The ledger remembers what the market forgets. In 2021, when everyone was piling into Ethereum L1 tokens, the smart money was already rotating to L2s and application chains. Last week, JPMorgan’s Gabriela Santos advised clients to diversify AI investments across regions and sectors. The parallel is not accidental. It is a structural signal that the first phase of any technology super-cycle—infrastructure buildout—is nearing its peak, and the next phase (application diffusion) demands a different playbook.
Context: The AI capex cycle from 2023 to 2025 was the crypto equivalent of the 2020-2021 L1 and GPU mining boom. Global hyperscalers spent over $200B on AI hardware. Nvidia’s data center revenue exploded from $15B to $130B. But just as Ethereum’s gas fees eventually collapsed after the Merge, inference costs for AI models have dropped 80-90% in two years. The barrier to entry for building AI applications has fallen. The result: the value capture is shifting from the pickaxe sellers (Nvidia, cloud providers) to the miners (AI application companies). This is exactly what happened in crypto when DeFi protocols and NFTs captured value after the L1 infrastructure was built.
Core Insight: Santos’s diversification advice is not about reducing risk for its own sake. It is a data-driven recognition that the correlation structure of AI assets is changing. In the infrastructure phase, every AI-related stock moved in lockstep with GPU supply news. Now, the correlation matrix is fragmenting. Healthcare AI, manufacturing AI, and financial AI have different revenue drivers, different regulatory tailwinds, and different adoption curves. The same fragmentation occurred in crypto during 2022: the correlation between BTC and DeFi tokens dropped from 0.85 to 0.45 as the market matured. Based on my experience auditing DeFi protocols in 2020, I saw that the winning L2s were not just Ethereum clones—they were specialized chains (Arbitrum for DeFi, Polygon for gaming). The same logic applies to AI today. A diversified AI portfolio should own exposure to data infrastructure, model inference, vertical SaaS, and AI security—not just a single GPU miner.
We do not build on hype; we build on consensus. The consensus among institutional capital allocators is that the next 12-18 months will be about AI application revenue verification. The data supports this: PitchBook reports that AI venture funding in 2024 shifted to 60% application-layer deals, up from 35% in 2023. Public markets are following. Microsoft’s AI revenue run rate grew to $20B annualized, but the growth rate is decelerating. Meanwhile, niche players like Palantir and C3.ai are showing accelerating revenue from government and manufacturing contracts. This is the same pattern we saw in crypto after the 2021 infrastructure boom: DeFi protocols like Uniswap and Aave started generating sustainable fee revenue, while L1 tokens like AVAX and SOL entered a re-rating phase.
Contrarian Angle: The decoupling thesis. Santos’s diversification implicitly assumes that AI assets will remain correlated to the same macro factors (liquidity, interest rates, regulation). That is a dangerous assumption. Crypto taught us that when the macro environment tightens, even the most diversified portfolio of risk assets gets crushed together. In 2022, my 60% crypto hedge fund lost 40% in a week despite holding 10 different Layer-1 tokens. The diversification failed because the systemic risk—Fed tightening, FTX contagion—washed out all correlations. For AI, the systemic risk is not interest rates alone; it is a potential paradigm shift (e.g., a new model architecture that renders current hardware obsolete) or a regulatory bombshell (e.g., the EU AI Act’s liability provisions). The real contrarian bet is not diversification across AI sectors, but a hedge against the AI hype cycle itself—by allocating to assets that benefit from AI failure, such as cybersecurity or automation insurance. The ledger remembers what the market forgets: in 2022, the best crypto hedge was not a basket of altcoins, but shorting perpetuals on the largest exchange. The AI equivalent is shorting AI infrastructure futures or buying puts on Nvidia.
Takeaway: The market is telling us that the AI infrastructure trade is crowded. The next move is toward application-layer value capture. But do not confuse diversification with risk reduction. The real question is: which AI application subsectors have the strongest unit economics and the least regulatory overhang? Based on my experience designing the compliance framework for the first spot Bitcoin ETF, I know that regulatory clarity is the ultimate catalyst for institutional capital. The AI sector with the most regulatory clarity today is healthcare AI (FDA frameworks are established) and financial AI (SEC’s Reg SCI guidelines). The sectors with the least clarity are autonomous vehicles and generative media. The ledger remembers: the market rewards those who position into the regulated infrastructure before the rest of the herd. Follow the data, ignore the noise. Diversify, but with a clear thesis on where the next wave of liquidity is flowing.