Hook
When JPMorgan's global market strategist Gabriela Santos publicly urged investors to diversify their AI bets across regions and industries, it wasn't just a portfolio recommendation—it was a signal that the AI narrative has entered a new phase. But here's the question that no one in the mainstream financial press is asking: What if the most powerful diversification opportunity lies not in spreading your NVIDIA shares across geographies, but in the collapsing wall between artificial intelligence and blockchain? Over the past 12 months, the market cap of AI-focused crypto tokens has surged over 300% while the broader crypto market remained sideways. Yet institutional investors still treat AI and crypto as separate silos. That is a mistake. The convergence of AI and decentralized infrastructure is not a fringe narrative; it is the next logical evolution of both industries, and the diversification playbook Santos outlined applies here with even greater urgency.
Context
To understand the significance, we need to unpack Santos's core thesis. In her analysis, she argued that AI investment is moving from a 'concentration phase'—where the entire value chain was captured by a handful of infrastructure giants like NVIDIA and the hyperscalers—to a 'dispersion phase' where growth is increasingly driven by application-layer deployment, vertical industry adoption, and regional differentiation. She recommended cross-region and cross-sector allocation to mitigate the risk of a single technological or regulatory shock. This is sound advice for traditional equities. But in the crypto AI space, the same logic applies with an added layer of complexity: the assets themselves are not just stocks but tokens that represent both ownership of a protocol and a claim on future network activity. The crypto AI ecosystem already mirrors the value chain Santos described: there are decentralized GPU networks (Render, Akash), on-chain inference platforms (Bittensor, Autonolas), data provenance and verification protocols (Ocean Protocol, Filecoin for AI), and emerging AI agent marketplaces. Each of these sits at a different stage of maturity, faces different regulatory risks, and is driven by different technical and economic drivers. The diversification opportunity is real, but it requires a framework that goes beyond simple sector labels.
Core
Let me walk you through the seven dimensions that matter for crypto AI diversification, drawing from my own experience auditing smart contracts for decentralized compute projects and building educational content for thousands of learners.
Technology Routes: Not All AI Is Built on the Same Chain
The crypto AI sector is not monolithic. The technology stack splits into at least three distinct routes: (1) decentralized compute networks that aggregate idle GPUs for training and inference, (2) on-chain AI agents that perform autonomous tasks using smart contracts, and (3) data and model provenance layers that ensure verifiability and sovereignty. Each has a different risk profile. For example, decentralized GPU networks are highly sensitive to the supply-demand balance of hardware—when NVIDIA's Blackwell shipments soared in 2025, the utilization rates of some networks dipped, causing token price volatility. On-chain agents, on the other hand, depend on the maturation of the underlying blockchain's scalability—if Ethereum's L2s fail to achieve sub-second finality, agent-based use cases stall. Data provenance protocols are less capital-intensive but face adoption hurdles because enterprises are slow to trust blockchain-based data audits. The key insight is that these three routes are not perfectly correlated; a shock to one (e.g., a hardware breakthrough that makes centralized GPUs cheaper) may leave the others unaffected. That is the foundation of any meaningful diversification.
Commercialization: From Testnet to Revenue
The first phase of crypto AI (2023–2024) was dominated by speculative token launches and testnet incentives. That phase is ending. Based on publicly available data from leading projects, several have crossed the threshold into real revenue. Render Network reported over $10 million in annualized compute fees by mid-2025, driven by AI rendering workloads. Bittensor's subnet mechanism has attracted developers who pay for inference in TAO, creating a self-sustaining economic loop. But here is the nuance: revenue is not evenly distributed. The largest share flows to the most liquid networks with the strongest developer ecosystems. This is exactly where Santos's 'dispersion' logic applies—the next wave of value will come from vertical-specific AI solutions built on top of these base layers. For instance, a decentralized medical imaging AI model on Akash may generate predictable revenue years before a general-purpose model does. Diversification across use cases (healthcare, finance, logistics) reduces the risk of betting on a single application that fails to find product-market fit.
Competition: The Landscape Is Still Fragmented
In traditional AI, the competitive landscape is converging around a few large models and cloud providers. In crypto AI, the opposite is true: fragmentation is a feature, not a bug. There are over 200 live AI-related tokens, and the top 10 account for only 40% of the total market cap. This is both a risk and an opportunity. The risk is that many projects will fail—tokenomics are often poorly designed, and governance battles can paralyze progress. The opportunity is that a diversified portfolio of 10–15 carefully selected protocols can capture the upside of the winners without being wiped out by the losers. From my own experience researching tokenomics for a crypto education platform, I've seen that the projects with the strongest community governance and transparent treasury management tend to outperform those with flashy tech but no distribution. Community is not a user base; it is a shared soul. That is why I advise investors to look beyond the whitepaper and examine the on-chain activity of the DAO—does it vote on upgrades? Does it fund ecosystem grants? That is the true signal of long-term viability.
Investment and Valuation: The Elephant in the Room
Santos's implicit warning about AI stock valuations applies even more acutely to crypto AI tokens. The average fully diluted valuation of the top 20 AI tokens is over $5 billion, yet many have less than $100 million in annualized revenue. That is a price-to-sales multiple north of 50x—even by crypto standards, that is frothy. The diversification thesis here is not about chasing alpha; it's about risk management. If you own a concentrated position in a single AI token and it gets hit by a protocol exploit, a regulatory crackdown on decentralized compute, or a sudden shift in the underlying model architecture, you lose everything. A basket of tokens across different layers (compute, agent, data) with different geographic exposures (e.g., projects based in the EU vs. Asia vs. US) can reduce the impact of any single event. Moreover, the correlation between crypto AI tokens and the broader crypto market is still high—around 0.7 in recent months—but it is falling as these projects establish their own revenue streams and user bases. The next 12 months will likely see a decoupling, making diversification even more valuable.
Contrarian Angle
But let me offer a counter-intuitive perspective that Santos likely did not consider: diversification in crypto AI may be a trap if it lulls investors into a false sense of security. The reality is that many of these projects share a common underlying dependency: the availability of cheap, unregulated GPU compute. If a major government (say, the US or China) imposes export controls on GPUs that are used in decentralized networks, the entire sector could suffer a systemic shock. No amount of diversification across tokens will protect you if the whole asset class is exposed to a single geopolitical event. Similarly, the regulatory landscape is still murky: the EU AI Act classifies some AI models as high-risk, and if decentralized inference networks are deemed to require licenses, many projects could be forced to shut down or relocate. The true hedge is not just diversification across tokens, but across asset classes—maintaining a portion of your portfolio in cash, bonds, or even Bitcoin, which has a lower correlation to AI-driven narratives. We build not for the token, but for the tribe. But the tribe must survive the winter.
Another blind spot is the assumption that 'AI' is a single category. In reality, the crypto AI tokens that rely on large language models (LLMs) have a very different risk profile from those that focus on computer vision or reinforcement learning. If the next breakthrough in AI bypasses the transformer architecture entirely, LLM-based protocols could become obsolete overnight. Diversification across different model types is essential, but few investors have the technical depth to evaluate this. That is where education becomes the ultimate utility—without understanding the underlying tech, you're just gambling on brand names.
Takeaway
Santos's advice to diversify AI investments is a wise starting point, but it is incomplete. The crypto AI sector offers a unique laboratory where the same principles apply with higher stakes and faster cycles. The next 18 months will separate the tribes from the tokens. The protocols that survive will be those that combine robust technology with resilient communities and adaptable governance. The question is not whether to diversify, but whether you are investing in the right protocols—and whether you have the patience to hold through the volatility that comes with building the future of decentralized intelligence. As I tell my students: trust is the only real asset, and in crypto AI, that trust is earned one block at a time.
