When a Wells Fargo strategist declared publicly that banks were becoming the 'AI periphery,' the statement barely registered in crypto circles. Yet for anyone who has watched the flow of capital through the 2024-2026 cycle, this was a seismic signal. The strategist was referring to a simple but powerful observation: investors were rotating out of chipmakers like Nvidia and into the companies financing AI data centers—Goldman Sachs, JPMorgan, Morgan Stanley. The logic was clean: single AI data centers cost between $1 billion and $3 billion, and most of that capital must be raised through syndicated loans, bond issuances, and project finance. Banks, as the intermediaries, collect fees and interest. But for a crypto native, the story is not about buying bank stocks. It is about understanding how infrastructure finance works, and where decentralized alternatives are missing the mark.
I have been tracking this narrative since early 2024, when I began consulting with three Nordic banks on their blockchain strategies. During those workshops, I saw firsthand how traditional financial institutions approach large-scale infrastructure lending. They have dedicated teams for project finance, risk assessment based on decades of data, and relationships with the largest tech firms. Crypto, in contrast, has lending protocols that barely scratch the surface of real-world asset financing. The AI data center boom is a stress test for both systems. And the results so far are revealing: centralized banks are winning the funding game, but their dominance comes with blind spots that crypto is uniquely positioned to exploit.
Let me ground this in numbers. According to Synergy Research, global AI-related capital expenditure exceeded $200 billion in 2024 and is expected to surpass $300 billion by 2026. Approximately 60-70% of that is financed externally, meaning banks and private credit funds. The largest data center projects require structured finance solutions that combine senior debt, mezzanine tranches, and equity. This is the bread and butter of investment banks. During my time analyzing the EU's MiCA regulation for my educational platform 'Crypto Compass,' I interviewed 40 policymakers and developers. One senior banker from a top-tier European institution told me off the record: 'We are seeing a new asset class emerge—AI infrastructure debt. It is like lending to a utility, but with higher margins because the technology risk is still mispriced.' That mispricing is exactly where crypto could step in, but only if it solves its scalability and trust problems first.
The core insight from the bank-as-AI-periphery thesis is that infrastructure financing is a lagging but sticky revenue stream. While chipmakers enjoy volatile multiples based on product cycles, banks secure recurring fees from loan servicing and bond coupons. The profitability is lower, but the risk-adjusted returns are often better. For crypto, this suggests that any decentralized protocol aiming to compete must offer not just lower costs, but also institutional-grade risk assessment and capital efficiency. Uniswap V2, which I personally audited during DeFi Summer 2020, revealed that gas fee fluctuations disproportionately hurt low-income users. That same inefficiency plagues on-chain lending today. If we cannot yet finance a $100,000 loan for a small business without paying $50 in gas fees, how can we expect to finance a $1 billion data center?
But the contrarian angle is more nuanced. The bank thesis assumes that AI capital expenditures will continue to grow for at least 2-3 years. That assumption is fragile. During my research on the 2022 bear market, I analyzed 120 retail investors who lost savings to rug pulls. The common thread was not a lack of technical literacy, but a failure of emotional resilience during market downturns. The same applies to institutional capital: if an AI recession hits—due to a technology bottleneck like energy availability or model improvement plateau—banks could face a wave of non-performing loans. Crypto's opportunity is not to compete head-on with banks, but to build the financial infrastructure for a parallel economy where AI agents and decentralized organizations can raise capital without relying on centralized gatekeepers. That requires solving identity, credit scoring, and dispute resolution on-chain, which remain unsolved for large-scale lending.
Let me go deeper into the competitive landscape. Private credit funds like Blackstone and Apollo have been aggressively entering the data center financing space. They offer faster execution and more flexible terms than traditional banks, but at higher interest rates. During my consultations with the Nordic banks, I noticed a pattern: they were losing mid-sized deals to private credit because their risk committees were too slow. However, for the largest projects—those above $1 billion—the banks still dominate due to their ability to syndicate risk across multiple institutions. Crypto lending protocols, by contrast, are limited by their reliance on overcollateralization. MakerDAO can issue DAI against real-world assets, but the process is cumbersome and the scale is tiny compared to a $500 million syndicated loan.
This is where the 'code is law, but empathy is truth' signature applies. The banking system works because of relationships, trust, and legal frameworks. Crypto has the code, but lacks the empathy—the ability to understand borrower needs beyond collateral ratios. During the 2024 bear market, I proposed a hybrid model to one of the banks: use a public chain for transparency and settlement, but keep the credit assessment and legal enforcement off-chain. They were intrigued, but ultimately rejected it because of regulatory uncertainty. The MiCA regulation I helped analyze is a step forward, but it treats crypto as a separate asset class, not as an integral part of infrastructure finance. Until that changes, banks will remain the default financiers of AI.
What about the on-chain data? Let's examine the TVL in the top lending protocols. Aave has roughly $10 billion in total value locked, Compound around $5 billion, and MakerDAO about $8 billion in DAI outstanding. Compare that to the $200 billion annual AI capital expenditure. Even if all crypto lending protocols combined funded 10% of that, they would need to scale by 20x. Worse, most of the TVL is in overcollateralized loans for trading purposes, not for real-world infrastructure. The RWA tokenization narrative, which I have followed since 2021, has been a three-year storytelling exercise. I have yet to see a single $100 million data center loan originate on-chain. Traditional institutions do not need your public chain, as I have argued repeatedly. They have their own settlement systems, and they trust them more than a decentralized validator set.
Yet there is a hidden opportunity. The AI data center boom will generate enormous amounts of energy consumption, carbon credits, and computing capacity that could be tokenized. During my work on the 'Crypto Compass' video series, I interviewed a data center operator who told me that their biggest headache is idle compute during off-peak hours. A decentralized market for compute time, similar to what Render Network does for GPU rendering, could solve that. But to finance such a market, you need loans against future compute revenue—a type of cash flow lending that is very difficult to do on-chain today. The banking system can do it because it relies on audited financial statements and legal contracts. Crypto needs to evolve to support such instruments, which requires a shift from overcollateralization to undercollateralized lending based on reputation and data analytics.
From a technical perspective, Layer2 scalability plays a role here. Post-Dencun, the blob data on Ethereum will be saturated within two years, and then all rollup gas fees will double again. That is my core prediction, based on my analysis of blob size and demand growth. If on-chain lending becomes even more expensive, it will push large-scale financing further away from crypto. The banks will not care about blob saturation; they settle on private systems with no gas costs. So what is the path forward? We need Layer2 solutions that prioritize stable execution costs for large transactions, not just cheap token swaps. This is a market failure that no one is addressing because the industry is obsessed with memecoins and speculation.
The contrarian angle I want to emphasize is that the bank thesis itself has a blind spot: it ignores the role of non-bank financing and the potential for a government-led infrastructure bank. The CHIPS Act in the US is already providing subsidies for semiconductor manufacturing, and similar policies could extend to data centers. If governments start financing AI infrastructure directly, the banks' role could diminish. Moreover, the AI companies themselves—Microsoft, Google, Amazon—are increasingly using their own cash reserves to fund data centers, bypassing banks entirely. In 2024, Microsoft announced $50 billion in self-funded data center expansion. That internal financing reduces the addressable market for banks. Crypto could learn from this: the real growth is not in lending to AI, but in building the tools for AI agents to manage their own treasuries, pay for compute, and settle disputes autonomously. That is the 'sovereign intelligence' era I have been writing about in my manifesto 'The Cognitive Commons.'
Surviving the winter to plant the spring. That is my motto for why this analysis matters. The bear market of 2022 taught me that resilience is not just a financial metric, but a narrative. Banks are resilient because they are diversified. Crypto lending protocols are fragile because they are overexposed to volatile collateral. If we want to finance the next generation of AI infrastructure, we need to build protocols that can handle the complexity of real-world cash flows, legal enforceability, and long-term capital commitment. That is not a simple technical upgrade; it is a philosophical pivot from 'trust no one' to 'verify everyone, feel everyone.' The ledger remembers, but the heart forgives. And the heart of this opportunity is to create a financial system that works for both machines and humans, without sacrificing the values of decentralization.
Let me conclude with a forward-looking thought. Over the next 12 months, watch for three signals. First, the AI capital expenditure guidance from the hyperscalers—Microsoft, Google, Meta, AWS. If they raise it, bank stocks will likely rally, and crypto lending volumes might stagnate as capital flows to traditional channels. Second, monitor the private credit market for data center loans. If Blackstone starts syndicating tranches on-chain via a tokenized debt product, that could be a bridge. Third, look at the MiCA implementation in Europe. If it allows regulated stablecoins to be used for large-scale lending, we might see the first $100 million RWA loan on-chain. Until then, the banks remain the AI periphery, and crypto remains the edge case. But edge cases have a way of becoming the center in the next cycle.
Behind every hash, a heartbeat. The hash today is the bank's settlement system, but the heartbeat is the community's desire for a fairer financial system. The tension between these two forces will define how AI infrastructure gets funded in the coming decade. I am not betting against the banks, but I am building for what comes after them.