Technology thesis · Semiconductors & Chips
low conviction conceptMemory-centric computing
The memory wall is now the dominant AI bottleneck, but it is being solved by HBM4 and CXL bandwidth scaling, not true processing-in-memory, which stays niche through 2027.
Position maintained continuously · last reviewed Jun 24, 2026
The thesis
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.
Everything below is live inside CanaryIQ
The full analysis behind the verdict — the structure is real; the content unlocks when you log in.
Signal stack
Evidence stacked leading → lagging
Technology-native KPIs
Metrics that predict trajectory, tracked over time
Landscape map
Who builds what — and who depends on whom
Catalyst calendar
Dated events that will move the position
Technology roadmap
Milestones on the path to maturity
Watchlists
Companies, people and papers — each with a remove-by condition
Decision frameworks
The same call, framed for your desk
Thesis changelog
When our view changed, and why
Change our mind
2 disconfirming conditions
The rest is inside
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 the changelog of every time our call on Memory-centric computing has changed — all live inside CanaryIQ.