Inference Cost Deflation & Value Capture Challenge

DeepSeek achieves 98% cost reduction vs OpenAI, compressing inference margins. AWS CEO confirms inference now 2/3 of AI compute demand. Akash token down 94% despite real usage, highlighting value capture problem. Runware's Sonic Inference Pod claims 30-90% lower GPU-hour pricing, further pressuring DePIN's value proposition. AI ETFs see $19B inflows in 2025 but mostly into traditional AI equities, not DePIN tokens. Structural shift challenges 'more AI = more GPU demand' narrative and pressures DePIN token economics.