Micron's financial data demonstrates that HBM4 memory commands 15x the value per unit weight compared to standard DRAM, reflecting its critical role in AI accelerator economics.

High-bandwidth memory (HBM) stacked DRAM may prove to be the most profitable product in the six-decade history of commercial information technology. IBM's mainframe engines in the 1970s occupied similar territory, depending on cost and revenue allocation, but HBM stacked DRAM packaging likely surpasses the monolithic systems technology of the System/370s and the thermal conduction module packaging of the System/3081 in driving revenues and profits. Advanced packaging has been invaluable historically and remains equally critical today.

The HBM market's growth and the revenues of Micron Technology, Samsung, and SK Hynix are accelerating far faster than IBM's mainframe business ever did. This growth underpins expansion at Nvidia and AMD. Without the high-bandwidth stacked DRAM these companies produce, the GenAI revolution would not have occurred.

Micron's fiscal 2026 results, ended September 3, demonstrate the scale of this opportunity. Quarterly revenues surged 4.8X year-on-year to $54.23 billion. Operating income rose nearly 12X to $43.75 billion, and net income increased 12X to $37.7 billion. Micron is converting 69.5 percent of revenue to the bottom line while maintaining disciplined capital spending.

The company holds $73.45 billion in cash, up 7.6X annually, yet invested only $10.77 billion in capital expenditures. Memory makers are constraining fab construction deliberately. All three major manufacturers believe AI demand will sustain through 2028 at minimum, likely into 2030, and are determined to avoid the boom-bust cycles that have historically plagued the memory business. Building foundries cannot be accelerated through parallelism—the process runs serially from conception to production. Moreover, with memory prices and profits climbing, manufacturers prefer maximum revenue extraction at lowest cost.

The DRAM business, encompassing low-power LPDDR for AI servers, server memory, and consumer devices, grew 4.4X to $39.77 billion. Flash storage proved the fastest-growing segment, with sales up 6.2X to $14.1 billion. Expanding capacity remains the binding constraint. NAND flash foundries are cheaper and faster to deploy than DRAM facilities, so Micron is expected to build flash capacity aggressively while moving more cautiously on DRAM given the technical difficulty and expense of bringing advanced DRAM fabs online.

Micron's chief technology officer Scott DeBoer stated that the company maintains a two-generation technology lead over China's ChangXin Memory Technologies (CXMT), which went public on the Shanghai Stock Exchange in July after raising $8.6 billion. Micron will work to preserve this advantage and prevent customer migration to Chinese suppliers, though CXMT's emergence signals China's serious commitment to domestic memory supply and eventual global competition.

The Cloud Memory business unit, dominated by HBM, is smaller than the Core Datacenter unit but growing rapidly. In fiscal Q4, Cloud Memory generated $16.28 billion in sales—up 3.6X year-on-year and 18.3 percent sequentially—with gross income of 83 percent ($13.52 billion) and operating income of 76 percent ($12.38 billion). Gross income grew 5X and net income 4.7X year-on-year, reflecting extraordinary pricing power.

According to CFO Mark Murphy, DRAM bit shipments rose by single-digit percentages in Q4 while prices increased in the high-teens range. Flash bits shipped rose 10 percent year-on-year with prices up 30 percent. These figures span all products; HBM, LPDDR, and high-end server DRAM price increases and shipment growth exceeded these averages substantially.

The Core Datacenter business rose 11X year-on-year to just over $18 billion, demonstrating how DRAM and flash are essential to AI training clusters, inference systems, and agentic AI execution environments. Combined, datacenter revenue reached $34.29 billion, representing 35.6 percent sequential growth, with operating income of $27.68 billion (81 percent of datacenter product revenues).

Micron no longer discloses HBM segment specifics, but industry observation suggests HBM memory drove approximately $14.33 billion in fiscal Q4, up 7.6X year-on-year, while high-capacity server DRAMs and LPDDR combined for $8.05 billion, up 6.1X. Remaining DRAM sales are estimated at $17.39 billion, growing 3.1X annually.

CEO Sanjay Mehrotra indicated that Micron has "completed agreements for the vast majority of our calendar 2027 HBM bit supply with significant price increases year over year, narrowing the gross margin gap with conventional DRAM." The HBM4 ramp is proceeding on schedule, and the company has a "strong roadmap for future HBM products," including collaboration with Nvidia on the industry's first custom-HBM4E implementation (NVHBM) for next-generation GPUs and NVLink Fusion platforms.

Looking ahead, memory and flash will face tighter constraints in 2027 and 2028 than in 2026. Micron expects DRAM bit shipments to grow in the mid-20s percent range this year, decelerating to the low 20s percent in 2027 and 2028, while flash capacity deployment will accelerate as supply chains become less constrained.