Overview
- D‑Matrix announced in June 2026 that Corsair has entered volume production and the company has started shipping four‑chip PCIe cards to committed hyperscalers, neoclouds and frontier AI labs.
- The startup claims Corsair can run some small inference tasks up to 10 times faster and use up to five times less energy than a standalone Nvidia GPU and cites Gimlet Labs research for larger hybrid gains when paired with Blackwell GPUs.
- Corsair relies on an in‑memory design that embeds SRAM next to compute to reduce data movement, which lowers latency and power but restricts on‑chip capacity compared with DRAM‑backed GPU systems.
- D‑Matrix packages four Corsair dies per server card, is working with Arista, Broadcom and Super Micro on a rack system called SquadRack, and has the chips produced at TSMC on a 6 nm process.
- Key uncertainties remain because independent third‑party benchmarks are not yet public, experts note SRAM designs cannot host the largest trillion‑parameter models, and reports conflict on how much capital the company has raised.