Yeah, no, look - we frame three AI fabrics: scale-up within a rack, scale-out across racks, scale-across data centers, right? Scale-up's a 2027 entry for us - today that's NVLink from NVIDIA plus some PCIe switching, and most Ethernet scale-up doesn't show until '27, '28. XPO bridges it: 12.8T a module, 204.8T an OCP rack, cold plate to 400W, over 100 vendors. CPO's still science experiments, very proprietary; open CPO years out, but XPO runs ten years where liquid cooling's the ticket. Sits with OSFP at 400 and 800G, not instead of. That's the way to look at it.