d-Matrix partners with NVIDIA to plug Raptor XPUs into NVLink Fusion rackscale systems
bookwormengr · x · 2026-09-12
Inference chip startup d-Matrix announced a multi-year collaboration with NVIDIA to integrate its next-gen Raptor XPUs, built on 3D DRAM stacked memory, into NVIDIA MGX rack-scale systems via NVLink Fusion, targeting ultra-low-latency premium token services for AI labs, hyperscalers, and neoclouds. The rack design spans NVIDIA's Vera CPUs, NVLink switches, BlueField-4 DPUs, ConnectX-9 SuperNICs, and Spectrum-X Ethernet, with Astera Labs providing connectivity solutions. Customers can run Raptor racks standalone or alongside NVIDIA GPUs, matching compute to each workload phase. The deal signals specialized inference silicon entering NVIDIA's ecosystem rather than competing against it.
More from Venture
- Dev says his AI token EXAMCN was launched without consent, with ~45% taken — examachine · 2026-09-13
- Transformer newsletter says its July call on an impending AI slowdown is aging well — ShakeelHashim · 2026-09-13
- 25-year design veteran breaks down text-first landing pages: Copilot Money's H1 fix — michalmalewicz · 2026-09-12
- AI firms lease 2.9M sq ft of SF office in H1 2026, topping all of 2025 — atShruti · 2026-09-12
- YC Partner: Do Your First 100 Outreaches by Hand to Learn What Actually Converts — ycombinator · 2026-09-12
- AI has created ~1M US jobs vs ~200K AI-attributed layoffs, Economist estimates — lemire · 2026-09-12