In d-Matrix’s April newsletter we talk about the rapid growth of agentic AI tools like Claude Code and Codex is driving up inference costs and straining GPU resources. In response, disaggregated pipelines use smaller, specialized models and techniques like speculative decoding to improve efficiency—delivering faster results at lower cost while maintaining high-quality outputs as AI continues to scale. Read More
In d-Matrix’s March newsletter we explain how AI infrastructure is evolving beyond SRAM toward stacked 3D DRAM, enabling significantly higher bandwidth, lower latency, and improved power efficiency for next-generation workloads. We highlight a shift to disaggregated compute pipelines—combining specialized hardware—which can deliver up to 10× performance gains over traditional GPU-only systems. We confirm real-world validation of its 3D DRAM technology (3DIMC), demonstrating a viable path to scaling AI performance for larger models and growing demand. Read More
d-Matrix’s February newsletter highlights its seven-year investment in a hybrid DRAM–SRAM memory architecture that places SRAM alongside compute to deliver ultra-low latency, energy-efficient AI inference optimized for emerging agentic, multi-model workloads. As AI shifts from massive monolithic models to networks of smaller, specialized models at scale, the company positions hybrid memory and in-memory compute as the solution to GPU bottlenecks, power constraints, and performance tradeoffs limiting next-generation AI infrastructure. Read More
Welcome to our November newsletter. This month, we’re celebrating a major milestone: a $275M Series C round at a $2B valuation to accelerate our high-performance, energy-efficient AI inference roadmap. We also highlight why modern AI workloads benefit from a disaggregated architecture, announce our new partnership with AIchip, and share our plans for Supercomputing 2025. Enjoy the read! Read More
