The market didn't flinch when Gabriela Santos, JPMorgan's global market strategist, told Bloomberg last week that AI investing is entering a diversification phase. The tape barely moved. AI tokens like FET, AGIX, and RNDR continued their slide from February highs. Nvidia held steady. The crowd yawned. That's exactly when you should pay attention.
Santos didn't drop a hot take. She gave a structural signal. When a top-tier institutional strategist publicly advises spreading AI bets across regions, sectors, and market caps, she's not being clever. She's reading the same order flow I see on-chain. The first phase of AI โ the infrastructure land grab โ is pricing in its peak. The second phase โ application diffusion and vertical integration โ is where the alpha lives. But the catch is that most of that alpha won't flow through the tokens you're holding.
Where the smart money is rotating
Let's step back. Santos's thesis is simple: AI value creation is dispersing from a handful of GPU makers and hyperscalers into a broader set of industries โ healthcare, finance, manufacturing, and emerging markets. The logic is sound. If you look at the 2024-2025 capital flows, PitchBook data shows venture funding shifting from pure-play foundation model companies to AI application layers and vertical SaaS. The money is following the use case, not the infrastructure.
But here's where the crypto parallel gets ugly. The AI token ecosystem โ Render (RNDR), Bittensor (TAO), Akash (AKT), Fetch.ai (FET), and a dozen others โ is still primarily a bet on infrastructure. Render is decentralized compute for rendering, but its AI compute use case is years away from meaningful revenue. Bittensor is a decentralized machine learning network, but its tokenomics are opaque and the network's actual ML output is dwarfed by centralized alternatives. Akash is a cloud marketplace, but its AI workload adoption is negligible. These are not diversified bets. They are leveraged bets on the same narrative: that decentralized infrastructure will somehow capture a slice of the AI boom.
I've been in this space long enough to remember the 2017 ICO audit sprint. I reverse-engineered Golem's smart contract during that era and found an integer overflow that could have drained 15% of the raised funds. The lesson was clear: code is law, but human greed is the bug. Today, many AI token projects have the same shiny wrapper โ decentralized, permissionless, community-owned โ but the underlying code and business models are still in the pre-revenue stage. The 2020 DeFi yield farming experiment taught me that impermanent loss is a brutal reality check. The 2021 NFT floor sweep โ buying 12 CryptoPunks at $1.2 million โ taught me that holding through the dip requires a spine of steel. The 2022 Terra Luna collapse taught me that official narratives are worthless when the math breaks. And the 2024 ETF arbitrage taught me that institutional-grade strategies require precise execution, not hype.
All of this experience points to one conclusion: the AI token sector is a concentrated bet on a single narrative, not a diversified portfolio. Santos is telling you to spread your risk. The crypto market is telling you the opposite โ it's all-in on AI tokens. That asymmetry is a red flag.
Decoding the order flow
Let's look at the on-chain data. I've been tracking whale wallets for AI tokens since late 2024. The pattern is clear: large holders are distributing to retail. FET's top 100 wallet concentration dropped from 68% to 52% between November 2024 and March 2025. TAO saw a similar trend. Meanwhile, retail inflow into AI token pools on decentralized exchanges surged. The classic retail-whale divergence. Smart money is taking profits, and the crowd is chasing the narrative.
The same dynamic is playing out in traditional equities. Nvidia's P/E ratio is over 50, and while its earnings are strong, the market is discounting future growth that may not materialize at the same rate. Microsoft's AI revenue is growing, but it's still a fraction of its cloud business. Santos's recommendation to diversify is essentially a warning that the easy money in pure AI plays has been made. The next leg requires granular analysis of which verticals actually adopt AI profitably.
For crypto, the implication is starker. The AI token sector has a total market cap of roughly $30 billion, but the underlying revenue is negligible. Render's annualized revenue from its compute network is under $10 million. Bittensor's subnet rewards are mostly emission tokens, not real earnings. These are not assets; they are lottery tickets. And the lottery is rigged โ the house always wins.
The retail vs. smart money chasm
The contrarian angle here is that the biggest risk isn't that AI fails. It's that AI succeeds โ but the value accrues to traditional equities, private markets, and maybe a few well-positioned crypto infrastructure projects that nobody is talking about. The real AI winners in crypto will likely be the boring ones: data availability layers (Celestia), decentralized storage (Filecoin, Arweave), and settlement layers (Bitcoin, Ethereum). These are the "picks and shovels" of the AI gold rush, not the gold itself. The "sell water to miners" strategy is timeless.
Retail traders are fixated on the AI token narrative because it's exciting. They see Nvidia's 500% run and want a crypto version that will 10x in a week. But the smart money is rotating into assets with real yield, real users, and real regulatory clarity. Real-world asset tokenization, stablecoins, and DeFi lending are quietly absorbing the liquidity that was once chasing AI tokens. The data doesn't lie: total value locked in AI-related DeFi protocols is flat, while lending protocols like Aave and Maker are hitting new highs.
What the JPMorgan report really means
Santos's report is a signal that the AI investment cycle is maturing. In the early stage (2022-2024), the optimal strategy was to concentrate on the infrastructure leaders โ Nvidia, hyperscalers, and the few foundation model companies. That's when you buy the monopolist. Now, the market is transitioning to a diffusion phase where value creation spreads across many industries. The winners will be harder to pick, and the risk of overpaying for a single stock is higher. Diversification becomes a risk management tool, not a return-enhancing strategy.
For crypto, the parallel is that the infrastructure phase of the AI token boom is over. The next phase will be about applications that actually solve real problems โ not decentralized compute for AI training (which is still a fantasy), but AI agents that execute trades, AI-based credit scoring for lending, or AI-driven content verification on decentralized social networks. Those applications are being built, but they are not yet tokenized. The tokenized AI infrastructure we have today is a speculative narrative, not a productive asset.
I've seen this movie before. In 2020, everyone was yield farming on Uniswap V2, chasing 300% APYs. I deployed $20,000 and achieved 340% APY for three months before the pool diluted. The profit was real, but it was a product of early mover advantage and high risk. The same goes for AI tokens today. The early movers โ those who bought TAO at $30 or RNDR at $1 โ made outsized gains. But the latecomers are buying at 10x the price with no revenue growth to justify it. The order flow is telling you that the smart money is exiting.
The liquidity fragmentation myth
Let me address a common narrative pushed by VCs in the AI token space: "Liquidity fragmentation is a problem that needs to be solved by our new bridging protocol." That's nonsense. Liquidity fragmentation is a feature of a decentralized ecosystem, not a bug. The real problem is that most AI token projects have no real demand for their tokens. They create supply and hope retail buys the story. The 2024 ETF arbitrage taught me that clean, risk-free profits exist in gaps between spot and futures. Those gaps are not found in AI tokens; they are found in basis trades on Bitcoin and Ethereum. The institutional money is smart enough to know that.
Santos's recommendation to diversify across regions and sectors is a tacit admission that the easy alpha in AI is gone. The same applies to crypto AI tokens. The next 12 months will see a brutal shakeout. Most AI tokens will go to zero. A few will survive and become the infrastructure of the next cycle. But you can't just buy the index and hope. You need to audit the code, check the revenue, and understand the tokenomics. Most investors don't do that. They rely on hype and influencer endorsements. That's why they lose.
Actionable levels and the takeaway
So what do you do? First, recognize that the AI token thesis is a crowded trade. The best time to buy was 2023. The best time to sell is now. If you're holding significant positions in AI tokens, consider reducing exposure by 50% and rotating into blue-chip crypto assets (Bitcoin, Ethereum) and DeFi protocols with real yield. The risk/reward is not in your favor.
Second, watch for the next catalyst. The AI narrative will get a temporary boost when a major tech company announces a partnership with a crypto AI project. But that's noise. The real signal is when the JPMorgan report's full details come out โ specifically the weightings and the asset classes recommended. If they recommend exposure to AI through private equity or venture funds, that's a sign that public AI tokens are seen as too risky.
Third, set your price levels. For TAO, a break below $200 would confirm the distribution pattern. For FET, $0.50 is a critical support. If those levels fail, the drawdown could be 70-80%. For Render, $5 is the line in the sand. Below that, the infrastructure narrative collapses.
Risk is the only currency that never depreciates. That's what 2022 taught me. When the Terra Luna collapse happened, I shorted Luna futures based on my intuition about the algorithmic stability's fragility. I closed positions at the peak and secured $150,000 while others lost everything. The lesson was simple: act on real-time data, not official narratives. The JPMorgan report is real-time data. The retail enthusiasm for AI tokens is a narrative. Don't confuse the two.
Volatility isn't risk, it's opportunity. But only if you have the discipline to act when the crowd is wrong. Right now, the crowd is all-in on AI tokens. The smart money is rotating out. The on-chain data, the institutional signals, and the valuation math all point to the same conclusion: it's time to diversify.
Speculation ends where strategy begins. Santos's recommendation is a strategy. The AI token mania is speculation. The choice is yours.
Holding through the dip requires a spine of steel. But the dip hasn't even started yet. When it does, the steel will be tested. Make sure you have enough dry powder to buy back in at lower prices. That's the real alpha.
The great AI diversification play is not about spreading your bets. It's about recognizing that the game has changed. The infrastructure phase is over. The application phase is beginning. The only question is whether you're still holding the shovels when everyone else is digging for gold. I'm not.
Let the crowd have their tokens. I'll take the liquidity.