AI is moving from experimentation into business-critical production. As capable models become widely available and the gap between open and closed narrows, access to a good model is no longer the advantage. The advantage comes from how well a model is chosen, optimised and improved for your specific workload. In this session, Alan Hsia, SVP at Nebius, looks at where open and custom models earn their place in production, and where closed frontier models remain the right choice. He will explore why public benchmarks and token price can’t establish production fit, and how production feedback and proprietary data can become a durable, compounding advantage that organisations own and control. Attendees will leave with a clearer view of where open models fit in their AI strategy, and a practical way to test it against their own workloads.