Day 1 of Ethernet Alliance Technology Exploration Forum. A few key takeaways: 1. 400G/lane is becoming an architecture problem, not just a SerDes problem. PAM4 vs PAM6, FEC, copper reach, retimers and the optical handoff are all getting coupled together. 2. Moving optics closer to the ASIC may simplify the system. The harder we push long-reach copper, the more complexity we add. NPO/CPO shortens the electrical channel and avoids some of that pain. 3. VCSEL/MMF is still very much alive. 200G/lane multimode remains relevant for short-reach AI links, and “wide and slow” is still part of the discussion. 4. Fiber itself is becoming a bottleneck. Connector density, fiber count, cable routing, contamination and conduit capacity are becoming real infrastructure constraints. 5. Oracle had some of the most interesting production data. LPO at 800G appears to work better than many expected, and a huge portion of optical failures may simply come from dirty connectors. That creates a surprisingly strong reliability argument for NPO/CPO. Lots of deep discussions on Multi-Core Fiber too. Initial use cases would only fit scale-across. It has its challenges. I'll write a full report at a later time.