Baseten CEO Tuhin Srivastava on the AI Inference Crunch, Custom Models, and Building the Inference Cloud
AI inferenceGPU capacityCustom modelsOpen source AIApplication layerEnterprise AI adoptionPost-trainingModel customizationSupply chainGeopolitical AIChip diversificationNvidia ecosystemRuntime optimizationSandboxingContinuous learningJevons ParadoxOperational excellenceLeadership in startupsAI agentsUnits of cognition
Baseten CEO Tuhin Srivastava discusses the explosive growth in AI inference demand and Baseten's 30x growth, arguing that inference is becoming the strategic "last market." He explains why the application layer will persist due to unique user signals and the ability to post-train specialized models, despite severe GPU capacity constraints. The conversation also covers the geopolitical implications of open-source models, Baseten's operational challenges at scale, and the impact of Jevons Paradox on AI consumption.