Micron Pushes Supply-Demand Visibility Out to 2028 as Agentic AI Adds a New Memory Demand Vector
Fiscal Q4 2026 earnings call, September 30, 2026
Micron Technology used its fiscal fourth-quarter call to deliver one of its most aggressive multi-year demand statements in years, telling analysts that memory market tightness now extends through 2028, a year further out than management had previously been willing to commit to. The signal was unmistakable: with more than 75% of fiscal 2027 shipments already committed, Micron's conversations with customers have shifted from allocating 2027 supply to negotiating 2028 commitments. Executives said they "really don't have line of sight to when supply and demand balance," a phrase that sets a notably bullish tone for a company that has historically been cautious about extrapolating cycle length.
President and Chief Operating Officer Manish Bhatia attributed the extended tightness to two forces compounding each other. On the demand side, server unit growth is running in the high teens, and agentic AI workloads are creating what he called "a CPU-driven demand stream" layered on top of accelerator-driven HBM demand. On the supply side, the structural constraints Micron has flagged before remain intact: diminishing bit-growth returns from each new technology node, HBM capacity growing faster than conventional DRAM, and rising trade ratios as the industry moves from HBM3 to HBM4 and eventually HBM4E. Clean-room construction timelines compound the problem further, since new capacity takes multiple quarters to ramp into meaningful shipments even after a facility opens.
Agentic AI Emerges as a Distinct Demand Driver
One of the more novel threads on the call was management's explicit framing of CPU-attached memory, rather than just HBM tied to accelerators, as a newly important swing factor. Bhatia pointed to the speed at which agentic workloads are scaling in both enterprise and consumer settings, citing Meta's recently released Muse product as evidence of "how quickly the agentic workloads are realizing value" for end users. He argued this dynamic is pulling through higher attach rates of LPDRAM, DDR and SSD content around CPUs, reinforcing Micron's view that DRAM, not logic or power, is now the primary bottleneck in AI data center buildouts. Management also acknowledged that customers are currently "de-speccing" memory content in AI racks out of necessity given supply constraints, implying meaningful latent demand that could surface quickly if more supply becomes available, a dynamic Morgan Stanley's Joseph Moore pressed on directly during the call.
HBM Pricing Reset and the NVIDIA Custom Silicon Relationship
On HBM specifically, Micron confirmed it has raised pricing significantly for calendar 2027, a move designed to narrow the profitability gap between HBM and conventional DRAM, though the company declined to update its longer-term HBM total addressable market forecast. More notable was new detail on the company's custom silicon relationship with NVIDIA. Chief Technology and Product Officer Scott DeBoer disclosed that Micron has been co-developing "NV HBM," a customized HBM4E product, with NVIDIA for more than a year, describing it as "really the first major custom HBM product out in the market." Unlike Micron's in-house base die approach used on standard HBM4, the NV HBM design runs on a foundry process for both the customized and JEDEC-standard variants. DeBoer was careful to frame the competitive differentiation in power efficiency and speed performance rather than the base die itself, arguing that quality execution, not architecture alone, has historically separated HBM suppliers and will continue to do so.
Long-Term Supply Contracts Expanding in Scope and Duration
Micron now has 26 strategic commercial agreements (SCAs) covering roughly 35% of bit volume through 2030, with DRAM slightly below that figure and NAND slightly above it. Of the 10 new SCAs signed since the last call, Bhatia said the agreements range from small to large customers and now span every business unit, including the mobile and client segment, which had previously been slower to adopt these structures. Pricing frameworks remain largely built around floor-and-ceiling bands, but newer agreements are being negotiated with explicit reference to current tight market conditions rather than the calendar second-quarter pricing environment that underpinned earlier contracts. Roughly three-quarters of SCA revenue now sits under a defined pricing framework, with the remainder subject to periodic market-based renegotiation. Management also noted conversations are now extending commitments beyond 2030, which it described as central to its confidence in committing capital to new capacity.
CapEx Mix Shifts Toward Construction, Pushing Capacity Into 2028 and Beyond
Chief Financial Officer Mark Murphy confirmed that the bulk of Micron's fiscal 2027 CapEx increase versus prior plans is going toward construction rather than equipment, specifically clean-room space that will not come online until late calendar 2028 or later. That spending does not translate into bits any time soon, a point Wolfe Research's Chris Caso pushed on, but it does signal how far out Micron's demand visibility now extends. Equipment will only be installed once clean rooms are built, and actual production capacity will be calibrated to demand trends closer to the time, with SCAs serving as the company's primary read on how much to build. Micron's first wafer output from its Idaho facility is expected mid-calendar 2027, with a new clean room in Singapore slated for the second half of calendar 2028.
China Exposure Shrinking, Technology Lead Holding at Two Nodes
Micron reiterated that its China sales exposure has been declining steadily and will fall to a single-digit percentage of revenue in fiscal 2027. DeBoer characterized Micron's technology lead over Chinese competitors at "at least two nodes," with the 1-gamma DRAM node now representing the majority of bit output and positioned to become the largest node in company history, an outcome dependent on EUV lithography. The next-generation 1-delta node is in development, with ramp targeted for the second half of calendar 2027.
Near-Term Guidance Details
For modeling purposes, Murphy added that first-quarter guidance assumes single-digit sequential bit growth alongside double-digit cost increases across both DRAM and NAND. He also flagged that fiscal 2027 year-over-year R&D spending growth, excluding incentive compensation effects, will exceed the $1 billion figure given last quarter, reflecting additional research activity the company has layered in since. NAND bit shipments underperformed the industry in fiscal 2026 as Micron repurposed some existing clean-room space in Singapore for HBM pilot operations and advanced R&D, though management expects NAND market conditions to stay tight even as industry bit growth runs in the mid-20% range through calendar 2027 and 2028.