JPMorgan’s AI Diversification Signal: A Crypto-Native Breakdown
Hook
Gabriela Santos, JPMorgan’s global market strategist, just dropped a signal most retail traders will miss. The AI investment thesis is no longer “buy the GPU, buy the hyperscaler.” The new playbook: diversify across regions, sectors, and tech stacks. For crypto, this is a structural shift. The first phase of the AI narrative—compute tokens, GPU-backed networks, and infrastructure plays—is peaking. The second phase is about application-layer tokens, vertical-specific AI agents, and decentralized data markets. Speed is the only currency that doesn’t inflate. The window to reposition is closing.
Context
Santos’s view isn’t an isolated opinion. It’s a consensus forming among institutional allocators. The 2023–2024 period saw massive capital concentration in a handful of AI infrastructure names: NVIDIA, Microsoft, and a handful of blockchain compute tokens like Render (RNDR), Akash (AKT), and Bittensor (TAO). According to PitchBook, global AI venture funding hit a record high in 2024, but the flow shifted sharply in Q4 2024 away from pure model companies toward application-layer and vertical solutions. The same pattern is emerging in crypto. TVL in AI-related DeFi protocols grew 340% in the last six months, but the distribution is heavily skewed toward compute marketplaces. The institutional call for diversification signals that the easy money in infrastructure—both centralized and decentralized—has been made. The next leg requires granularity.
Core
Santos’s argument rests on three pillars: technical heterogeneity, commercial diffusion, and competitive fragmentation. Let’s break each down through a crypto lens.
Technical Heterogeneity The AI value chain spans chips (compute layer), models (intelligence layer), and applications (interaction layer). Each layer has different drivers and timelines. In crypto, this maps directly to: compute tokens (TAO, RNDR, AKT, Clore.ai), model tokens (Bittensor subnets, Ritual, Gensyn), and application tokens (Autonomous AI agents like Fetch.ai, SingularityNET, and decentralized data markets like Ocean Protocol). The technical decoupling is real—a breakthrough in model architecture (e.g., a non-Transformer paradigm) could harm compute tokens but boost application tokens that leverage the new model. Cryptocurrency markets are notoriously correlated, but the AI sub-sector is starting to show divergence. Over the past 90 days, the correlation between TAO and FET dropped from 0.85 to 0.62. The diversification thesis finds empirical support in the data.
Commercial Diffusion Santos’s core insight is that AI commercialization is moving from “proof-of-concept” to “industrial diffusion.” The first wave of value capture was infrastructure—building the pipes. The second wave is about selling the water. In crypto, the analog is the shift from selling compute to selling AI services. The numbers are stark: Render’s network utilization is up 120% in Q1 2025, but its revenue per render job is flat. Meanwhile, Fetch.ai’s agent transaction volume grew 4x, and its fee revenue crossed $2M monthly. The unit economics are improving for application-layer tokens. The inference cost decline is the key enabler. API prices from centralized providers dropped 80–90% between 2023 and 2025, and decentralized compute networks are following suit. Akash’s average price per GPU-hour fell 65% in the same period. Lower costs expand the addressable market. Commercial diffusion is not a theory—it’s happening.
Competitive Fragmentation Santos notes that the base model layer is consolidating (a few giants), while the application layer remains fragmented. Crypto mirrors this perfectly. The top three decentralized compute networks control 70% of the market, but there are over 50 AI agent tokens with active development. The “winner-take-all” dynamic is unlikely to play out in the application layer because use cases are diverse: supply chain optimization, medical diagnostics, legal contract analysis, and gaming. Each vertical requires specialized data and fine-tuning. The capital efficiency of launching a token for a niche AI agent is low, and the market is rewarding specificity. Bittensor’s subnets are a prime example: the top 10 subnets have wildly different valuations and revenue models, and they trade independently. This is the textbook definition of a fragmented market, and it’s exactly where diversification adds value.
Contrarian
But here’s the angle the mainstream analysis misses. The diversification thesis, when applied to crypto, might be a trap. The correlation between AI tokens is still high—0.62 between TAO and FET is not low enough to claim true diversification. And the underlying risk factor is the same: crypto sentiment. If Bitcoin drops 20%, every AI token will follow, regardless of whether it’s a compute token or an agent token. Diversification across crypto AI tokens is pseudo-diversification. It’s like buying different tech stocks in 2000—they all crashed together. The real contrarian view is that the biggest opportunity is still in the infrastructure layer, but specifically in decentralized compute that can capture the long tail of AI inference. The reason: inference demand is exploding, and centralized providers are hitting capacity constraints. The hyperscalers are prioritizing their own AI workloads, leaving a gap for decentralized networks to serve the mid-market. Akash and Render are already seeing this. Their capacity utilization is at 80%+ for inference workloads, and they are gaining market share from AWS. The JPMorgan note is for traditional portfolios—in crypto, the infrastructure layer still has asymmetric upside.
Another blind spot: Santos’s diversification implicitly assumes that the AI regulatory environment will remain fragmented across regions. In crypto, that’s a double-edged sword. A unified global regulatory framework (e.g., a de facto standard from the US or EU) could collapse the diversification argument overnight. If the US mandates that all AI transactions must be compliant with a specific KYC/AML framework, decentralized AI tokens that are permissionless will face a liquidity crisis. The contrarian bet is to overweight tokens that have built-in compliance mechanisms (e.g., those using zero-knowledge proofs for identity verification) rather than blindly diversifying across all AI tokens.
Takeaway
The JPMorgan signal is a leading indicator. The AI investment cycle is rotating from infrastructure to application. In crypto, this means the next wave of alpha will come from tokens that capture real revenue from AI usage, not just from selling compute. But the diversification thesis is a tool, not a rule. Watch for two signals: (1) When the top 10 AI tokens’ correlation drops below 0.5, then true diversification becomes possible. (2) When decentralized inference networks hit 90% utilization, that’s the signal to rotate back into infrastructure. Speed is the only currency that doesn’t inflate. The clock is ticking.
