News

[News] NVIDIA Teams up with TSMC for Silicon Photonics Switches, Expected in 2025-26


2025-03-19 Semiconductors editor

At GTC 2025, NVIDIA unveiled Spectrum-X and Quantum-X silicon photonics switches, scaling AI factories to millions of GPUs while cutting energy and costs. Taiwan’s supply chain plays a key role, with TSMC’s COUPE (Compact Universal Photonic Engine) integrating 65nm electronic and photonic ICs in the switches using SoIC-X packaging, per Tom’s Hardware.

TSMC’s Progress on CPO

According to Economic Daily News, TSMC has integrated CPO (Co-Packaged Optics) with CoWoS and SoIC advanced packaging. It plans to enter the 1.6T optical era in late 2025 and ramp up shipments in 2026, aligning with NVIDIA’s rollout, the report adds.

NVIDIA plans to launch the Quantum-X InfiniBand switch in late 2025 and the Spectrum-X Photonics Ethernet switch in 2026, according to NVIDIA’s press release.

Media reports highlight this as NVIDIA’s first major CPO product, marking the AI era of optical communication. A TechNews report notes NVIDIA’s silicon photonics ecosystem spans Taiwan’s TSMC, Foxconn, SPIL, Browave, and global players like Coherent, Corning, Lumentum, and SENKO.

Quantum-X Photonics Switches Debut with Liquid Cooling

In detail, NVIDIA silicon photonics networking switches are available as part of the NVIDIA Spectrum-X Photonics Ethernet and NVIDIA Quantum-X Photonics InfiniBand platforms.

According to NVIDIA, the Spectrum-X Ethernet networking platform delivers superior performance and 1.6x bandwidth density compared with traditional Ethernet for multi-tenant, hyperscale AI factories.

Meanwhile, Quantum-X Photonics switches feature 144 ports of 800Gb/s InfiniBand, liquid cooling, and 2x speed, 5x scalability over the previous generation.

Silicon photonics is a next-gen chip technology that uses light instead of electrical signals for data transmission between semiconductors. According to a previous Business Korea report, it offers data speeds hundreds of times faster than copper, making it crucial for AI data centers that require fast and efficient data transfer.

Read more

(Photo credit: NVIDIA)

Please note that this article cites information from Tom’s HardwareEconomic Daily NewsTechNews, Business Korea and NVIDIA.

Get in touch with us