Two and a half years ago, custom silicon was crazy talk. AI was NVIDIA. Recorded live at W.Media Sydney 2026 in Darling Harbour, JP Vasquez, founder of AI Foundry, explains why that has changed: purpose-built inference chips can use as little as a quarter of the power while delivering three to five times more tokens per second.

JP breaks down where GPUs excel and where custom silicon takes over, why air-cooled 40kW racks like Positron AI’s can land in data centres that already exist, and how a few gigawatts of legacy capacity across Asia Pacific could be turned into AI factories. He also covers why Australia and New Zealand are well placed to serve both US and Asian markets around the clock, what it takes to get your money back on AI infrastructure, and where to start if you want to learn the space. Plus: Utes, Toyota Corollas, and hot dog or not a hot dog.