by Chris DePuy, 650 Group / October 9, 2026
This week, Broadcom announced its 2026 OCP Global Summit slate, placing its AI networking lineup — the Tomahawk 6 and Tomahawk Ultra switch families with Jericho 4, Thor NICs at 800G (Thor Ultra) and 400G (Thor 2), and the third-generation TH6-Davisson co-packaged optics — into ORV3 rack designs for scale-up, scale-out, and scale-across AI clusters at the October 12–15 show in San Jose, with 15-plus partners demoing the gear, including Celestica, Delta, and Micas running Tomahawk 6-Davisson 102.4T CPO systems.
As 650 Group (https://www.650group.com) has been tracking in its AI infrastructure research, Ethernet keeps taking traffic from proprietary fabrics in both scale-out and early scale-up deployments, and this slate is the merchant-silicon side of that shift. We think the telling number sits in the keynote promise — switch designs aimed at clusters beyond 100,000 accelerators — because it puts the scale-up Ethernet fight on this show floor. Co-packaged optics moving into demonstration Tomahawk 6-Davisson systems matters as much: bringing optics onto the switch package attacks the power and reach penalties that have kept scale-up fabrics copper-bound, and we expect CPO attach to shift from differentiator toward requirement as rack densities climb.
The session slate — joint sessions with OpenAI on multi-path reliable connections and with Arista Networks on SUE-T scale-up transport — reads as ecosystem proof rather than a solo product push.
650 Group researches several of the technologies we covered in this blog, including Data Center, Co-Location and Cloud and Enterprise.
