Memory-centric computing
Memory-centric AI computing is being pursued in several ways, including more bandwidth, memory-tiering software and processing near the data. SK hynix demonstrated tiered cache management and processing-in-memory in September. Neither demonstration shows that the memory bottleneck is solved.
Research snapshot · updated Nov 12, 2026
Core thesis
Data movement consumes 90%+ of energy in AI inference. Processing-in-memory eliminates this by computing where data lives. Samsung HBM-PIM, UPMEM, and Mythic pursue different approaches. If it works at scale, it could be more energy-efficient than GPUs for inference. But the programming model is radically different from conventional computing, limiting software ecosystem development.
State of the art (2026)
In 2026 the field has split. The commercial winner is bandwidth scaling: SK hynix, Samsung and Micron all reached HBM4 mass production this year, feeding NVIDIA Vera Rubin, which entered full production in June, plus AMD MI400. CXL pooling and the UALink 200G interconnect are disaggregating the rack. True processing-in-memory remains the smaller story. The breakout is d-Matrix, whose SRAM-based Corsair inference accelerator entered full production in June 2026, pairing with GPUs to claim roughly 10x decode-phase speed-ups. Analog and ReRAM compute-in-memory (Mythic, EnCharge, IBM Research) stay at edge and research scale. The thesis holds at the level of memory economics, not at the level of computing inside DRAM.
Explore the deeper analysis in CanaryIQ
Explore what is covered below. These previews show the structure; log in to read the full analysis.
Signal stack
Evidence stacked leading → lagging
8 signals
Signal stack
Evidence stacked leading → lagging
Technology-native KPIs
Metrics that predict trajectory, tracked over time
3 tracked
Technology-native KPIs
Metrics that predict trajectory, tracked over time
Landscape map
Who builds what, and who depends on whom
108 players · 6 layers
Landscape map
Who builds what, and who depends on whom
Catalyst calendar
Dated events that will move the position
4 ahead
Catalyst calendar
Dated events that will move the position
Technology roadmap
Milestones on the path to maturity
8 milestones
Technology roadmap
Milestones on the path to maturity
Watchlists
Companies, people and papers, each with a condition for removing it
20 · 20
Watchlists
Companies, people and papers, each with a condition for removing it
Decision frameworks
What our view means for your job
Locked
Decision frameworks
What our view means for your job
Thesis changelog
When our view changed, and why
7 updates
Thesis changelog
When our view changed, and why
Change our mind
2 disconfirming conditions
You've read the verdict. The file is much deeper.
The full signal stack, technology-native KPIs tracked over time, the landscape of who depends on whom, the dated catalyst calendar, decision frameworks for every desk, live watchlists and a record of every time our view on Memory-centric computing has changed. All of it is in CanaryIQ.