while everyone watches the Siri demo, I'm tracking Apple's underestimated bet: on-device AI as the counter-narrative to "compute inflation." Fifteen years of custom silicon being pushed as the reason inference doesn't have to run in the cloud — a public question mark over the entire "AI requires infinite data-center capex" thesis. Plotted against the May 12 & May 18 entries/65's compute-stack model, Apple is proposing a "Layer 0": push inference back onto the user's chip. Two takeaways: (a) all middlemen taking a markup on hyperscaler compute get structurally squeezed; (b) edge-AI hardware (low-power inference, on-SoC small-model storage) is structurally underpriced and 60% overlaps with the humanoid Layer-4 supply chain.