The ledger remembers what the market forgets. On a Tuesday morning in early 2025, JPMorgan global market strategist Gabriela Santos stood before a quiet room of institutional investors and said what many had suspected but few had dared to state publicly: the era of concentrated AI bets is over. Her recommendation—diversify AI investments across regions, sectors, and asset classes—landed like a block confirmation in a congested mempool. The market had been pricing AI as a single narrative; Santos was arguing that the narrative had already fractured into a thousand subplots.
For those of us who have spent years auditing DeFi protocols, the parallel is immediate. In DeFi, we learned the hard way that putting all liquidity into a single yield farm—even one audited by the best firms—is a recipe for disaster. The same logic applies to AI. The ledger remembers what the market forgets: concentration conceals risk until the moment of failure.
Context: The Mainstreaming of AI as an Asset Class
Santos’s statement is not a casual remark. JPMorgan is the largest bank in the United States by assets, and its global market strategist team influences trillions in capital allocation. When a JPMorgan strategist publicly advocates for diversification, it signals that the first phase of AI investing—the infrastructure-led boom of 2023–2024—has likely priced in most of the easy gains. The next phase requires a more surgical approach.
AI investment has evolved rapidly. In 2023, the narrative was simple: buy NVIDIA, buy Microsoft, buy any company that touches large language models. By 2025, that simplicity has eroded. The AI value chain now spans chips, cloud services, foundation models, specialized applications, data pipelines, and AI governance tools. Each segment moves on its own cycle. The tech market’s S&P 500-style concentration risk is now repeating in AI: a handful of mega-cap stocks account for the majority of AI-related market cap gains.
Santos’s recommendation reflects a growing consensus among institutional investors that the AI opportunity set is no longer a single sector but a multi-dimensional ecosystem. The question is whether this diversification is a tactical rebalancing or a structural shift.
Core Analysis: Seven Dimensions of AI Diversification
1. Technology Route Dispersion
The assumption underlying Santos’s strategy is that AI technology is not a winner-take-all game. While large language models have dominated headlines, specialized models, edge AI, and symbolic AI are gaining traction. The technical drivers for chips, models, and applications are misaligned in time: model breakthroughs typically precede application adoption by 12–24 months. This temporal gap creates opportunities for investors to rotate along the value chain.
Formal verification is the only truth in code. In my own audits of AI-driven smart contracts, I have seen how a single model architecture choice—Transformer vs. state-space model—can introduce entirely different security profiles. Diversification across technology routes is not just financial prudence; it is a hedge against paradigm shifts that could render entire model families obsolete.
2. Commercialization Maturity Gradient
Santos’s call is a classic signal that AI commercialization is moving from proof-of-concept to industrial diffusion. According to PitchBook, global AI venture capital reached an all-time high in 2024, but the flow has shifted from foundation model companies to application-layer startups. The unit economics of vertical AI solutions—healthcare diagnostics, financial fraud detection, manufacturing quality control—are being validated. Diversification here means capturing the long tail of use cases that are still undervalued.
3. Industry Impact Dispersion
The AI adoption curve varies dramatically across sectors. Financial services and technology have already crossed 50% adoption, while manufacturing and construction are still in single digits. This gradient is a massive mispricing opportunity. Capital directed toward underpenetrated industries can earn a first-mover advantage. However, the risk is that these industries may face structural barriers—regulatory, cultural, or infrastructural—that slow adoption.
4. Competitive Landscape Fragmentation
At the foundation model layer, the competitive window is narrowing. The top players—OpenAI, Anthropic, Google—are burning capital at a rate that makes new entrants nearly impossible. But in the application layer, the market is fragmented. No single AI assistant or vertical solution has achieved dominance. This is the ideal environment for diversification: many horses, each with a chance to win.
5. Ethical and Security Considerations
Every blockchain security auditor knows that centralization is a vulnerability. In AI, concentration of model power in a few corporations creates systemic risks—bias amplification, single points of failure, and regulatory backlash. Diversification, even if motivated by financial risk management, has the side effect of reducing ethical concentration. The EU AI Act’s risk-based approach aligns with this logic: different applications require different oversight.
6. Valuation and Investment Strategy
This is the core of Santos’s argument. AI valuations have expanded dramatically. The forward P/E of the AI basket is well above the broader market. In such an environment, stock-picking error is costly. Diversification reduces the impact of any single miss. Moreover, the first phase of AI investing (2023–2024) was a beta trade—buy the whole sector. The second phase (2025–2027) is an alpha trade—select winners within segments. Diversification is the bridge between these two phases.
7. Infrastructure and Compute
Stress tests reveal the fractures before the flood. The AI compute landscape has shifted from extreme scarcity to relative abundance. GPU supply has improved, inference costs have dropped 80–90% since 2023, and edge AI is reducing reliance on centralized data centers. This democratization of compute makes diversification feasible. Without it, only the largest players could afford to participate.
Contrarian Angle: The Blind Spots in Diversification
Every strategy has hidden assumptions. Santos’s diversification advice implicitly assumes that AI will not experience a winner-take-all outcome. But what if AGI-level intelligence emerges from a single model? Then diversification across weaker models is a loser’s game. Also, the diversification strategy is itself a signal of peak optimism—when strategists start recommending diversification, it often means the easy money has been made.
Chaos is just unverified data. The most dangerous blind spot is the “pseudo-diversification” trap. Many AI companies share the same risk factors: dependence on NVIDIA chips, sensitivity to interest rates, exposure to regulatory changes. Cross-sector diversification may not truly reduce correlation if all holdings are exposed to the same underlying tail risks. For example, an AI healthcare stock and an AI fintech stock both suffer if the Fed raises rates aggressively.
Another blind spot is geopolitical. Cross-regional AI diversification is appealing in theory, but the U.S.-China technology war imposes real constraints. Investing in Chinese AI companies carries the risk of export controls, while investing in European AI companies may face slower growth due to regulatory overhead. The diversification benefits may be offset by higher transaction costs and uncertainty.
From my own experience auditing DeFi protocols, I recall a 2023 incident where a supposedly diversified yield aggregator turned out to be invested in multiple protocols that all relied on the same oracle provider. When that oracle was compromised, the entire portfolio collapsed. The same can happen in AI: a diversified portfolio of AI stocks may all depend on the same cloud provider, the same model API, or the same chip supply chain.
Takeaway: What to Watch Next
Verification precedes value. The true test of Santos’s strategy will come in the next 12–18 months. Investors should watch for three signals: (1) whether more large institutions publish similar diversification frameworks; (2) whether AI application-layer revenue growth accelerates relative to infrastructure; and (3) whether the compute cost curve continues to decline, enabling broader adoption.
For blockchain and crypto investors, the parallel is clear. Just as AI is moving from a single-narrative trade to a multi-dimensional ecosystem, the crypto market is also evolving beyond Bitcoin dominance. The lessons of DeFi diversification—yield farming across multiple protocols, using different risk models, and hedging with uncorrelated assets—apply directly to AI investing.
The block height does not lie. Over the next 18 months, the market will reveal whether Santos’s advice was prescient or premature. Either way, the conversation has shifted. AI is no longer a single story. It is a library of stories, each with its own risks and rewards. And the only way to read them all is to diversify.