The creative-tools frontier dropped a layer this quarter: the decisive moves happened in infrastructure—real-time pipelines, composable workflows, local inference tooling—while foundation-model fundraises confirmed that middleware, not just weights, attracts growth capital. [[c:9da179b9-ca5f-4847-b7ad-97e02411c359|Together AI]] shipped sub-500ms voice agents and GPU autoscaling that close the latency gap for synchronous creative work. [[c:37b91758-337e-4eb5-b415-c4ea4dc6a975|Comfy Org]] solved last-mile distribution by abstracting node graphs into shareable apps. [[c:4d5ce541-ed4e-4e33-97d9-03118526704e|ElevenLabs]] and [[c:68857c58-c7bf-4e96-ad5c-4ee0edef902a|Runway]] raised two of the largest creative-AI rounds in history—$500M and $315M respectively—validating enterprise API traction. [[c:df0325d4-a345-475e-a732-f97b4b1356d8|Hugging Face]] absorbed GGML and llama.cpp, cementing its position as the open-source infrastructure layer. The read: production bottlenecks—not model performance—now define competitive moats. Watch whether incumbents like [[c:859651b5-e9f3-451f-8503-c695d6182870|Adobe]] can compete on workflow velocity or cede ground to composable platforms.