AI's Decade-Long Cycle: A Macro Liquidity Mirage?

Franklin Templeton's David Dudley reiterates his bullish stance on AI infrastructure spending, calling it a 'decade-long cycle'. The statement reverberates through both traditional and crypto markets, but the on-chain data tells a different story about capital flows. Context demands a dissection of the underlying assumptions. Dudley's optimism aligns with rising capital expenditure from hyperscalers—Microsoft, Google, Amazon collectively pledged over $200 billion in 2024 for data centers and GPU clusters. The narrative is seductive: infinite demand for compute, scaling laws sustaining moats, and a new industrial revolution. Yet, when we map this to crypto, the picture is fragmented. DePIN projects like Render Network, Akash, and io.net have absorbed significant speculative capital, trading at multiples that imply decades of future revenue. But the actual utilization of these networks remains low—often below 30% for GPU compute marketplaces. From my fund's analysis, we've identified a pattern: the AI narrative inflates valuations, but real organic demand lags behind. Core insight: AI infrastructure spending is a macro liquidity phenomenon, not a purely technological one. The current cycle is fueled by zero-interest-rate legacy—excess capital from tech giants seeking growth arbitrage. When we examine on-chain metrics, we see stablecoin flows rotating away from DeFi and into AI-themed tokens. This is not a healthy sign; it's a rotation of speculative capital from one narrative to another. In 2021, the same pattern played out with NFTs and gaming tokens. Now, AI compute tokens are the new shiny objects. The 'rug pull' here is not necessarily malicious but structural: projects raise capital via token sales, build infrastructure, but fail to achieve product-market fit. The tokenomics often rely on continuous inflation to pay node operators, creating a loop that resembles a slow-moving liquidity drain. Based on my experience auditing Uniswap V2's constant product formula, I learned that any system with asymmetric incentives eventually collapses if the underlying demand doesn't sustain the supply. AI tokens are no different—they are essentially non-dividend assets, where the only hope for holders is a greater fool. The contrarian angle: The decoupling thesis—AI infrastructure spending may not benefit crypto directly. In fact, it could represent an exogenous shock that siphons liquidity from the crypto ecosystem. The narrative that AI and crypto converge is overhyped by marketing teams. The real synergy lies in decentralized verification of AI computations (zk-proofs) and decentralized physical infrastructure networks (DePIN). But the current generation of AI tokens are little more than governance tokens without dividends—a Ponzi-like structure where early adopters prey on latecomers. This mirrors our earlier analysis of DAO governance tokens: they offer no claim on revenue, only governance rights, which are rarely exercised effectively. The market is pricing in a future that may not materialize for 5-10 years, if ever. Takeaway: Position for the cycle, but separate signal from noise. The AI super-cycle is real, but the crypto-side beneficiaries are few. Focus on protocols with genuine utility, like those offering verifiable compute or decentralized data storage. Use technical signals: track LP outflows from AI token pools, monitor token velocity, and compare network revenue to market cap. The current sideways market is perfect for positioning—accumulate when others are euphoric and sell into liquidity when macro conditions tighten. The decade-long cycle may be true, but the crypto leg of it will be a story of survival, not abundance.