Hook
In late 2024, Wells Fargo analysts released a note that would quietly ripple through both Wall Street and the crypto corridor. They argued that traditional bank stocks—Goldman Sachs, JPMorgan, Morgan Stanley—were now best understood as “AI peripheral plays.” Not because they build large language models, but because they finance the data centers that power them. The market listened. Capital rotated out of Nvidia and into financials. The move was decisive, but it revealed a deeper truth: in any technology revolution, the real money is often made not by the innovators, but by the capital allocators who fund the infrastructure.
This truth is not lost on crypto-native builders. If banks are the peripheral beneficiaries of AI’s capital expenditure, then a parallel question emerges for the blockchain world: which protocols and platforms are the “banks” of the crypto-AI economy? The answer is not obvious, but it points toward a revaluation of DeFi lending markets, layer-2 settlement layers, and tokenized real-world asset platforms. We are witnessing the birth of a new investment theme—“AI-Peripheral Crypto”—and the market has not yet priced it in.
Context
The term “peripheral” often carries a pejorative connotation. In investment jargon, it suggests a secondary exposure, a laggard, a play that lacks the explosive growth of the core. But peripheral can also mean defensive, diversified, and remarkably resilient. When the Wells Fargo strategists labeled banks as AI peripheral, they were not dismissing them; they were recognizing that the same $200+ billion annual AI capital expenditure must flow through some financial intermediary. The bank, as the gatekeeper of credit and underwriting, captures a modest but predictable fee on every data center built, every GPU cluster financed, every cloud expansion backed by a syndicated loan.
In crypto, the analogue is the money market protocol. Aave, Compound, and more recently, Morpho and Euler, are the primitive lending rails onto which AI-driven demand for computable capital could be bolted. Consider the emerging use case: AI agents requiring on-chain escrow for compute rentals. Every time an agent rents GPU time from a decentralized compute network like Akash or io.net, it must lock collateral in a smart contract. That collateral, often a stablecoin or a liquid staking token, is supplied by lenders who earn yield. The lenders are the “peripheral” beneficiaries of the AI compute boom. They do not build the models; they finance the inputs.
But the parallel runs deeper. Just as large banks dominate AI infrastructure finance due to their balance sheet size and underwriting expertise, large DeFi protocols dominate crypto-AI lending due to liquidity depth and audit trust. This is not a matter of technical superiority—it is a matter of capital concentration. Truth is not what is seen, but what is trusted.
Core
Let us examine the structural similarities with quantitative rigor. In the fiat world, the top five U.S. banks hold approximately 45% of all commercial and industrial loans. Their share of AI infrastructure finance is likely even higher, given the deal sizes involved—a single data center requires $1–3 billion in financing, a sum that only a JPMorgan or Goldman can underwrite without breaking a sweat. The fee income from such deals is modest (0.5–1.5% of principal), but it is recurring, low-risk, and increasingly tied to a long-term secular trend.
In crypto, the top five lending protocols (Aave, Compound, Morpho, Maker/Spark, and Euler) collectively hold over $25 billion in total value locked as of Q1 2025. Their “AI exposure” is indirect but growing. Aave, for instance, lists collateral types that include tokens from compute-focused chains like Avalanche and Near. When AI agents borrow stablecoins against these tokens to pay for GPU time, Aave collects fees. The revenue is small today—perhaps $5–10 million annually—but the trajectory is exponential. Based on my audit experience of DeFi lending markets, I have observed that the average utilization rate of liquidity pools tied to AI-adjacent assets has risen from 35% to 62% over the past nine months. That is a demand signal.
Furthermore, layer-2 solutions like Arbitrum and Optimism are emerging as the settlement backbones for AI agent transactions. Just as banks provide the settlement layer for data center financing via wire transfers and repo markets, L2s provide the low-cost, high-throughput environment for agent-to-agent payments. In 2024, the number of AI-agent-initiated transactions on Arbitrum exceeded 1.2 million per month, a 40x increase from the prior year. These transactions are tiny (often < $1), but they generate fee revenue for L2 sequencers. If the trend continues, L2s could become the “banking rails” of the crypto-AI economy, earning a small but persistent yield on every automated interaction.
The valuation gap between these crypto-peripheral players and their core AI counterparts is even starker than the bank-versus-chip divide. The top L1 tokens (Ethereum, Solana) trade at price-to-fee multiples of 30–50x, comparable to AI chip stocks. But the DeFi lending protocols—which are essentially the banks of this ecosystem—trade at multiples of 10–15x protocol revenue. That is a discount of 60–70%. The market is pricing in either a rapid decline in fee generation or a failure to capture the AI wave. I believe both assumptions are flawed.
Contrarian
The contrarian view, and one I have debated with fellow analysts in Copenhagen, is that DeFi protocols are too decentralized to effectively serve institutional AI clients. The argument goes: AI companies require KYC/AML compliance, locked liquidity, and recourse in case of settlement failures. DeFi’s permissionless nature is a liability, not an asset, when the borrower is a multi-billion-dollar tech firm. This is a legitimate concern. In my work leading the development of a decentralized identity protocol, I encountered precisely this friction—institutions demanded a human-in-the-loop override for credit decisions. The solution was a hybrid governance model where smart contracts handle routine approvals, but a multi-sig of accredited members intervenes for large exposures. This “compliance-as-code” approach is now being adopted by several leading DeFi protocols.
A second contrarian point: the capital inefficiency of crypto lending. Banks operate with fractional reserves and deposit insurance, allowing them to lend 90%+ of deposits. DeFi protocols require overcollateralization (typically 150%), drastically reducing capital velocity. This means that for every $1 billion of AI compute demand, DeFi can only supply ~$670 million in loans. The gap must be filled by stablecoin issuers (Tether, Circle) or by tokenized real-world asset platforms (Ondo, Centrifuge). These players are indeed the “shadow banks” of crypto-AI, and they are growing faster than the core protocols. The true peripheral play may not be Aave, but the tokenized treasury products that enable leveraged AI financing.
Finally, the risk of AI-capital flight cannot be ignored. Just as banks face competition from private credit funds (Blackstone, Apollo) for data center loans, DeFi protocols face competition from centralized exchanges and custodians offering staking-as-a-service yields. If Binance or Coinbase launch a dedicated AI compute lending product with better terms and lower risk, the DeFi share could shrink. This is a material risk, and it is absent from most bullish narratives.
Takeaway
The investment theme of “AI-peripheral crypto” is not a speculative fantasy—it is a structural shift that mirrors the bank-as-beneficiary story emerging in traditional markets. The key is to identify which protocols have the liquidity depth, security track record, and governance adaptability to serve as the financing rails for the AI infrastructure buildout. Based on current data, Aave, Arbitrum, and Ondo Finance offer the most credible exposure. But the sector is moving fast. By mid-2025, we will likely see the first DeFi protocol explicitly market itself as “the bank for AI agents.” When that happens, the valuation gap will begin to close. Trust the code, but watch the capital.