DeFi moats in an AI world and the implications for value capture

Five years ago, building a DEX or borrow/lend protocol meant raising $3 million, hiring five engineers, and spending six months on an MVP. Now, a founder can spin up most of that in a weekend with Claude Code. Ask a frontier AI model to build a perps exchange, and you can get a matching engine, risk engine, trading interface, APIs, and most of what you need out of the box.

AI is making it easier to compete across almost every part of the software industry, but financial services will be among the hardest businesses to commoditize. That's because the most valuable parts of a financial business have very little to do with the software itself. Building a loyal customer base, accumulating a large balance sheet, and developing proprietary data and underwriting capabilities all take years of building relationships, establishing a track record, and earning trust. AI can make the software much cheaper to build, but it doesn't make any of those things particularly easier to replicate.

Said another way, a model can write a lending protocol in an afternoon. What it can’t do is manifest $10 billion of stablecoin deposits into the contracts or sign up 28 million customers for a new neobrokerage.

We think this is incredibly bullish for DeFi. Trillions in financial activity will move onchain over the next decade, and AI will make it much cheaper to build the products that serve that activity. But the things that make financial businesses valuable and defensible aren't getting any easier to replicate.

A New Pillar of our DeFi Value Capture Framework

We've spent years thinking about value capture in DeFi, and our views have evolved as the market has grown. Early on, we looked at how app tokens capture value using models such as work tokens and burn-and-mint equilibrium (BME). Later, we focused on DeFi defensibility and unforkable state. The idea was that anyone can fork open-source code, but they can’t copy liquidity or network effects. More recently, we argued that DeFi protocols capture the most value when they have some form of risk that needs to be managed. All of these theses have played out to some extent, and this essay is the next iteration of our thinking.

We’ll preface the rest of this essay with the fact that we believe DeFi is going to continue swallowing CeFi, and eventually all of traditional finance too. DEX market share against CEXs is at an all-time high and still climbing:

Sources: The Block, DefiLlama

We're also now seeing real adoption of tokenized assets, whether that's treasuries, private credit, or equities. Importantly, it's not just crypto-native users fueling the growth. Incumbents, fintechs, institutional asset managers, and traditional brokerages are building products onchain and bringing their customers along.

Source: rwa.xyz

The catch is that today's blue-chip DeFi protocols won't necessarily be the ones to capture this growth. As software and data analysis get cheaper, value flows to whoever controls the scarce inputs software can't create.

We see three primary moats in DeFi: 1) distribution, 2) capital, and 3) proprietary data. There's a fourth, risk tolerance, that works differently and is really a wedge, and we'll get to it at the end.

Moat #1 - Distribution

One of the seminal early crypto theses was “Fat Protocols,” written in 2016 by Joel Monegro. The idea was that, while the internet produced enormous value at the app layer (Google, Facebook, Amazon, etc.), blockchains would invert that relationship, with most of the value accruing to the underlying protocols.

The evidence points in the opposite direction, and this trend is accelerating. As protocols get cheaper to launch, value moves to whoever owns the customer relationship. That business can route users and balances to whichever venue pays the most. Sometimes that's an app, and sometimes, like Hyperliquid, it's a chain that's also the app.

We define distribution simply: it's about controlling where order flow or balances go. The caveat is that not all distribution is equally sticky. Funded accounts that are difficult to move create a real barrier, while an interface on its own is much more vulnerable to competition. We’ll get into this more below.

Exchanges and other stablecoin distributors

The clearest place distribution shows up is the stablecoin value chain. Issuance has historically been a fantastic business, assuming you hit scale and interest rates are not near zero. Tether and Circle hold dollar reserves, earn interest on them, and in most cases pay holders nothing. Tether alone has earned over $20 billion of nearly 100%-margin interest income since 2021 by doing this:

Source:DefiLlama

But take Hyperliquid, which has over $6 billion of USDC sitting on its platform as derivatives collateral. Those balances generate substantial reserve income for Circle and its distribution partners. At some point, Hyperliquid realized that it was the reason those balances existed, and therefore had leverage: if Circle wasn’t willing to share the economics, Hyperliquid could replace USDC with a stablecoin that would or theoretically create their own.

Earlier this year, Hyperliquid made a deal with Coinbase using its Aligned Quote Asset (AQA) framework. Under this deal, roughly 90% of the reserve yield from USDC on Hyperliquid flows to the protocol's Assistance Fund, which uses it to buy HYPE, rather than to Circle or Coinbase. With $5.5 billion of USDC in the treasury (the 90% that goes to HYPE), that's a ~$185 million revenue line for Hyperliquid today.

Hyperliquid doesn't maintain any of the plumbing here, whether it be mint/redeem, treasury management, or regulatory licenses. It owns the distribution, and that's enough to demand more of the economics.

Wallets and frontends

Wallets and frontends are the other side of this. In the early days, noncustodial wallets struggled a lot with monetization. They eventually realized that the king moneymaking business in crypto is trading, and so they made it a first-class product a few years ago. On trades, MetaMask charges a 0.875% fee, compared to Uniswap's average trading fee of 0.13% over the last 12 months. And most of that value goes to liquidity providers (LPs) rather than the Uniswap protocol itself.

In short, the wallet earns far more than the venue that executes the trade. On the other side, the exchange has to attract LPs, maintain complex infra, and compete on execution; MetaMask has to win and keep the customer relationship. If one DEX doesn't offer a good price, MetaMask can route the trade somewhere else. Over time, if they get enough order flow, they can internalize a lot of it and keep even more of the economics.

We're also starting to see this dynamic on the perps side. Phantom lets users trade perps on Hyperliquid and charges an additional 5 bps on top of Hyperliquid's 4.5 bps fee. Hyperliquid runs the risk engine, matching, and liquidity, but Phantom earns more just by owning the interface and the user.

There’s a limit here, though. Among Hyperliquid builders, a new red-ocean market right now, there’s already a pricing war going on and teams are lowering fees to improve retention. In other words, even the frontend layer isn’t immune to commoditization.

More generally, crypto wallets are somewhat vulnerable because switching costs are so low. Users can easily connect their private key or seed phrase into another wallet and access the same funds through a new interface in minutes. Switching banks or fintechs is much more involved. Users need to open a new account, complete KYC, move funds over traditional rails, reconnect payroll, and potentially update other connected services.

Fintechs and Neobanks

To illustrate the fintech example, take Robinhood, which has 28 million funded accounts. Right now, Uniswap is the main DEX on Robinhood Chain, and Robinhood is fine with that. But as AI drives down the cost of building financial software, nothing stops Robinhood from building and launching its own DEX protocol and sending users there. With their reach and state-of-the-art coding tools, they could spin up one of the biggest layer-2 DEXs in DeFi virtually overnight. We're not saying Robinhood will 100% do this, by the way, as owning the liquidity hasn’t historically been Robinhood’s primary goal (though they did start routing a lot of flow to Rothera, their prediction market, over Kalshi). We're simply pointing out that building the protocol is no longer the hard part. Attracting users, capital, and proprietary data is.

You can see a DeFi waterfall value chain forming. Frontends (Phantom) are capturing more of the economics from trading venues, while trading venues (Hyperliquid) are capturing more from stablecoin issuers (Circle). In all cases, the business with distribution has pricing power over the infrastructure underneath.

But how much each layer retains comes down to how easily its users can leave. Hyperliquid got roughly 90% of the reserve yield, while frontends on Hyperliquid are already competing fees down to low single digit bps.

Moat #2 - Capital

The natural conclusion from the previous section is that headless protocols and backend infrastructure are disadvantaged. And yes, in some cases, that’s true. But there are other moats beyond distribution, and one of the most important is capital.

Capital is the lifeblood of financial markets. Traders need it to take on leverage, market makers need it to keep quoting prices, and lending protocols need deposits to make loans. AI can help write code and automate workflows, but it can’t create money out of thin air, which is what these systems rely on. And when something breaks (e.g., Aave’s KelpDAO incident or MakerDAO’s Black Thursday), someone has to eat the loss.

Before we go further, let's talk about liquidity. There's a log(n) network effect: traders go where liquidity is, and liquidity follows them. Hyperliquid's open interest is a real advantage, but that capital belongs to users who can move it elsewhere. We've seen liquidity move from one protocol to another in DeFi, often driven by token incentives and vampire attacks. And when those incentives dry up, a lot of times the liquidity leaves too.

The two forms of capital below can actually create defensible long term value.

DeFi Prime Brokerage

Prime brokers (PBs) are critical market participants in tradfi markets. They provide leverage, finance trades, and enable hedge funds to use a single pool of capital across markets and strategies, rather than posting separate collateral at each venue.

It's an absolutely gigantic business. Industry-wide PB revenues grew from ~$15 billion in 2020 to ~$27 billion in 2024 on the back of growth in the hedge fund industry and multi-manager pod shops that rely a ton on financing.

We think the opportunity in DeFi could be even bigger. Traditional PBs mostly serve large institutions because dealing with smaller customers is operationally expensive. Smart contracts can automate much of that work and offer it to anyone, from someone trading $500 to a pod at Millennium running a few hundred million.

Today, traders generally have to post separate collateral on Hyperliquid, Aave, Lighter, Morpho, Variational, and Kamino. That's extremely capital-inefficient, and the problem only gets worse as tokenized stocks, treasuries, private credit, and other RWAs move onchain. A fund that's long tokenized equities on one venue and short equity index perps on another shouldn't have to fully collateralize both positions if some of the risk offsets.

Building the software to manage this is increasingly easy, whereas the hard part is having the balance sheet to finance those positions, a low cost of capital, and the risk-management know-how to avoid blow-ups. Over time, PBs with cheaper capital can offer better financing terms, attract more customers, and use that scale to secure even cheaper financing. In sum, there should be great returns to scale in this business.

Backstopping risk in DeFi protocols

The second way capital can serve as a moat is by backstopping risk. This ties directly to our previous essay around DeFi DAOs managing risk, where we argued that tokens affiliated with protocols that don't manage risk are much more susceptible to being forked out than those that do.

For example, look at Hyperliquid's HLP vault, which provides liquidity by market making and taking over liquidated positions. Depositors earn a share of the profits but also absorb losses. In March 2025, HLP lost roughly $4 million unwinding a large ETH position, and the subsequent JELLYJELLY incident exposed it to further risk. Both of these situations showed why derivatives DEX insurance funds are so important. Aave's Umbrella serves a similar purpose, allowing users to earn yield in exchange for backstopping bad debt.

Anyone can fork HLP or Umbrella contracts, but not the capital inside. The catch: this is third-party capital, and depositors could leave if they find better yields. The counter is that some capital is sticky because there may be overlap between token holders and depositors. HYPE holders, for example, might keep some USD in HLP to make the exchange safer and protect their token investment.

The most durable version of this strategy is a protocol's own balance sheet, built from retained earnings. That capital won’t walk out the door chasing newer, shinier yields or points programs.

Binance recently illustrated why this approach matters. After losing $6.3 billion in a hack on October 29, it covered the losses with its SAFU reserve and corporate funds (mostly the latter), reopened withdrawals within days, and saw only net 3% of assets leave. It’s still early, but the read so far is that its perceived balance sheet helped prevent a bank run.

As DeFi moves into private credit, tokenized equities, and other RWAs, demand for insurance and risk capital will keep growing. The net result is that protocols that actually provide and manage capital will capture way more value than those that just ship software.

Moat #3 - Proprietary data

The third moat is information, with a caveat. DeFi was built on the idea that financial data should be public, which works well for things like crypto-backed lending and spot trading. But as more RWAs move onchain, we're going to see more businesses built around info that isn't public. The moat is the combination of proprietary data and the ability to underwrite risk with it.

Credit curators

Lending is the obvious place this shows up first. Curators on platforms like Morpho, Kamino, and Euler assess risk on behalf of depositor capital to the tune of $9.5 billion today:

Source: DefiLlama

These curators are granted limited controls in the vault smart contract to adjust which lending markets are allocated to, set risk parameters, and monitor collateral positions. Despite managing $9.5 billion between them, they generate very little revenue, especially compared to traditional asset managers with greater discretion and responsibility. In fact, many charge nothing at all. Steakhouse Financial, the largest DeFi risk curator by AUM, has a take rate of 22 bps on its managed assets, and the average across curators is 19 bps.

Part of this is that risk-adjusted yields in DeFi are unattractive right now (likely due in part to latent capital sitting in Aave). Part of it is that Morpho and Kamino have commoditized the vault layer and sit between the curator and depositor, which is our distribution argument again.

But the bigger issue is that curators are working from mostly the same data. If you’re lending against ETH or BTC, the prices, liquidation rules, and the collateral are all public. AI makes it easy to replicate risk modeling, leaving curators to compete on judgment (e.g., risk-taking, much of which is algorithmically predefined) and track record. This inevitably puts pressure on fees over a sufficiently long time horizon.

Compare that to a protocol like OnRe. It's an onchain reinsurance company that takes capital from investors and uses it to underwrite real-world insurance risk. To do that, it needs historical claims, loss probabilities, and the insurance contracts themselves. Not to mention, it needs people who understand that data. None of this is on a block explorer, and a frontier model without access to a reinsurance book can’t underwrite it, no matter how smart it is.

That's why OnRe can charge a 75 bps management fee plus 20% of the yield. At current returns, that works out to roughly 3% annually, 16x more than the average DeFi risk curator.

These are different businesses with different risks, and some of that fee premium reflects illiquidity and the potential for losses. But that’s part of the point, and businesses with proprietary data, underwriting expertise, and relationships can charge meaningfully more. If you have the same data and tools as everyone else, you'll likely end up being a commodity.

What does all this mean for value capture?

The obvious question arising from this framework is who is best positioned to capture value if DeFi grows significantly over the long term. Some groups will benefit, while others will have a harder time monetizing.

L1s and L2s

A cynical interpretation here is that infrastructure tokens like L1s and L2s are in a tough spot. The reason is that blockchains used to give new DeFi apps two of these moats: distribution and capital.

Historically, blockchains had distribution because DeFi users kept their assets on the chain where they were most active. If you built on Ethereum, you had access to Ethereum users. If you built on Solana, you had access to Solana users. If you were starting Uniswap, Compound, Aave, Kamino, or MakerDAO from scratch, all of this mattered a lot. You had no existing customers, no assets, and no financial relationships, and therefore launching into an ecosystem where users already were made it much easier to get started.

But now, interoperability improvements are chipping away at that. Today, a fomo user onboards via their app and can trade any asset on any chain without even thinking about it. And more and more, new users will come into DeFi via fintechs like Revolut, Nubank, and Robinhood, which already have tens of millions of customers and don’t require funded wallets to access them.

L1s also supplied capital through the ecosystem wealth effect. As SOL and ETH prices appreciated, users had more money to deploy into apps on those chains. Downstream of that, capital stuck around because bridging was painful and users tended to be tribal. However, bridging is easier now, and capital comes in via stablecoins/tokenized assets/institutions, so new protocols are less dependent on the wealth effect from any one token. RWA issuers also want to launch on every chain, and we’ve already seen that from Securitize, Superstate, Ondo, and others.

On the data side, there was never much of a moat. Virtually all onchain financial information is public, so chains and apps never had much proprietary data to defend.

AI is also eating another historical advantage of blockchains: their developer ecosystems. One of Ethereum’s biggest advantages over newer chains like Solana or Aptos was its mature tooling, audited libraries, and integrations. But as AI makes it easier to port applications between VMs, that advantage matters less.

More broadly, general-purpose chains are losing their advantages in distribution and capital and never had much of a data moat to begin with. That leaves them competing primarily on execution and settlement, where blockspace is abundant and fees continue to fall.

That doesn't mean chains can't be valuable, though. As Tushar laid out in his summit presentation a few years back, L1 tokens are capital assets, commodity assets, and store-of-value assets. If Solana eventually processes a billion payments a day at a penny per transaction, that's $3.65 billion in annual fees. And as the chart below shows, transaction volumes keep growing:

Source: Blockworks Research

The takeaway is that the reasons they're valuable may look very different from what investors originally expected.

Apps and protocols

For apps, the businesses that capture the most value will be those with at least one of three moats: distribution through a direct customer relationship that’s hard to move (e.g., Robinhood or Hyperliquid); capital through a balance sheet that can absorb losses; or information through the ability to underwrite risks others can't. All three are much harder to replicate than the software, no matter how good the models get.

We don't think it matters whether the value goes to equity or tokens, and our view is that value will flow to both. Recent SEC guidance and rule proposals have mapped compliant ways for tokens to capture value, and we fully intend to invest in these opportunities. What really matters is whether a business (token or equity) controls something scarce, has pricing power because of it, and can turn that into returns for investors, whatever the instrument.

As more financial activity moves onchain, the gap between protocols with a moat and protocols with only software will widen. As one of the largest investors in the space, we spend a lot of time figuring out which projects in the stack have these moats and can keep them.

Doesn’t this all just favor incumbents?

A reasonable conclusion from everything above is that incumbents win. Robinhood has the customers, Goldman Sachs has more capital than any DeFi PB likely ever will, and Moody's has the data and the licenses. If software is becoming free and the moats are the expensive part, the firms that already own those moats should have the advantage.

But startups still have one edge: they’re willing to take risks incumbents won’t. Part of this is that AI cuts both ways here. The same tools that let Robinhood launch a DEX in a weekend also let a three-person startup do it. As the cost of building a credible product falls, the real opportunity for startups is going after markets and customers that incumbents aren't willing to pursue.

There are actually a few compelling case studies here already.

Ethena built a synthetic dollar product around delta-neutral hedging, a strategy traditional firms obviously understand (basis trading) but were unable or unwilling to bring to crypto exchanges. They’ve now scaled it to $5.4 billion of AUM.

Pump.fun and fomo came out of nowhere and created net new trading experiences in a novel asset class (memecoins) and built a ton of distribution around that wedge. They’re now two of the biggest success stories of this crypto cycle.

Hyperliquid entered a derivatives market dominated by Binance, Bybit, and OKX and rapidly took market share. It's also listing perps on pre-IPO companies like OpenAI and Anthropic, something traditional exchanges won't do. These markets attract new traders and liquidity, as you can see in the wallets created on the day of the SpaceX IPO:

Source: HyperTracker

On capital, Project 0 finances DeFi collateral that traditional PBs won’t lend against. And in terms of proprietary data, USD.ai can underwrite loans backed by GPUs, an asset class with which banks have almost no experience.

The common thread is that these businesses went after markets incumbents weren't serving, whether due to regulation, operational complexity, reputational risk, or simply because the opportunity seemed too small. There will always be openings like these for startups to compete asymmetrically, as we've seen throughout crypto's history and expect to keep seeing.

That's what makes the combination of AI and DeFi so compelling. AI is significantly lowering the cost of building financial products, while DeFi enables startups to reach customers and access capital without rebuilding the entire financial stack.

Of course, risk tolerance isn't a durable moat on its own. Regulation catches up, and incumbents follow once a market is proven (see BTC spot markets). But startups don't need a moat on day one. They just need an opening that incumbents won't pursue and enough time to turn it into one (or several) of the three moats, and then they can start compounding from there.

Summary

We believe trillions of dollars of financial activity will move onchain over the next decade, and AI will make it easier than ever to build the products that serve it.

But while software is becoming commoditized, the things that make financial businesses valuable aren't. Customers, capital, and proprietary data remain difficult to acquire, and the businesses that control them will have pricing power.

This is why we're so bullish on the next generation of DeFi companies. With the proliferation of AI developer tooling, startups can now compete with incumbents faster and more cheaply than ever. Once a new company builds one of these moats, it can create an incredibly durable business.

Ultimately, you can fork the code and generate it with the latest and greatest AI models. But you can't simply generate any of these moats, and that’s where the opportunities are for us as crypto investors.