Cisco Silicon One P200 and Cisco 8223 for 51.2T Scale Across Networking

This week, Cisco announced its new chip for 51.2T scale-across networking. The Cisco Silicon One P200 powers the Cisco 8223 switch. This announcement is being made ahead of OCP Summit 2025 next week, where scale-across AI networking is set to be a major topic. Companies need to locate compute close to power generation to minimize transmission losses and ensure they have enough power. Meanwhile, AI clusters are becoming big enough that the quest for more power will mean tying together multiple compute sites. That new multi-site and large site AI cluster deployment is pushing the industry to take scale-across networking seriously.

Cisco Silicon One P200 and Cisco 8223 for 51.2T Scale Across Networking

51.2Tbps switches are not necessarily groundbreaking these days, as we have switch silicon running at twice those speeds. Still, a challenge with scaling to multiple data centers at high speeds is that large buffers are needed to buffer all of that data in transit. That is what makes the Silicon One P200 different from your average data center switch chip.

Worth noting is that Cisco has a P4 programmable network processor to allow developers and administrators to manage flows.

Cisco showed off a few different potential deployment scenarios with the new parts.

Cisco showed the two year evolution from the Cisco 8804 chassis switch to the Cisco 8223-64EH. This one feels a bit dramatic, but the idea across the industry is consolidating to do the cross data center routing directly in these 51.2T switch platforms.

The company also said that the Cisco 8223 supports SONiC and will support IOS XR in the future. Other models in the Nexus 9000 series with Silicon One P200, will support NX-OS as well.

Final Words

Beyond the P200 and the switch, Cisco is highlighting the security features of the P200 saying that they are important in AI clusters both to prevent unauthorized access to data but also to prevent potential malicious activity.

We previously covered NVIDIA Co-Packaged Optics with Silcion Photonics for Switching and Spectrum-XGS Scale-Across, Marvell COLORZ 800, and Broadcom is also not sitting still in this space either. This is another announcement to help networks bridge geographies so compute can be colocated alongside power sources. Building enormous future training clusters means the physical distances are going up even inside mega campus data centers, so technologies like the P200 will be key.