Nvidia and TSMC Revolutionize AI Data Centers with Cutting-Edge CPO Switches

Nvidia, in partnership with TSMC, has unveiled next-generation co-packaged optics (CPO) switches to tackle data transmission and power challenges in AI data centers. These advanced Spectrum-X CPO switches are now shipping to select partners, promising enhanced bandwidth and power efficiency critical to AI performance.

Nvidia Corporation has announced the start of shipping for its latest co-packaged optics (CPO) switches, created in collaboration with Taiwan Semiconductor Manufacturing Company (TSMC). This development addresses key data transmission and power consumption issues in AI data centers, as reported by Focus Taiwan.

Speaking at the Nvidia GTC Taipei event on Wednesday, Gilad Shainer, Senior Vice President of Networking at Nvidia, confirmed the distribution of the new Spectrum-X CPO switch to a select group of partners. These switches feature Nvidia’s cutting-edge CPO technology, achieving throughput up to 400 terabits per second, with plans to ramp up production in the latter half of the year.

CPO switches, integral for networking, incorporate optical engines directly with switch chips for efficient data transmission, significantly lowering power consumption. This innovation is crucial as AI factories expand rapidly, with bandwidth and power use in data centers being critical to optimizing AI operations.

Shainer emphasized Nvidia's commitment to CPO technology, marking a shift from traditional designs that separate optical transceivers from switches. By integrating optical engines within switch chips, Nvidia reduces power usage and transmission distances. TSMC has been pivotal in this integration process, utilizing its COUPE platform for packaging.

The seamless packaging of optical engines, switch chips, and related components streamlines production, according to Shainer. At the event, Shar Narasimhan, Nvidia's Director of Product Marketing for Data Center GPUs and AI Training, highlighted the crucial support from Taiwan’s highly skilled technology ecosystem, which is essential in meeting the soaring global demand for AI computing power.