Yeah, so silicon dollar content climbing past $1,000 per accelerator as racks move from retimers to full AI fabric — PCIe Gen 6, 800G and 1.6T Ethernet driving that. Our X-Series 320-lane switch adds Hypercast and in-network compute; management says it can boost collective ops up to 2x by offloading accelerator-to-accelerator traffic onto the switch. Initial volumes shipping, full production ramp H2 2026. We intercept the fabric ramp; what the economics look like at scale remains to be seen.