SF : Cloudflare IRL

Cloudflare IRL at Replicate's Mission District office, with PlanetScale joining the lineup. The timing made the room feel different. Cloudflare had announced acquiring Replicate in November, and the deal was tracking to close right around this event, so the live demos were not just product walkthroughs but a preview of what a single AI cloud actually looks like when the inference layer, the database layer, and the edge runtime stop being three separate procurement decisions. Replicate brings more than fifty thousand production ready models that will land inside Workers AI, and the engineering conversations on the floor were focused on exactly this kind of consolidation. The most useful conversations I had were with engineers building on Hyperdrive, talking through the latency wins they have been measuring against their previous Postgres setups, and with PlanetScale folks on how Vitess scales next to embedded inference paths. Demos ran continuously through the evening, which made it easier to corner people while something interesting was on screen. The shift in mindset, from stitching three providers together to building on a unified developer surface, is the part that will stick. If you have been putting off consolidating your AI infrastructure stack, this is a reasonable moment to revisit that decision. Grateful to the Cloudflare, Replicate, and PlanetScale teams for hosting and to the engineers who stayed late to actually answer questions.