Building a Lending Protocol Takes Just an Afternoon: In the AI Era, Which Three Moats Does DeFi Rely On to Truly Profit?
Author: Spencer Applebaum
Editor: TechFlow
TechFlow Summary: In the past, building a DEX or lending protocol often required millions in funding and a six-month development cycle. Today, with cutting-edge AI tools like Claude Code, founders can write matching and risk control engines over a single weekend. The software barriers created by open-source code are accelerating toward disappearance.
In this latest essay published on Multicoin, author Spencer Applebaum breaks down the path for value reconstruction in DeFi as software approaches zero cost: when technical implementation is no longer scarce, true pricing power belongs to those who control the three key moats of user distribution, proprietary risk capital, and non-public proprietary data.
Five years ago, building a DEX or lending protocol meant raising $3 million, hiring five engineers, and spending six months to launch an MVP. Today, founders can build most of the framework over a single weekend using Claude Code.
Ask a cutting-edge AI model to build a perpetual contract exchange, and it can nearly out-of-the-box generate matching engines, risk control engines, trading interfaces, APIs, and the vast majority of core components.
AI is lowering competitive barriers across virtually every software sub-sector, but financial services will remain one of the hardest businesses to commoditize. This is because the most valuable parts of a financial institution have little to do with the software itself.
Building a loyal customer base, accumulating a massive balance sheet, and refining proprietary data and underwriting capabilities require years of relationship building, establishing track records, and earning trust. While AI can drastically reduce software development costs, it cannot make these capabilities easier to replicate.
In other words, a large language model can write the code for a lending protocol in an afternoon. But it cannot conjure $10 billion in stablecoin deposits flowing into contracts, nor can it attract 28 million customers for a new internet brokerage overnight.
We see this as a major tailwind for DeFi. Over the next decade, trillions of dollars in financial activity will migrate on-chain. AI will significantly reduce the cost of building products that serve these activities, but the core elements that support the high value and high barriers of financial businesses have not become any easier.
New Pillars of the DeFi Value Capture Framework
We have been contemplating DeFi value capture for years, and our perspective has evolved alongside market maturity. Early on, we researched how application tokens could capture value through work tokens and the Burn-Mint Equilibrium (BME) model.
Later, we focused on DeFi defensibility and unforkable states, arguing that while anyone can fork open-source code, they cannot replicate liquidity or network effects.
Moving forward, we argued that DeFi protocols capture maximum value only when they must manage some form of material risk. These arguments have been validated to varying degrees; this article represents the next iteration of our thinking.
Let us first state the overarching premise of this piece: We believe DeFi will continue to cannibalize CeFi, ultimately absorbing the entire traditional financial system. Currently, the spot trading market share of DEXs relative to CEXes is at an all-time high and continues to climb:
Chart: Trend of DEX spot trading volume as a percentage of CEX volume, recently rising to 24% (Source: The Block, DefiLlama)
Meanwhile, whether it’s U.S. Treasuries, private credit, or equities, we are witnessing genuine adoption of tokenized assets. More importantly, the drivers of this growth are no longer limited to crypto-native users.
Traditional industry giants, fintechs, institutional asset managers, and legacy brokers are all building products on-chain and bringing their existing customer bases along into the on-chain world.
Chart: Scale structure and growth trend of on-chain tokenized real-world assets (RWAs) (Source: rwa.xyz)
The critical question, however, is whether today’s blue-chip DeFi protocols will naturally absorb this growth premium. As software and data analytics costs decline, value will flow to those who control “scarce elements that software cannot create.”
We identify three core moats in DeFi: distribution channels, capital, and proprietary data. There is also a fourth element—risk tolerance—that functions differently, acting more like a wedge to enter the market, which we will discuss further at the end.
Moat #1: Distribution Channels
One of the most influential theories in early crypto comes from Joel Monegro’s 2016 essay, Fat Protocols. The thesis posited that the internet left enormous value at the application layer (Google, Facebook, Amazon, etc.), whereas blockchain would invert this relationship, anchoring most value in the underlying protocols.
Real-world evidence points in the opposite direction, and this trend is accelerating. As protocol deployment costs fall, value flows to whoever controls customer relationships. These companies can route users and account balances to any underlying venue offering the highest rebates. Sometimes it’s a standalone app; other times, it’s both a chain and an application itself, like Hyperliquid.
Our definition of distribution is straightforward: the ability to control where order flow or funds go. But not all distribution enjoys equal customer stickiness. Hard-to-migrate, KYC-verified funded accounts build genuine barriers, whereas plain front-end interfaces face fiercer competition, as we will analyze further below.
Exchanges and Other Stablecoin Distributors
The clearest demonstration of pricing power in distribution channels lies in the stablecoin value chain. As long as scale is achieved and interest rates aren’t near zero, stablecoin issuance has always been a lucrative business. Tether and Circle hold dollar reserves and earn interest, rarely sharing rewards with holders.
Since 2021, Tether alone has generated over $20 billion in interest income from this model, operating with margins nearing 100%:
Chart: Cumulative revenue trend for Tether from May 2021 to October 2026, totaling $20.2 billion (Source: DefiLlama)
By contrast, Hyperliquid holds over $6 billion in USDC on its platform as derivative margin. These retained balances generate substantial reserve yield income for Circle and its distribution partners.
At a certain point, Hyperliquid realized it was the fundamental reason these balances existed, giving it negotiating leverage: if Circle refuses to share yields, Hyperliquid can easily swap USDC for stablecoins willing to split revenue, or theoretically issue its own.
Earlier this year, leveraging the Aligned Quote Assets (AQA) framework, Hyperliquid struck a partnership deal with Coinbase.
Under the agreement, roughly 90% of the USDC reserve yields on Hyperliquid no longer go to Circle or Coinbase, but instead flow to the protocol’s Assistance Fund to buy back HYPE.
Based on the $5.5 billion USDC holding representing this 90% allocation in the vault, this arrangement now generates approximately $185 million annually for Hyperliquid.
Hyperliquid doesn’t need to maintain mint/redemption rails, treasury management, or compliance licenses itself. By firmly controlling the distribution gateway, that alone gives it leverage to demand a larger slice of the economic pie.
Wallets and Frontend Gateways
The flip side emerges in wallets and frontend gateways. Non-custodial wallets initially struggled with monetization before realizing trading is actually the most profitable business in crypto, prompting them to elevate swaps to a core product a few years ago.
Today in swap transactions, MetaMask charges fees up to 0.875%, while Uniswap’s average swap fee over the past 12 months sits at just 0.13%, much of which goes to liquidity providers (LPs) rather than the Uniswap protocol itself.
In short, wallets earn far more than venues that merely match and execute trades. Underlying exchanges must constantly attract LPs, maintain complex infrastructure, and compete on execution slippage; MetaMask simply needs to win and retain customer relationships. If a DEX offers poor pricing, MetaMask can instantly reroute orders elsewhere.
In the long run, once wallets and frontends accumulate sufficiently massive order flow, they can even internally match a large portion of trades, capturing a bigger share of the economic upside.
A similar dynamic is emerging in the perpetual contract space. Phantom allows users to trade perpetuals on Hyperliquid, adding a 5-basis-point fee on top of Hyperliquid’s native 4.5 basis points.
While Hyperliquid bears the heavy lifting of risk engines, matching, and liquidity, Phantom extracts a larger cut simply by controlling the interface and user base.
This model, however, also hits a ceiling. In the Hyperliquid Builders ecosystem, which has turned into a fierce red ocean, price wars have erupted, with teams slashing fees to secure user retention. In other words, the frontend layer is not immune to commoditization either.
More broadly, crypto wallets are relatively fragile due to exceptionally low user migration costs. By simply importing a private key or mnemonic phrase into another wallet, users can seamlessly take over their identical asset portfolio within minutes.
Changing banks or fintech apps, by comparison, is vastly more cumbersome: users must reopen accounts, complete KYC, transfer funds through traditional clearing rails, re-link payroll accounts, and update all associated services.
Fintech Companies and Digital Banks
Take Robinhood, the fintech giant with 28 million funded accounts. Currently, Uniswap remains the primary DEX on the Robinhood Chain, a status Robinhood willingly permits.
Yet as AI continuously depresses financial software development costs, nothing prevents Robinhood from building its own proprietary DEX protocol and diverting traffic there. Leveraging massive reach and top-tier AI coding tools, they could spin up one of DeFi’s largest Layer 2 DEXs overnight.
We aren’t saying Robinhood will definitely do this; historically, acting as the house and holding inventory hasn’t been core to their model (though they have indeed started routing significant prediction order flow to their proprietary markets like Rothera rather than Kalshi).
We are simply pointing out: writing protocol code is no longer the hard part. Acquiring users, capital, and proprietary data is.
A clear DeFi “waterfall value chain” is taking shape. Frontends (like Phantom) are reclaiming larger cuts from trading venues, while venues (like Hyperliquid) are recapturing yields from stablecoin issuers (like Circle). Across all layers, whoever controls distribution wields pricing power over the infrastructure beneath it.
Ultimately, how much profit each layer retains depends entirely on how difficult it is for users to leave. Hyperliquid captures roughly 90% of the reserve yields, while frontend apps running on it are already being forced to slash add-on fees down to mere basis points.
Moat #2: Capital
The natural deduction from the previous section seems to be that backend-heavy, headless protocols and infrastructure are at a distinct disadvantage. While true in some scenarios, there are other moats beyond distribution, with capital being by far the most critical.
Capital is the lifeblood of financial markets. Traders need it to leverage positions, market makers need it to quote continuously, and lending protocols need deposits to originate loans. AI can write code and automate workflows, but it cannot conjure capital from thin air. When systems fail—as seen in events like Aave’s KelpDAO exploit or MakerDAO’s Black Thursday liquidation—someone must step in to cover the losses.
Beyond this baseline, we must address liquidity. Liquidity undeniably exhibits logarithmic (log(n)) network effects: traders follow liquidity, and liquidity follows traders. While Hyperliquid’s open interest is a massive advantage, those funds belong to users who can withdraw at any time.
We’ve repeatedly witnessed liquidity migrating across protocols driven by token incentives and vampire attacks. Once incentives dry up, massive liquidity flees. Of the two capital forms discussed below, they constitute the true moats capable of building long-term defensive barriers.
DeFi Prime Brokerage
Prime brokers (PBs) are key players in traditional markets. They provide leverage and trade financing, allowing hedge funds to utilize a shared pool of capital across different markets and strategies without locking up separate margins at each venue.
This is a massively scaled business. Fueled by the expansion of hedge funds and highly financed multi-strategy shops (Pod Shops), global PB revenues grew from ~$15 billion in 2020 to ~$27 billion in 2024.
We believe this opportunity will be even larger in DeFi. Traditional prime brokers mainly service large institutions because servicing retail is operationally expensive; smart contracts can automate most processes, enabling everyone from retail traders posting $500 to Millennium’s research desks managing hundreds of millions to access unified margin services.
Currently, traders typically must post independent margins across Hyperliquid, Aave, Lighter, Morpho, Variational, and Kamino. This is incredibly capital inefficient, and the problem will only worsen as tokenized equities, Treasuries, and private credit RWAs come on-chain.
If a fund longs tokenized equities on one venue while shorting index perpetuals on another, netting out partially offsetting risks, it shouldn’t be forced to post full collateral on both sides.
Writing software to manage this logic is increasingly easy. The real difficulty lies in holding a balance sheet capable of cross-margining, accessing low-cost capital, and applying risk control experience that avoids cascading liquidations. Prime brokers with cheaper funds can offer better financing terms, attract more clients, leverage scale to secure even cheaper capital, creating powerful network economies.
Backstopping Protocol Risk
A second way capital builds moats is by providing backstops for protocol risk. This aligns with our earlier discussion on DAO risk management: we previously argued that protocol tokens that don’t actively manage risk are more easily forked by competitors than those that do.
Take Hyperliquid’s HLP Vault, which provides liquidity via market making and taking over liquidated positions. Depositors share profits but bear loss risks. In March 2025, HLP lost ~$4 million while unwinding large ETH positions, and the subsequent JELLYJELLY exploit further stressed its exposures.
Both incidents underscore the importance of insurance funds for derivatives DEXs. Aave’s Umbrella mechanism plays a similar role, allowing users to assume bad debt risk in exchange for extra yields.
Anyone can fork the smart contract code for HLP or Umbrella, but they cannot fork the actual cash locked inside. There is a catch, however: since these are third-party funds, depositors can flee at any moment if higher yields emerge elsewhere.
Conversely, some funds exhibit stickiness because token holders and vault depositors heavily overlap. For instance, HYPE holders may deliberately retain USD assets in the HLP to bolster exchange solvency and protect their token exposures.
The most robust endpoint of this strategy is a protocol establishing its own balance sheet using retained earnings. This proprietary capital will never flee just because a competitor launches a higher short-term yield or loyalty program.
Binance’s recent response proves the weight of an owned balance sheet. After suffering a $6.3 billion hack on October 29, Binance deployed its SAFU fund and corporate treasury (primarily the latter) to fully cover the shortfall, reinstating withdrawals within days and limiting net outflows to just 3%.
Though still early stages, market confidence in its formidable balance sheet clearly prevented a systemic bank run.
As DeFi penetrates deeper into private credit, tokenized equities, and other RWAs, market demand for insurance capital and risk capital will only grow. The inevitable outcome: protocols that genuinely provide and operate capital will capture vastly more value than pure software delivery vendors.
Moat #3: Proprietary Data
The third moat revolves around information and data, but requires a crucial caveat. DeFi has historically operated on the principle that “financial data should be publicly transparent,” which works exceptionally well for overcollateralized crypto lending and spot trading. Yet as more RWAs come on-chain, business models built around “non-public information” will proliferate.
The true moat lies in the combination of “proprietary non-public data” and the “ability to price risk and underwrite based on that data.”
Credit Risk Curators
Lending has been the earliest sector to reveal this divergence. On platforms like Morpho, Kamino, and Euler, risk curators currently manage $9.5 billion in depositor capital for risk assessment:
Chart: Growth trend of DeFi risk curator managed assets, currently reaching $9.5 billion (Source: DefiLlama)
These curators are granted limited permissions within vault smart contracts to allocate capital across lending markets, set risk parameters, and monitor collateral positions. Despite collectively managing $9.5 billion, their captured revenues remain pitifully small, especially compared to traditional asset managers who wield greater discretion and accountability.
In fact, many managers charge almost nothing. Steakhouse Financial, the largest DeFi risk curator by AUM, takes a mere 22 basis points in carry on managed assets, while the industry-wide average curator fee sits at just 19 basis points.
This stems partly from unappealing current risk-adjusted returns in DeFi (possibly amplified by idle capital sitting in Aave); and partly because Morpho and Kamino have commoditized the vault layer, inserting themselves between curators and depositors—again validating our earlier distribution channel argument.
The deeper issue, however, is that the vast majority of risk curators rely on identical public data. If you’re doing collateralized lending backed by ETH or BTC, prices, liquidation thresholds, and collateral types are universally visible. AI makes replicating risk modeling effortless, eventually forcing managers to compete solely on singular speculative judgments and historical performance, which structurally compresses long-term management fees.
In contrast, consider protocols like OnRe. It operates as an on-chain reinsurer, raising investor capital to underwrite real-world insurance risks. To do so, it must master historical claims data, loss probability curves, and specific policy wording, requiring specialized actuaries capable of interpreting these datasets.
None of this information is queryable on block explorers. Regardless of how advanced a frontier LLM is, without access to internal reinsurance ledgers, it simply cannot calculate underwriting pricing.
This explains why OnRe commands a 75-basis-point management fee plus 20% performance carry. At current return levels, its annualized fee rate reaches approximately 3%, roughly 16 times that of the average DeFi risk curator.
While these undeniably represent fundamentally different businesses with distinct risk profiles—and the fee premium partially compensates for illiquidity and principal loss risk—it perfectly validates our core thesis: entities wielding proprietary data, underwriting expertise, and client relationships command vastly higher prices. If you only possess the same public data and tools as everyone else, the endgame is invariably commoditization.
What Does This Mean for Value Capture?
Following this analytical framework leads to an obvious question: If DeFi experiences long-term mass adoption, who is best positioned to capture the value premium, and who will struggle to monetize?
Public Chain Infrastructure (L1s & L2s)
A somewhat harsh reading is that infrastructure tokens like L1s and L2s currently face headwinds. Historically, chains inadvertently endowed nascent DeFi applications with the moats of “distribution” and “capital,” but this free ride is rapidly evaporating.
Historically, chains controlled distribution because users naturally anchored their capital on the most active chain. Building on Ethereum reached Ethereum users; building on Solana reached Solana users. When launching Uniswap, Compound, Aave, Kamino, or MakerDAO from scratch, booting up within an established ecosystem was critical.
But the evolution of cross-chain interoperability is diluting this lock-in. Today, a fomo user can simply onboard via a mobile app to trade any asset across chains without caring which underlying chain powers the transaction.
In the future, growing waves of new users will flood into DeFi through Revolut, Nubank, and Robinhood. These tech giants already command tens of millions of users and don’t require customers to self-custody wallets.
Past L1s fueled capital circulation via ecosystem wealth effects: as SOL or ETH appreciated, users had more on-chain investable capital, while cross-chain friction and community tribalism kept it trapped.
Today, cross-chain flows are seamless, with fresh capital entering directly via stablecoins, tokenized assets, and institutions. Furthermore, RWA issuers like Securitize, Superstate, and Ondo prefer multi-chain deployments across every major chain.
As for data, public chains have essentially never built meaningful moats. The vast majority of on-chain financial information was inherently public, meaning chains lacked defensible proprietary data right from the start.
AI is even eroding blockchain’s other traditional advantage: the developer ecosystem. One of Ethereum’s historic moats over newer chains like Solana or Aptos lay in its mature tooling, battle-tested codebases, and vast integrations. But when AI can effortlessly port and refactor apps across VMs, that edge diminishes sharply.
Macroscopically, generic public chains are losing their original distribution and capital premiums, while lacking inherent data walls. This forces them to ultimately compete rawly on execution and settlement efficiency. With abundant block space, unit transaction fees continue to compress.
None of this implies public chains lack intrinsic value. As Tushar articulated years ago in his summit talk, L1 tokens function simultaneously as capital assets, commodity assets, and stores of value. Even charging just 1 cent per transaction, if Solana ultimately processes 1 billion payments daily, annual fee revenue would hit $3.65 billion.
As reflected in the data below, on-chain processing volumes continue to climb steadily:
Chart: Daily transaction volume on Solana, reaching 162 million in 2026 (Source: Blockworks Research)
The core takeaway is this: the underlying thesis for why public chains will hold value in the future may differ drastically from what investors originally imagined.
Application & Protocol Layers
Regarding applications and protocols, the ultimate winners capturing the most value will inevitably master at least one of these three moats: building distribution barriers through hard-to-migrate direct customer relationships (e.g., Robinhood or Hyperliquid), securing capital moats via balance sheets capable of absorbing losses, or monopolizing proprietary data moats through unmatched underwriting capabilities.
No matter how advanced AI models become, these three factors remain exponentially harder to replicate than software itself.
Whether value ultimately flows to corporate equity or protocol tokens does not dictate the winner; both structures will share in this premium. Recent SEC guidance and proposed rule drafts have already mapped paths for compliant token value capture, and we will actively position ourselves to capitalize on these opportunities.
There is only one truly critical metric: regardless of whether wrapped in token or equity form, does the enterprise control scarce elements, wield pricing power as a result, and successfully convert that advantage into tangible investor returns?
As more financial activity migrates on-chain, the valuation and profitability gap between protocols wielding genuine moats and those possessing “nothing but source code” will widen dramatically. As one of the largest investors in this space, we dedicate immense effort to identifying which full-stack projects truly possess and defend these critical barriers.
Isn't This All Just Traditional Incumbent Advantage?
Following this analysis, a natural conclusion emerges: legacy incumbents seem poised to dominate forever. Robinhood holds millions of direct customers, Goldman Sachs likely possesses a stronger balance sheet than any DeFi prime broker, and Moody’s commands deep data reservoirs and regulatory licenses.
If software approaches zero cost while moat construction remains prohibitively expensive, established corporations already sitting atop these barriers logically hold the upper hand.
Startups retain one ace card, however: the willingness to take risks too daunting for incumbents. AI serves as a double-edged sword here: it enables Robinhood to spin up a DEX in a weekend, but grants a three-person startup team the exact same development velocity.
As the cost of shipping reliable products plummets, startups' real historical opportunity lies in boldly targeting white spaces and customer segments that legacy giants avoid due to regulatory friction, operational overhead, or reputational risk. Several compelling examples already exist in practice.
Ethena engineered synthetic dollar products around delta-neutral hedging. Basis trading strategies aren’t foreign to Wall Street, but traditional firms lack the appetite or compliance clearance to deploy them on crypto exchanges. Ethena leveraged this blind spot to rapidly scale to $5.4 billion AUM.
Pump.fun and fomo launched from near-zero, inventing entirely new trading mechanics around memecoins—a novel asset class—and seamlessly captured massive distribution entry points, emerging as two of the brightest winners of this cycle.
Hyperliquid invaded the derivatives space traditionally monopolized by Binance, Bybit, and OKX, rapidly snatching market share. It also launched pre-market perpetuals for unlisted tech giants like OpenAI and Anthropic—products conventional regulated exchanges would never dare touch.
Such niches continuously attract fresh traders and incremental liquidity. As the stats below illustrate, newly created wallets on the Hyperliquid platform spiked dramatically on the day of the SpaceX IPO:
Chart: Daily new wallets on Hyperliquid from May 16 to June 14, 2026, surpassing 40k on the day of the SpaceX IPO (Source: HyperTracker)
On the capital moat, Project 0 specializes in financing DeFi collateral that traditional prime brokers refuse to touch. Regarding proprietary data, USD.ai underwrites and prices loans secured by GPU compute hardware—an asset class conventional banks lack any experience handling whatsoever.
The common thread weaving through these success stories is clear: they target micro-verticals that incumbents deliberately ignore due to regulatory friction, operational complexity, reputational concerns, or initial skepticism over scale. Historical precedent suggests such asymmetric competitive entry points will continually emerge.
This precisely explains the profound excitement surrounding the intersection of AI and DeFi: AI drastically slashes the manufacturing costs of financial products themselves, while DeFi empowers startup teams to bypass reinventing traditional financial stacks from scratch, granting immediate access to global clients and capital mobilization.
Naturally, elevated risk appetite alone does not constitute a durable moat. Over time, regulation inevitably catches up, and once a new market proves viable, legacy giants eventually pour in en masse—as witnessed in the spot Bitcoin ETF landscape.
But startups don’t need to construct a complete fortress on day one. They merely require an entry wedge abandoned by incumbents, seize the temporal window to fortify it into one or more of the aforementioned moats, and subsequently ignite compounding, snowball-style expansion.
Conclusion
We firmly believe that trillions of dollars in financial activity will migrate on-chain over the next decade, while AI drives the software development costs serving these activities to unprecedented lows.
Yet, as software increasingly becomes a commodity, the core elements underpinning a financial enterprise’s intrinsic value have not depreciated. Direct customer relationships, genuine risk capital, and non-public proprietary data remain exceptionally difficult to acquire. Enterprises controlling these scarce assets will firmly anchor pricing power.
For these reasons, we remain profoundly optimistic about next-generation DeFi startups. Democratized AI development tools arm emerging teams with faster, lower-cost weapons to challenge incumbents. Once validated and fortified with even a single moat, they stand to evolve into immensely resilient, enduring franchises.
Ultimately, code can be forked at any time or generated directly by top-tier AI models. But no model can conjure these moats out of thin air for you. As investors in crypto, the true alpha opportunities reside here. (Full disclosure statement available in the original Multicoin publication.)