of next-generation GPU designs, Micron could find its HBM market share gains stalling or reversing. The competitive dynamic in HBM is intensely capital-intensive and winner-take-most for each GPU design cycle.
Micron's Idaho and New York fabs represent tens of billions in committed investment. Construction delays, cost overruns, or difficulties ramping new facilities to competitive yields could weigh on profitability. Building greenfield fabs in the US also carries higher labor and construction costs compared to Asian sites, though CHIPS Act funding partially offsets this premium.
Micron is the cheapest way to play the AI memory supercycle at just ~12x forward earnings. While NVIDIA trades at 30x+ and AMD at 25x, Micron offers direct exposure to the same AI infrastructure buildout at a fraction of the valuation. The reason for the discount is simple: memory stocks carry cyclical risk that logic chipmakers do not. But this cycle is different — HBM supply constraints, disciplined industry capex, and structural AI demand provide a more durable floor than in previous memory upcycles.
The investment thesis boils down to three core pillars: (1) HBM3E is sold out through 2026 with pricing that delivers 3-5x the margins of commodity DRAM, (2) $13.6B in CHIPS Act funding de-risks the next decade of capex while building a US manufacturing moat, and (3) the industry's supply discipline is keeping DRAM pricing at multi-year highs even outside of HBM. If you believe AI infrastructure spending continues to grow — and every major hyperscaler's capital expenditure budget suggests it will — Micron sits at the intersection of unprecedented demand and constrained supply.
The risks are real: memory is cyclical, NAND oversupply is a drag, and China restrictions have cost Micron billions in lost revenue. But at 12x forward earnings with 80% revenue growth, the market is pricing in significant cycle risk that may not materialize in the way bears expect. For investors willing to accept memory-sector volatility, MU offers a compelling risk/reward at current levels.