Why Big Tech Is Hoarding Custom AI Inference Chips

For years the AI chip race has been about training—but the bottleneck is shifting to inference, the moment an actual user asks a model a question. This episode explains why Meta, Apple, and Amazon are now designing their own inference silicon, and why that could reshape the economics of the entire cloud. Hosts Lucas and Luna break down a single surprising data point: Meta's capital expenditure guidance for 2026, which suggests the company will spend more on inference than training by year-end. They also touch on what this means for NVIDIA's dominance and for the next wave of consumer AI features that will run on your phone or laptop instead of in a distant data center. #AIInference #CustomSilicon #Meta #Apple #Amazon #NVIDIA #BigTech #Semiconductors #DataCenters #ChipDesign #CloudComputing #EdgeAI #CapitalExpenditure #AITraining #Hardware #Technology #FexingoBusiness #BusinessPodcast Keep every episode free: buymeacoffee.com/fexingo