[Azalia Mirhoseini] The golden age of model-software-hardware co-design

Inference demand is projected to grow 10-100x per year, and at the scale AI is now trained and served, merely building more datacenter capacity is impractical. Meeting this demand will require new sources of performance. In this talk, I will argue that customization through co-design is the key to unlocking those gains, and that the biggest wins will come from full-stack optimization across model, software, and hardware. I will present recent research on co-design, including methods that pair LLMs with linear programming to efficiently search the vast design space spanning model, software, and hardware; new techniques for training on long-context agentic trajectories; and how AI is beginning to remove the traditional compiler layer, paving the way for an explosion of custom chips. Together, these results point to a new era in which co-design becomes the key driver of AI progress.