Baseten CEO Tuhin Srivastava on the AI Inference Crunch, Custom Models, and Building the Inference Cloud

AI inference represents the critical bottleneck and final frontier for capturing value in the current AI landscape. Tuhin Srivastava, founder and CEO of Baseten, explains that the market is shifting from vanilla model usage toward highly specialized, custom-trained models that leverage unique user signals and proprietary workflows. With Baseten experiencing 30x growth, the focus has moved to managing an extreme compute supply crunch by abstracting infrastructure across 18 different clouds. While frontier labs compete for raw compute, the long-term competitive advantage lies in the software layer that optimizes inference, enables continuous learning loops, and integrates specialized sandboxes. Ultimately, the decreasing cost of intelligence drives higher consumption, as companies increasingly embed agentic workflows into their products to deliver superior user experiences and maintain a defensible position against model providers.