Micron is reportedly working on a new kind of graphics memory that could reshape how future GPUs handle AI-heavy workloads. According to a report from ET News, the company is developing vertically stacked GDDR (graphics DRAM), a move aimed at keeping up with rapidly changing demand for memory in artificial intelligence applications.
Traditional GDDR memory has long been a go-to choice for graphics cards because it delivers strong bandwidth at a cost and supply profile that works well for gaming and mainstream computing. But the AI boom is shifting priorities across the industry. As more AI tasks move onto GPUs and other accelerators, memory bandwidth, capacity, and efficiency are becoming even more critical. That shift is pushing memory makers to explore designs that can deliver more performance without requiring dramatically larger boards or higher power draw.
Vertically stacked GDDR essentially points to a packaging approach where memory is built upward rather than spread out across the PCB. By stacking memory dies, manufacturers can potentially increase density and improve data movement in a smaller footprint. For GPU makers, that could translate into more flexibility in how they design cards, especially in products that need to balance high bandwidth with space constraints, cooling limits, and power targets.
The report suggests Micron’s effort is directly tied to evolving AI memory needs. As AI training and inference workloads grow, the competition for high-performance memory intensifies. Solutions that can offer better bandwidth and capacity scaling are increasingly valuable, not only for data centers but also for professional workstations and high-end consumer hardware that’s expected to handle AI features locally.
While details like timelines, final specifications, and which GPUs might adopt stacked GDDR haven’t been confirmed in the report, the direction is clear: memory technology is adapting quickly as AI changes what “performance” means. If Micron’s stacked GDDR development progresses, it could become an important option alongside existing high-bandwidth memory approaches, helping GPUs meet the next wave of demand for faster, smarter computing.
