Overview
- Nvidia targets early 2026 for Quantum‑X InfiniBand switches rated at 115 Tb/s with 144 ports at 800 Gb/s, integrating 14.4 TFLOPS of in‑network processing and 4th‑gen SHARP, with liquid cooling.
- Spectrum‑X Photonics is slated for the second half of 2026 using the Spectrum‑6 ASIC, with the SN6810 at 102.4 Tb/s (128×800 Gb/s) and the SN6800 at 409.6 Tb/s (512×800 Gb/s), both liquid‑cooled.
- Nvidia cites end‑of‑row switch architectures that extend server‑to‑switch distances as the driver for moving from copper to optical links at speeds such as 800 Gb/s.
- The company claims co‑packaged optics cut electrical loss from about 22 dB to 4 dB and per‑port power from roughly 30 W to 9 W versus pluggable modules, with reported gains in efficiency, signal integrity, resiliency, and deployment speed based on Hot Chips presentations.
- Product timing and capabilities are aligned with TSMC’s three‑stage COUPE plan from OSFP optical engines to CoWoS CPO and ultimately on‑package photonics, and Nvidia is positioning CPO as a competitive differentiator for future AI data centers.