Micron Targets 2028 for Vertical HBM AI Memory Production

What Happened

Micron Technology has disclosed plans to produce a next-generation vertical high-bandwidth memory architecture designed to increase data throughput for artificial intelligence workloads, with a production target set for 2028. The company has identified thermal management as a primary engineering constraint that will shape how and where the technology is adopted.

Background

High-bandwidth memory, known as HBM, has become a critical component in AI accelerator systems. Graphics processing units and AI training chips from companies including Nvidia and AMD depend on HBM to move large volumes of data rapidly between memory and compute elements. Demand for HBM has surged alongside the expansion of large language models and other AI infrastructure, placing memory suppliers at the center of the AI hardware supply chain.

Micron is one of three major DRAM manufacturers globally, alongside Samsung Electronics and SK Hynix. SK Hynix has held a leading position in HBM supply, particularly as a key provider to Nvidia. Micron has been working to close that gap and has previously supplied HBM3E memory for use in Nvidia's H200 and related products.

Vertical HBM refers to an architectural approach that stacks memory dies in a taller configuration than current designs, with the goal of increasing memory density and bandwidth within the same or smaller footprint. The approach differs from existing HBM generations primarily in how the stacking geometry is extended.

The Cooling Problem

The central technical challenge Micron has identified is heat dissipation. Taller memory stacks generate more heat in a concentrated area, and existing cooling infrastructure used in AI server deployments may not be sufficient to manage thermal loads at scale.

Data center operators and system integrators have been expanding liquid cooling deployments to handle the rising power density of AI hardware. However, vertical HBM would place additional demands on those systems, and Micron has indicated that cooling constraints could influence which customers and platforms adopt the technology when it reaches production.

The company has not specified a particular cooling solution or stated whether it is developing proprietary thermal management technology alongside the memory architecture itself.

Market Context

The global HBM market has grown rapidly as AI infrastructure spending has increased across cloud providers, enterprise data centers, and government-funded research facilities. Analysts have projected continued HBM demand growth through the latter half of the decade, tied directly to AI accelerator shipment volumes.

Micron's 2028 production timeline places vertical HBM beyond the current product cycles of HBM3E and the anticipated HBM4 generation, which multiple suppliers are targeting for production in 2025 and 2026. Vertical HBM appears to represent a longer-horizon product development effort rather than an immediate competitive response.

The company has not disclosed pricing, yield targets, or specific customer partnerships associated with the vertical HBM program.

What It Means in Practice

For data center operators planning AI infrastructure beyond the current generation of hardware, Micron's 2028 timeline and the identified cooling constraints are relevant procurement and facilities planning considerations. Systems designed to use vertical HBM may require upgraded or purpose-built thermal management infrastructure.

For chip designers developing AI accelerators in that timeframe, the memory architecture represents a potential bandwidth increase that would need to be accounted for in system design, including power delivery and physical packaging.

Micron has not yet detailed the specific bandwidth figures or capacity per stack that vertical HBM is expected to deliver compared with HBM4 or subsequent conventional generations.

What Happens Next

Micron is expected to provide additional technical disclosures on the vertical HBM program at upcoming investor and industry events ahead of the stated 2028 production target.