Multicoin: Building a DEX Takes a Weekend — What Wins in DeFi in the AI Era?

Five years ago, building a decentralized exchange (DEX) or lending protocol meant raising $3 million, hiring five engineers and spending six months to ship a minimum viable product (MVP). Today, using Claude Code, a founder can get most of that done in a single weekend.

Just point a frontier AI model at building a perpetuals exchange, and you get the matching engine, risk engine, trading interface, API and most of the essential core functionality out of the box.

AI is lowering barriers to competition across almost every corner of the software industry, but financial services may be one of the hardest industries to commoditize. The reason: the most valuable parts of a financial business have little to do with the software itself.

Building a loyal customer base, accumulating a large balance sheet, and developing proprietary data and risk-assessment capabilities all require years of relationship-building, track record and trust. AI can dramatically cut the cost of software development, but it cannot make any of those capabilities easier to replicate.

In other words, a model can write a lending protocol in an afternoon, but it cannot conjure $10 billion of stablecoin deposits into a smart contract out of thin air, nor recruit 28 million customers for a newly founded digital bank.

We believe this trend is extremely bullish for DeFi. Over the next decade, trillions of dollars of financial activity will move onchain, and AI will sharply reduce the cost of building the products needed to service that activity. But at the same time, the core elements that determine a financial firm's value and competitive moat have not become any easier to replicate.

A New Pillar in the DeFi Value Capture Framework

For years we have been thinking about value capture in DeFi, adjusting our views as the market has evolved.

Early on, we focused on how application tokens capture value through models such as work tokens and the burn-and-mint equilibrium (BME). We then shifted our focus to DeFi's defensibility and unforkable state — the idea that anyone can copy open-source code, but not liquidity and network effects.

More recently, we argued further that DeFi protocols can only capture more value when they take on and manage some form of risk. Each of these views has been validated to some degree, and this piece extends our thinking further.

First, some context: We believe DeFi will continue to take market share from centralized finance (CeFi) and will eventually cover the entirety of traditional finance.

DEX market share relative to centralized exchanges (CEXs) is already at an all-time high, and it is still growing.

Source: The Block, DefiLlama

Meanwhile, tokenized assets are also starting to see genuine market adoption, whether US Treasuries, private credit or equities. Importantly, native crypto users are no longer the only driver of this growth. Traditional financial institutions, fintechs, institutional asset managers and traditional brokerages are all building onchain products and bringing their customers into the onchain world.

Source: rwa.xyz

The catch is that today's leading DeFi protocols won't necessarily be the ultimate beneficiaries of this growth trend. As the cost of software development and data analysis keeps falling, value will flow to players that control scarce resources software cannot create.

We see three primary moats in DeFi:

1. Distribution
2. Capital
3. Proprietary Data

There is also a fourth factor — risk tolerance. It works differently, functioning more like a wedge into the market, and we will discuss it at the end of this piece.

Moat 1: Distribution

Exchanges and other stablecoin distributors

One of the most influential theories of the early crypto industry was Joel Monegro's "Fat Protocols" thesis, put forward in 2016.

The theory held that most of the internet's value was generated at the application layer — companies like Google, Facebook and Amazon — but that blockchains would upend this relationship, with most value captured by the underlying protocols.

Reality has been moving in the opposite direction, and the trend is accelerating.

As building protocols gets cheaper and cheaper, value starts shifting to whoever owns the customer relationship. Firms that own the customer relationship can route users and capital to the highest-yielding venue. Sometimes that entity is an application; sometimes, as with Hyperliquid, it is itself a blockchain that doubles as an application.

Our definition of distribution is simple: control over where order flow or capital balances go.

Note, though, that not all distribution is equally sticky. Funded accounts that are hard to migrate can form a real competitive barrier, whereas a mere user interface is far more vulnerable to competitive threats. We discuss this further below.

The stablecoin value chain is the clearest expression of distribution.

Historically, once you reach a certain scale and rates aren't near zero, stablecoin issuance is a very profitable business.

Tether and Circle hold dollar reserves and earn interest on them, and in most cases they do not pay interest to holders. Tether alone has generated more than $20 billion of interest income at near-100% margins since 2021 through this model.

Source: DefiLlama

Take Hyperliquid, for example: more than $6 billion of USDC sits on its platform as collateral for derivatives trading. That capital generates substantial reserve income for Circle and its distribution partners.

Hyperliquid gradually realized that it is its own platform that brings those USDC balances into existence, and that gives it bargaining power: if Circle is unwilling to share the revenue, Hyperliquid can replace USDC with another stablecoin that is willing to — and in theory it could even issue its own stablecoin.

Earlier this year, Hyperliquid struck a deal with Coinbase through its Aligned Quote Asset (AQA) framework.

Under the agreement, roughly 90% of the reserve income on USDC held on the Hyperliquid platform flows into the protocol's Assistance Fund to buy HYPE, rather than going to Circle or Coinbase.

On the $5.5 billion of USDC in the treasury, the income corresponding to that 90% share currently generates about $185 million of revenue for Hyperliquid.

Hyperliquid is responsible for none of the underlying infrastructure here — token minting and redemption, treasury management or regulatory licenses. What it holds is the distribution channel, and that alone is enough for it to demand a larger share of the economics.

Wallets and front ends

Wallets and front ends are another expression of distribution.

In crypto's early years, non-custodial wallets struggled to monetize effectively. They later came to realize that the most profitable business in crypto is trading, and so a few years ago they began making trading their core product.

On trading fees, MetaMask charges 0.875%, while Uniswap's average trading fee over the past 12 months is just 0.13%. What's more, most of Uniswap's trading fees flow to liquidity providers (LPs), not to the protocol itself. In short, the wallet earns far more revenue from a trade than the platform that actually executes it.

On the other side, exchanges must attract liquidity providers, maintain complex infrastructure and compete on execution quality, while MetaMask only has to win and keep the customer relationship.

If a given DEX can't offer competitive pricing, MetaMask can route the trade to another platform. As its order flow scale keeps growing, it can even gradually internalize more trades, thereby retaining more of the economics.

A similar trend is starting to emerge in perpetual futures trading.

Phantom lets users trade perpetuals through Hyperliquid and charges an additional 5 basis points on top of the 4.5 basis points Hyperliquid charges.

Hyperliquid runs the risk engine and matching system and provides liquidity, while Phantom earns more revenue simply by owning the interface and the customer relationship.

That model has limits, though.

Today, development teams in the Hyperliquid ecosystem have entered a fiercely competitive red ocean, and a price war is underway, with teams cutting fees to improve user retention.

In other words, even the front-end layer cannot escape commoditization.

More broadly, crypto wallets themselves are fragile, because the cost of switching is low for users. A user simply imports a private key or seed phrase into another wallet and can access the same funds through a new interface within minutes.

By contrast, migrating a bank or fintech relationship is far more complicated. A user must open a new account, complete KYC, move funds through traditional payment rails, re-link their payroll account and potentially update other connected services.

Fintechs and digital banks

Take Robinhood, which currently has 28 million funded accounts.

Today, Uniswap is the main DEX on Robinhood Chain, and Robinhood has no objection to that.

But as AI lowers the cost of building financial software, nothing stops Robinhood from building and launching its own DEX protocol and routing users to its own platform.

With its enormous user reach and advanced coding tools, Robinhood could launch one of the largest Layer 2 DEXes in DeFi almost overnight.

Of course, we are not saying Robinhood will definitely do this, because historically owning liquidity has not been Robinhood's primary goal. But the company has indeed begun shifting a large volume of trading flow from Kalshi to its own prediction market, Rothera.

What we really want to emphasize is this: building the protocol is no longer the hardest part. The genuinely hard part is attracting users, capital and proprietary data.

As a result, a DeFi value-distribution chain is taking shape:

• Front ends (such as Phantom) capture more of the economics from trading platforms;
• Trading platforms (such as Hyperliquid) capture more from stablecoin issuers (such as Circle).

In all of these cases, the party that controls distribution has stronger pricing power over the underlying infrastructure.

But how much value each layer ultimately retains depends on how easy it is for users to leave.

Hyperliquid succeeded in capturing about 90% of the reserve income, while front-end applications in the Hyperliquid ecosystem have already started competing on price, pushing fees down to single-digit basis points.

Moat 2: Capital

A natural conclusion from the previous section is that protocols and back-end infrastructure without direct user access are at a disadvantage.

In some cases that is true. But beyond distribution there are other moats, and one of the most important is capital.

Capital is the lifeblood of financial markets. Traders need capital to use leverage, market makers need capital to keep quoting, and lending protocols need deposits to make loans.

AI can help write code and automate workflows, but it cannot create money out of thin air — and these financial systems depend on money to run.

When something goes wrong — Aave's KelpDAO incident, or MakerDAO's "Black Thursday" — someone always has to absorb the loss.

Before going further, let's talk about liquidity.

Liquidity has a log(n) network effect: traders flow to where liquidity is deep, and liquidity follows traders as they migrate.

Hyperliquid's open interest is a genuine competitive advantage, but that capital actually belongs to users, who can move it to another platform at any time.

We have already seen DeFi liquidity migrate from one protocol to another, often driven by token incentives and vampire attacks. And when those incentives disappear, a large amount of that liquidity often disappears with them.

Two forms of capital can actually create value with long-term defensibility.

DeFi prime brokers

Prime brokers (PBs) are key players in traditional financial markets.

They provide leverage, finance trades, and allow hedge funds to use the same pool of capital across markets and strategies without having to post collateral separately at each trading venue.

This is an enormous business.

Driven by growth in the hedge fund industry and the financing-dependent multi-manager pod shop model, industry-wide prime brokerage revenue grew from roughly $15 billion in 2020 to about $27 billion in 2024.

We think the opportunity in DeFi could be even larger.

Traditional prime brokers mostly serve large institutions, because servicing smaller clients is too operationally expensive. Smart contracts can automate much of that work, making these services available to everyone — from an individual trading only $500 to an investment team running hundreds of millions of dollars at Millennium.

Today, traders typically have to post collateral separately across platforms such as Hyperliquid, Aave, Lighter, Morpho, Variational and Kamino.

That model is wildly capital-inefficient. And as tokenized equities, US Treasuries, private credit and other real-world assets (RWAs) keep moving onchain, the problem will only get worse.

For example, if a fund holds a long position in tokenized equities on one platform and a short position in equity index perpetuals on another, it should not be required to fully collateralize both positions when the risks of the two can offset each other.

Today, building the software to manage this kind of business is getting easier and easier. What is genuinely hard is having enough balance sheet to finance those positions, access to low-cost funding, and the risk management capability needed to avoid major losses.

Over time, the prime broker with cheaper funding can offer more favorable financing terms, attract more customers and use its scale to lower funding costs further.

In short, this business should have significant economies of scale.

Backstopping risk for DeFi protocols

The second way capital forms a moat is by backstopping risk.

This connects directly to our earlier article on risk management in DeFi DAOs, where we argued that protocols able to manage risk have associated tokens that are harder to displace through forks than protocols that don't. Hyperliquid's HLP vault, for example, provides liquidity by market making and taking on liquidation positions.

Depositors share in the profits, but they must also bear losses. In March 2025, HLP lost roughly $4 million closing out a large ETH position, and the subsequent JELLYJELLY incident exposed it to further risk.

Both episodes highlighted the importance of insurance funds for derivatives DEXes.

Aave's Umbrella plays a similar role: users backstop bad debt in exchange for corresponding yield.

Anyone can copy the HLP or Umbrella smart contracts, but they cannot copy the capital inside them. There is a problem here, though: that capital comes from third parties. If depositors find higher yields elsewhere, they may withdraw their funds.

One counterargument is that some of this capital is somewhat sticky, because the interests of token holders and depositors may overlap. For example, HYPE holders may be willing to keep some of their dollar capital in HLP in order to make the exchange safer and protect their own token investment.

The most durable form of this strategy is a protocol using retained earnings to build its own balance sheet. That kind of capital does not easily flow out just because a newer, more attractive yield opportunity or points program appears.

Binance's recent experience illustrates the importance of this model.

After suffering a hack on October 29 that cost it $6.3 billion, Binance covered the losses with its SAFU reserve fund and corporate capital (mainly the latter) and reopened withdrawals within days.

In the end, net outflows amounted to only 3% of total assets.

It is still too early to draw firm conclusions, but on first read, the market's trust in the strength of its balance sheet may have helped it avoid a bank run.

As DeFi gradually moves into private credit, tokenized equities and other real-world assets, demand for insurance and risk capital will keep growing.

The end result: protocols that actually provide and manage capital will capture more value than protocols that merely provide software.

Moat 3: Proprietary Data

The third moat is information, but there is a caveat to note here.

DeFi is built on the idea that financial data should be public, and that works very well for businesses such as crypto-collateralized lending and spot trading. But as more and more real-world assets move onchain, we will see more businesses built around non-public information.

The real moat lies in combining proprietary data with the ability to use that data for risk assessment.

Credit curators

Lending is the clearest illustration of this trend. Curators on platforms such as Morpho, Kamino and Euler are currently responsible for assessing risk and managing depositor capital, with $9.5 billion in aggregate under management.

These curators receive limited control permissions within the vault smart contracts: they can adjust which lending markets capital is allocated to, set risk parameters and monitor collateral positions.

Even though they manage $9.5 billion combined, the revenue they generate is very limited — especially compared with traditional asset managers, who have greater discretion and responsibility.

In fact, many managers don't charge any fees at all. By assets under management (AUM), Steakhouse Financial, the largest risk curator in DeFi, charges just 22 basis points, and the average fee across all managers is 19 basis points.

Part of the reason is that DeFi's current risk-adjusted yields are not attractive, which may partly be related to the large amount of idle capital in Aave. Another reason is that Morpho and Kamino have commoditized the vault layer and sit between risk managers and depositors — again reflecting the distribution dynamic we discussed earlier.

But the bigger problem is that risk managers mostly use the same data.

If ETH or BTC is used as loan collateral, the asset prices, liquidation rules and collateral information are all public.

AI makes risk models very easy to copy, so risk managers can only compete on judgment (such as risk appetite, much of which is already predefined by algorithms) and on historical track record.

Over a long enough time horizon, this will inevitably put pressure on fee levels.

OnRe is a different case by comparison.

OnRe is an onchain reinsurance company that raises capital from investors and uses it to underwrite real-world insurance risk. To do so, it needs historical claims data, loss probabilities and the insurance contracts themselves — not to mention professionals who actually understand that data. None of this information can be obtained from a block explorer. A frontier AI model with no access to reinsurance business data, no matter how intelligent, cannot complete the related risk underwriting work.

That is also why OnRe can charge a 75 basis point management fee plus a 20% share of returns. At current return rates, that works out to roughly 3% annualized in fees, 16 times the average fee level of DeFi risk managers.

Of course, these are two businesses with different risk characteristics, and part of the fee premium also reflects asset illiquidity and the risk of potential losses. But that is precisely the point: firms with proprietary data, risk underwriting expertise and customer relationships can charge significantly higher fees.

If the data and tools you have are exactly the same as everyone else's, you will most likely end up as just another commoditized product.

What This Means for Value Capture

This framework naturally raises a question: if DeFi grows substantially over the long term, which participants are most likely to capture value? Some groups will benefit, while others will face greater difficulty monetizing.

L1s and L2s

One fairly bearish interpretation is that infrastructure tokens such as L1s and L2s are in trouble. The reason is that in the past, blockchains could provide new DeFi applications with two of the three moats above: distribution and capital.

Historically, blockchains had distribution because DeFi users typically kept their assets on the chain where they were most active.

If you build an application on Ethereum, you can reach Ethereum users; if you build on Solana, you can reach Solana users. If you were creating Uniswap, Compound, Aave, Kamino or MakerDAO from scratch today, this ecosystem advantage would matter a great deal.

Because you have no existing customers, no assets and no financial relationships, launching in an ecosystem where users are already gathered can significantly reduce the difficulty of starting a business. But today, advances in interoperability technology are eroding this advantage.

Now, users can enter DeFi directly through an application and trade assets on any chain with almost no need to think about the underlying blockchain.

More and more new users will also enter DeFi through fintechs such as Revolut, Nubank and Robinhood. These platforms already have tens of millions of customers, and users can use the relevant services without first funding a wallet.

L1s also provided capital through ecosystem wealth effects. As SOL and ETH prices rose, users had more capital to deploy into applications on those blockchains.

In addition, because cross-chain operations used to be more difficult and users often had a strong sense of belonging to a particular chain, capital was more likely to stay within its original ecosystem.

But today, cross-chain operations have become much easier, and capital can also enter DeFi through stablecoins, tokenized assets and institutional money. As a result, new protocols are becoming less dependent on the ecosystem wealth effect generated by any single token.

Issuers of real-world assets also want to deploy assets onto every blockchain, and projects such as Securitize, Superstate and Ondo have already demonstrated this trend.

As for data, blockchains never really formed much of a moat. Almost all onchain financial information is public, so blockchains and the applications on them have never held much proprietary data with which to build competitive barriers.

AI is also eroding another former advantage of blockchains: the developer ecosystem.

One of Ethereum's major advantages over newer blockchains such as Solana or Aptos was its mature development tools, audited code libraries and rich set of integrations. But as AI makes it easier to migrate applications between virtual machines (VMs), this advantage is becoming less important.

More broadly, general-purpose blockchains are losing their advantages in distribution and capital, and they never really established a data moat to begin with.

That means they will ultimately compete mainly on execution and settlement — where blockspace supply is ample and transaction fees keep falling. But that does not mean blockchains themselves have no value.

As Tushar laid out in a summit talk a few years ago, L1 tokens have the attributes of capital assets, commodity assets and store-of-value assets at the same time. If Solana eventually processes 1 billion payment transactions a day at a fee of $0.01 per transaction, its annual revenue would reach $3.65 billion.

As the chart below shows, transaction volume is still growing.

Source: Blockworks Research

The core takeaway: where L1s and L2s derive their value from may be completely different from what investors originally expected.

Applications and protocols

For applications, the businesses able to capture the most value will need at least one of the following three moats:

• Distribution: direct customer relationships that are hard to migrate, such as Robinhood or Hyperliquid.
• Capital: a balance sheet that can absorb losses.
• Information: the ability to underwrite risk using data that other participants cannot obtain.

No matter how powerful AI models become, all three capabilities are far harder to replicate than software itself.

We don't think it matters much in the end whether value flows to equity or to tokens. Our view is that both can capture value.

Recent SEC guidance and rule proposals have provided a path for tokens to capture value in a compliant way, and we fully intend to invest in opportunities in this area.

What really matters is whether a business — whether it takes the form of a token or equity — controls some scarce resource, whether that gives it pricing power, and whether it can convert that advantage into investor returns.

As more and more financial activity moves onchain, the gap between protocols with moats and protocols that merely have software will widen further.

As one of the largest investment firms in the space, we spend a great deal of time studying which projects across the entire stack truly have these moats and can sustain them over the long term.

Doesn't This All Ultimately Benefit Only the Industry Giants?

From the analysis above, it is easy to reach a reasonable conclusion: the ultimate winners will be the existing industry giants.

Robinhood has the customers, Goldman Sachs likely has more capital than any DeFi prime broker, and Moody's holds the data and the relevant licenses.

If software development is becoming free and the real moats are precisely those expensive resources, then firms that already control those resources naturally have a competitive edge.

But startups still have one advantage: they are willing to take on risks that the giants are not.

Part of the reason is that AI cuts both ways. The same set of tools that lets Robinhood launch a DEX in a weekend also lets a startup with only three employees do the same.

As the cost of building a competitive product keeps falling, the real opportunity for startups lies in going after markets the giants are unwilling to enter and serving customers whose needs remain unmet.

In fact, several compelling examples have already emerged.

Ethena

Ethena built a synthetic dollar product around a delta-neutral hedging strategy. Traditional financial institutions are obviously familiar with this strategy — it's the basis trade — but they were previously unable or unwilling to bring it to crypto exchanges. Today, Ethena's assets under management (AUM) have reached $5.4 billion.

Pump.fun and fomo

Pump.fun and fomo burst onto the scene, creating an entirely new trading experience around memecoins, an emerging asset class, and using that as a wedge to build massive distribution. Today they have become two of the most successful projects of this crypto market cycle.

Hyperliquid

Hyperliquid entered a derivatives market dominated by Binance, Bybit and OKX and quickly captured market share. It has also begun listing perpetual futures on pre-IPO companies such as OpenAI and Anthropic — a market traditional exchanges would not touch. These markets attracted new traders and liquidity. The number of wallets created on the day of the SpaceX IPO illustrates this.

Source: HyperTracker

On the capital side, Project 0 provides financing against DeFi collateral that traditional prime brokers are unwilling to accept.

On proprietary data, USD.ai can underwrite loans using GPUs as collateral, an asset class banks have almost no experience handling.

What these cases have in common is that they all entered markets that traditional industry giants were not yet serving.

The reason may be regulatory constraints, operational complexity, reputational risk, or simply that the opportunity looks too small to the giants. As long as these conditions persist, startups will always have opportunities to enter the market through asymmetric competition. Crypto's history has proven this many times, and we expect it will continue to do so.

This is exactly why the combination of AI and DeFi is so compelling. AI is dramatically lowering the cost of building financial products, while DeFi allows startups to reach customers and access capital directly without having to rebuild the entire financial system from scratch.

Of course, risk tolerance itself is not a durable moat.

Regulation will eventually catch up, and the giants will enter the competition once a market is proven (the spot Bitcoin market is one example). But startups do not need to establish a moat immediately at founding. They only need to find a market wedge that the giants are unwilling to pursue, and buy enough time to develop that breakthrough into one or more of the three moats.

After that, they can begin to accumulate and expand their advantages.

Conclusion

We believe that over the next decade, trillions of dollars of financial activity will move onchain, and AI will make the financial products that serve that activity easier to build than ever before. But even as software gradually commoditizes, the core elements that truly determine the value of a financial business have not changed.

Customers, capital and proprietary data remain hard to acquire, and the firms that control these resources will have pricing power. That is also why we are so bullish on the next generation of DeFi companies.

As AI development tools become increasingly widespread, startups can now compete with industry giants faster and at lower cost than ever before. Once a new company successfully establishes one of these moats, it has the opportunity to build a company with an extremely durable competitive advantage.

Ultimately, you can copy code, and you can use the latest and most advanced AI models to regenerate it. But you cannot simply generate these moats — and that is precisely where we see the opportunity as crypto investors.

Join the official Coincamps community:

X: https://x.com/coincamps

Telegram: https://t.me/coin_camps