Could the same computation that powers AI also secure a blockchain? Miners are paid to secure networks through computation. AI providers are paid to produce useful outputs for customers. Proof-of-Useful-Work (PoUW) attempts to combine the two by making useful computation part of mining. Earlier designs struggled to make useful work verifiable, difficulty-adjustable, and tightly coupled to the chain. Recent progress has made PoUW more technically credible. The bullish thesis is a flywheel between AI demand and blockchain security. AI customers provide external revenue for compute providers. Token rewards can subsidize their costs, enabling lower prices, which may attract more demand and compute supply. But AI demand does not automatically strengthen the blockchain. It only helps when useful work contributes directly to consensus and its revenue remains tied to participation in the canonical chain. Otherwise, providers may shift their compute elsewhere. More customer revenue reduces reliance on token rewards, allowing the AI business to grow without creating equivalent demand for the token. Token emissions can also bootstrap a compute market that would not survive on customer demand alone. PoUW becomes a flywheel only if AI demand strengthens the blockchain and can replace token subsidies without pulling compute away from security.