Liquidity doesn't follow narratives. It follows structural bottlenecks. That's the only way to interpret Citi strategists' recent call to separate AI investing from the 'Magnificent Seven' and redirect focus to chip makers.
This isn't a minor portfolio adjustment. It's a signal that the capital flows powering the tech rally are being rerouted at the most fundamental level. For those of us who watched the 2017 ICO boom collapse under its own weight, this feels eerily familiar. Back then, capital chased the shiniest tokens until it realized the real value was in the infrastructure—exchanges, mining rigs, and stablecoins. Now, the same logic is playing out in AI.
The Magnificent Seven—Microsoft, Google, Meta, Amazon, Apple, Nvidia, Tesla—were the darlings of the 2023 AI hype cycle. They were the application layer, the ones deploying GPT-4, Gemini, and Llama. But as I argued in my 2020 DeFi composability thesis, when every player has access to the same tools, the premium on any single application collapses. The market is finally internalizing that. Citi's move to decouple AI from the Mag7 is not bearish on AI. It's a recognition that the highest alpha now sits upstream—in the chip manufacturers that own the pickaxes and shovels.
The context is pure macro-liquidity mechanics. Global M2 is still tight. Institutional capital is looking for assets with inelastic supply curves. Nvidia's H100 and B100 have fixed production runs. TSMC's advanced packaging capacity is booked solid. Unlike a software product that can be replicated at zero marginal cost, a physical chip has a hard cap on supply. This creates a pricing power that is almost trustless—it's algorithmic, driven by supply and demand, not narrative.
Skepticism isn't about whether AI is revolutionary. It's about whether the current pricing of that revolution is sustainable. The Mag7's stock prices already assume decades of dominance. But if capital shifts to chip makers, what happens to the application layer's valuation? It gets repriced downward. This is exactly the liquidity vacuum I tracked during the Terra-Luna collapse: when capital exits one peg, it searches for the next hardest asset. Here, the hardest asset is compute.
The core insight for crypto observers is this: the pattern repeats. In 2021, capital rotated from DeFi apps to L1s like Ethereum and Solana, then to gas tokens like ETH and SOL themselves. The value capture mechanism shifted from application fees to base-layer security. Now, in AI, the same rotation is happening—from application providers (Mag7) to infrastructure providers (chip makers). The crypto parallel is unmistakable: the 'Magnificent Seven' are the dApps of the AI world; chip makers are the L1s.
Liquidity doesn't care about your favorite protocol. It flows to the point of greatest inelasticity. In AI, that's the silicon. In crypto, that's the block space. This convergence suggests that crypto assets tied to compute—like Render Network's RNDR, Akash Network's AKT, or even Filecoin's FIL as a data layer—could see unexpected demand as institutional allocators seek tokenized exposure to the AI infrastructure theme. After all, if you can't buy TSMC shares through a regulated ETF, you might buy a token that represents decentralized GPU compute. The narrative aligns.
But here's the contrarian angle, and it's critical: this decoupling thesis is being promoted by the very institutions that stand to benefit from it. Citi is a prime broker. They want volatility. They want rotation. Skepticism isn't about dismissing the logic; it's about understanding the timing. The chip maker trade is already crowded. Nvidia's PE ratio is over 70. AMD's is 40. The moment the next earnings cycle shows any slowdown in data center revenue, the rotation will reverse faster than a flash crash.
Moreover, the chip bottleneck isn't permanent. By 2026, ASIC customization by Google (TPU), Amazon (Trainium), and Microsoft (Maia) will erode Nvidia's monopoly. The AI infrastructure narrative will fragment into sub-themes: advanced packaging, liquid cooling, energy. The next liquidity rotation will then target those niches. In crypto terms, it's like moving from ETH to L2s like Arbitrum, then to specific DeFi protocols on those L2s. Each layer has its window.
Liquidity doesn't give you a second chance. The window for the chip maker trade is closing fast. Based on my experience auditing 50+ ICO whitepapers in 2017, I learned that when everyone agrees on a trade, the easy money is gone. The question now is: what comes after the chip maker mania? My bet is on the industrial metaverse—where AI agents interact with physical supply chains via blockchain-based identity and settlement. That's where the next structural bottleneck lies.
The takeaway is uncomfortable but necessary. The Citi call is a mirror for crypto investors. We are in a bull market, but the euphoria is masking technical flaws. The 'Magnificent Seven' of crypto—Bitcoin, Ethereum, Solana—are being challenged by new infrastructure narratives: AI compute tokens, zero-knowledge proofs, and decentralized physical infrastructure networks (DePIN). The capital rotation we see in traditional markets will inevitably bleed into crypto. Position for the infrastructure layer, but hedge your bets. The liquidity map is shifting, and the next vacuum is forming where you least expect it.