Micron leads the way in GDDR stacking technology, quietly igniting a new round of competition in the AI memory market.

Micron is looking to carve out a new track in the AI memory space—by vertically stacking graphics memory (GDDR) the way HBM is stacked, to fill the market gap between HBM and standard GDDR.

According to a report from Korea Electronic News (ETNews), Micron has kicked off development of a new vertically stacked GDDR product. The company plans to complete equipment deployment in the second half of this year and move into the process testing stage.

Initial expectations are to achieve roughly a four-layer GDDR stack, with sample units launching as early as next year. This move comes earlier than Samsung Electronics and SK hynix, and could give Micron a first-mover advantage in this emerging niche.

The backdrop for this move is that AI application scenarios continue to expand, driving a faster divergence in memory demand. Thanks to its cost advantages, GDDR is being used in an increasing number of AI inference accelerators, but its bandwidth bottleneck limits further penetration. By improving GDDR performance through stacking technology, this market may be opened up to new growth opportunities.

GDDR stacking: positioning it between HBM and standard memory by price

GDDR is a type of memory optimized specifically for video processing and 3D graphics rendering. For a long time, it has been used mainly in graphics cards and gaming devices. Compared with HBM, GDDR has lower bandwidth, but it is more competitively priced; in recent years, it has begun to enter some AI accelerators.

The stacked GDDR product Micron is developing aims to create a new offering with performance positioned between HBM and standard GDDR—higher bandwidth than existing GDDR, but lower cost than HBM. The product specifications are currently being discussed with customers such as AI accelerator makers as part of demand alignment. The market direction is clearly aimed at meeting customers’ customized requirements.

Analysts believe that stacked GDDR will occupy the middle market between HBM and non-stacked GDDR, and it may also capture substantial demand in the continuously growing high-performance discrete gaming graphics-card segment. As the AI market expands, it drives demand for diversified product offerings—NVIDIA’s use of SRAM in inference-focused chips is a typical example of this trend.

Technical barriers and cost control are key to mass production

Despite the attention on market prospects, vertically stacked GDDR technology is still in an early stage. So far, it has only been seen in academic papers and cutting-edge technology research, with no prior mass-production precedent.

Industry insiders note that Micron faces multiple technical challenges, including how to implement stacking interconnects between GDDR chips, power consumption management, and heat dissipation control. On the cost side, the additional manufacturing expenses introduced by the stacking process are equally not to be ignored. If the product cannot maintain a sufficiently strong cost-performance advantage versus HBM, its competitiveness in the market will be greatly weakened.

An industry source said:

“In the past, aside from technical issues, the unclear market positioning of stacked GDDR was an important reason why commercialization was delayed for so long. As the landscape of the AI memory market evolves, the necessity of stacked GDDR is starting to stand out. Micron’s attempt this time could become the starting point of a new round of memory stacking competition.”

Racing ahead of Samsung and SK hynix to move into an emerging niche

Micron’s entry into stacked GDDR clearly reflects a first-mover strategy. As of now, Samsung Electronics and SK hynix have not publicly announced similar plans. If Micron can achieve a technological breakthrough first and complete commercialization, it will establish competitive barriers in this segment.

From the perspective of market scale, stacked GDDR is still a niche market today. But as AI inference applications continue to become more widespread and demand for diversified accelerator hardware grows, its potential should not be underestimated.

Micron has obviously incorporated this judgment into its product strategy, choosing to make its move before the market landscape becomes clear.

Risk disclosure and disclaimer

        There are risks in the market; invest with caution. This article does not constitute personal investment advice, nor does it consider any specific user’s special investment objectives, financial situation, or needs. Users should consider whether any opinions, viewpoints, or conclusions in this article align with their specific circumstances. Invest at your own risk; you bear responsibility accordingly.