Bits Up

Code in a Weekend Is Not Adoption: Multicoin’s November 2026 Case for What Might Make Onchain Finance Stick

This is a structured reading of an essay by Spencer Applebaum, an investment partner at Multicoin Capital, published in Chinese compilation by 大钳子 on PANews on November 11, 2026. It is not a new development. Figures, comparisons, and uncertainty below are those stated in that essay. Where a growth claim appears, the metric, window, and baseline are named, and the limitation the essay itself attaches is kept in view. Higher token prices are not treated as evidence of adoption, and no user or address counts are inferred beyond what the text reports.

Original article: PANews, November 11, 2026.

The cost of the software fell; the scarce inputs did not

Applebaum’s starting contrast is operational, not a price call. Five years before the essay, he writes, shipping a decentralized exchange or a lending protocol meant raising about $3 million, hiring five engineers, and spending six months on a minimum viable product. By the time of publication, he argues, a founder using Claude Code could complete most of that work in a weekend. A frontier model asked to build a perpetual-futures exchange could, in his account, produce a matching engine, a risk engine, a trading interface, an API, and most of the basic functions.

The essay’s claim is narrower than “AI makes finance easy.” Applebaum treats financial services as unusually hard to commoditize because the valuable pieces are only loosely related to software: a loyal customer base, a large balance sheet, and proprietary data plus risk assessment. Those, he writes, take years of relationships, a track record, and trust. A model can draft a lending protocol in an afternoon. It cannot, on his examples, place $10 billion of stablecoin deposits into a smart contract, or recruit 28 million customers for a newly formed digital bank.

Multicoin’s stated view, as of that November 2026 essay, is that the shift is still favorable to DeFi: over the following decade, trillions of dollars of financial activity could move onchain, while AI cuts the cost of building products that serve it. The elements that set a financial firm’s value and its barriers, they argue, did not become easier to copy. That is an expectation they attribute to themselves, not a measured flow.

How their value-capture view changed

Applebaum situates the essay in a sequence of Multicoin arguments, each only partly borne out in their telling. Early on they focused on application tokens capturing value through work tokens and burn-and-mint equilibrium designs. They later emphasized defense and “unforkable state”: open-source code can be copied; liquidity and network effects cannot. More recently they argued that a DeFi protocol captures more value only when it takes and manages some form of risk. The November 2026 piece extends that line rather than replacing it.

He states as a premise that DeFi would keep taking share from centralized finance and, eventually, cover traditional finance. The only market-share fact offered for exchanges is relative: DEX share of volume versus centralized exchanges had reached a historical high and was still growing, with The Block and DefiLlama cited as sources. The essay does not give the percentage, the absolute volume, a user count, or an address count, so the claim cannot be read as a retention or adoption rate.

On tokenized assets, the essay says real market adoption had begun for U.S. Treasuries, private credit, and equities, citing rwa.xyz. The important qualifier in the text is who was pushing it: not only crypto-native users, but traditional financial institutions, fintech firms, institutional asset managers, and traditional brokers building onchain products and bringing their own customers. Again, no asset totals or holder counts are stated in the prose.

The caution that follows is the essay’s main adoption point. Today’s leading DeFi protocols are not guaranteed to be the parties that capture that growth. As the cost of writing software and analyzing data falls, Applebaum argues, value moves toward whoever holds scarce resources the software cannot create. He names three moats—distribution, capital, and proprietary data—and a fourth factor, risk tolerance, which he treats as an entry wedge rather than a durable barrier, discussed at the end.

Distribution: who directs flow, and how hard it is to leave

Applebaum defines distribution simply as control over where order flow or balances go. He immediately qualifies stickiness. Accounts that already hold funds and are hard to move can be a real barrier. A user interface alone is easier to contest.

The stablecoin stack is his clearest case. Historically, he writes, issuance was a very profitable business once scale was reached and rates were not near zero. Tether and Circle hold dollar reserves and earn interest, and in most cases do not pay that interest to holders. Tether alone, since 2021, had earned more than $20 billion of interest income at a margin close to 100 percent, with DefiLlama cited. That figure is reserve income over a multi-year window, not a count of new users, and it depends on rates and on holders not being paid.

Hyperliquid is the bargaining example. The essay says more than $6 billion of USDC sat on the platform as derivatives collateral, generating reserve yield for Circle and its distribution partners. Hyperliquid, having facilitated those balances, could demand a share: if Circle would not share, USDC could be replaced by a stablecoin that would, or Hyperliquid could issue its own. Earlier in 2026, under an Aligned Quote Asset framework, Hyperliquid and Coinbase agreed that about 90 percent of the USDC reserve yield on the platform would flow to the protocol’s Assistance Fund and be used to buy HYPE, rather than remain with Circle or Coinbase. Using $5.5 billion of USDC in the treasury as the base, Applebaum says the 90 percent slice was then worth about $185 million of revenue to Hyperliquid. Hyperliquid does not run minting and redemption, treasury management, or the regulatory licenses. The essay’s point is that distribution alone was enough to reprice the economics. Whether those balances stay is a separate question the text does not settle with a retention series.

Wallets and front ends are the second distribution layer. Early non-custodial wallets, he writes, struggled to monetize, then shifted toward trading because trading was the industry’s most profitable activity. On fees, MetaMask charged 0.875 percent, while Uniswap’s average trading fee over the prior 12 months was 0.13 percent, with most of Uniswap’s fee going to liquidity providers rather than the protocol. In that comparison, the wallet earned more from the trade than the venue that executed it. An exchange must attract liquidity providers, run complex infrastructure, and compete on execution. MetaMask, in this account, mainly has to keep the customer relationship, and can route flow elsewhere if prices are uncompetitive, or internalize more of it as flow grows.

A similar split had appeared in perpetuals. Phantom let users trade through Hyperliquid and charged an extra 5 basis points on top of Hyperliquid’s 4.5 basis point fee. Hyperliquid runs the risk engine, the matching system, and the liquidity; Phantom’s incremental revenue comes from the interface and the relationship. The limitation is stated in the same section: builder teams in the Hyperliquid ecosystem were already in a price war, cutting fees to hold users, including down toward single-digit basis points. Even the front end, Applebaum concludes, can be commoditized. Crypto wallets are also easy to leave: a key or seed phrase imported into another wallet reaches the same funds in minutes. Moving a bank or fintech relationship is harder—new account, KYC, traditional payment rails, payroll relinking, and other attached services.

Fintech balance sheets of users are the third case. Robinhood, at the time of the essay, had 28 million funded accounts. Uniswap was the main DEX on Robinhood Chain, and Robinhood was not objecting. Applebaum’s point is capability, not a prediction: nothing in the cost of software would stop Robinhood from shipping its own DEX and routing users there, potentially one of the largest layer-2 DEXes in DeFi, given its reach and the programming tools then available. He explicitly does not claim Robinhood will do this, noting that owning liquidity has not historically been its first goal. What the firm had started to do, in his account, was move a substantial amount of trading flow from Kalshi onto its own prediction market, Rothera. No volume for that shift is given.

The distribution chain he draws is therefore stacked: a front end such as Phantom can take more of the economics from a venue such as Hyperliquid, and the venue can take more from an issuer such as Circle. Pricing power sits with whoever holds the customer or the balance. How much of that power survives depends on how easy exit is. Hyperliquid had secured about 90 percent of the reserve yield in the arrangement above, while front ends in its ecosystem were already competing fees down.

Capital: balances that can leave, and capital that is supposed to stay

Protocols and back-end infrastructure without a direct user door are, in this framework, at a disadvantage—but not always, because capital is a separate moat. Traders need it for leverage, market makers need it to quote, and lending protocols need deposits in order to lend. Code and workflow automation do not create the funds. When something breaks—the essay cites Aave’s KelpDAO episode and MakerDAO’s Black Thursday—someone has to absorb the loss.

Liquidity, he notes, has a logarithmic network effect: traders go where liquidity is, and liquidity follows traders. Hyperliquid’s open interest is described as a real advantage, but the capital belongs to users who can move it. The essay points to liquidity that has already migrated between protocols, often pulled by token incentives and “vampire” attacks, and that has often left when the incentives ended. Incentive-driven balances are therefore not treated as durable use.

Two forms of capital are offered as longer-lived.

The first is a DeFi prime broker. In traditional markets, prime brokers supply leverage, finance trades, and let a hedge fund use one pool of capital across venues instead of posting collateral everywhere. Applebaum calls the business very large: prime-brokerage industry revenue rose from about $15 billion in 2020 to about $27 billion in 2024, alongside hedge-fund growth and financing-heavy multi-manager “pod shop” models. He thinks the onchain opportunity could be larger because smart contracts can automate work that made small clients uneconomic for traditional prime brokers, covering a $500 trader and a team running hundreds of millions at a firm such as Millennium. Today, he writes, a trader typically posts collateral separately on venues including Hyperliquid, Aave, Lighter, Morpho, Variational, and Kamino—an inefficient setup that gets worse as tokenized equities, Treasuries, private credit, and other real-world assets arrive. A fund long tokenized equities on one platform and short an equity-index perpetual on another should not, if the risks offset, post full collateral twice. The software to net that exposure is getting easier to build. The hard part is a balance sheet large enough to finance the positions, cheap funding, and risk management that avoids large losses. Lower-cost funding can then win clients and, with scale, get cheaper still. He expects meaningful economies of scale. That remains an argument about market structure, not a measured book of DeFi prime-brokerage revenue.

The second form is capital that backstops protocol risk, tied to an earlier Multicoin piece on DAO risk management. Protocols that manage risk, they argued, are harder to fork away than protocols that do not. Hyperliquid’s HLP vault is the example: it makes markets and takes liquidated positions. Depositors share profits and losses. In March 2025, HLP lost about $4 million closing a large ETH position; the later JELLYJELLY episode added further risk. Both, the essay says, underline why a derivatives DEX needs an insurance fund. Aave’s Umbrella plays a similar role, paying users who backstop bad debt. The contracts can be copied; the capital inside them cannot. The limitation is explicit: the capital is third-party money and can leave if yields are better elsewhere. A partial reply in the text is overlap between token holders and depositors—for instance HYPE holders who might leave some dollars in HLP to make the exchange safer and protect the token. The more durable version, Applebaum writes, is a protocol that builds its own balance sheet from retained earnings, capital that does not chase the next yield or points program.

Binance’s then-recent episode is the illustration of balance-sheet trust, not a DeFi result. After a hack on October 29 that cost $6.3 billion, Binance covered losses with its SAFU reserve and company funds, mostly the latter, and reopened withdrawals within days. Net assets that left were 3 percent of total assets. The essay says it was too early to conclude, but that trust in the balance sheet may have helped avoid a run. As DeFi moves into private credit, tokenized equities, and other real-world assets, demand for insurance and risk capital would keep rising, in their view. Protocols that actually supply and manage capital would then capture more than protocols that only ship software.

Proprietary data: public markets compress fees; private data does not

The third moat is information, with a premise the essay flags. DeFi assumed financial data should be public, which works for crypto-collateralized lending and spot trading. More real-world assets onchain would support businesses built on information that is not public. The moat is not the data alone but data plus the ability to underwrite with it.

Lending curators are the public-data case. Curators on Morpho, Kamino, and Euler were assessing risk and managing depositor funds with a combined $9.5 billion. They hold limited vault permissions: which markets to allocate to, which risk parameters to set, and how to watch collateral. Revenue on that $9.5 billion was very limited relative to traditional managers with more discretion. Many curators charged nothing. The largest by assets, Steakhouse Financial, charged a 22 basis point management fee; the average across managers was 19 basis points. Applebaum offers two reasons that are measurement caveats, not victories: DeFi’s risk-adjusted yields were not attractive, possibly linked to idle capital in Aave, and Morpho and Kamino had commoditized the vault layer, sitting between the risk manager and the depositor—the distribution problem again. The larger issue is that the managers mostly see the same data. With ETH or BTC as collateral, price, liquidation rules, and collateral facts are public. AI makes the risk model easy to copy, so competition falls back on judgment—much of it already set by the algorithm—and on a track record. Over a long enough window, he expects that to press fees down.

OnRe is the contrasting case: an onchain reinsurer that raises investor capital and underwrites real-world insurance risk. That requires historical claims, loss probabilities, the contracts themselves, and people who can read them—none of which a block explorer supplies. A frontier model without the reinsurance data cannot do the underwriting, however capable it is. OnRe could charge 75 basis points plus a 20 percent performance fee. At then-current returns, the essay puts the annualized take near 3 percent, about 16 times the average DeFi risk-manager fee. The businesses have different risk profiles, and part of the premium reflects illiquidity and loss risk. That is the point being made: proprietary data, underwriting skill, and client relationships support a higher fee. Identical public data and tools, in this telling, drift toward a commodity.

What that implies for chains and for applications

If onchain finance grows for a long time, the essay asks who is positioned to charge for it.

The more pessimistic reading is aimed at general-purpose L1 and L2 tokens. In the past, Applebaum writes, a chain supplied two of the three moats to a new DeFi application: distribution and capital. Users tended to keep assets on the chain where they were most active, so building on Ethereum reached Ethereum users and building on Solana reached Solana users. Launching something like Uniswap, Compound, Aave, Kamino, or MakerDAO with no customers, no assets, and no financial relationships was easier inside an ecosystem where users already were. Interoperability, in the essay’s account, has been eroding that. Users can reach DeFi through an application and trade assets on almost any chain without treating the base chain as the product. Newer users were also expected to arrive through fintechs such as Revolut, Nubank, and Robinhood, platforms that already had tens of millions of customers and did not require a wallet to be funded first. Those are reported platform scales in the essay, not new measurements of onchain retention.

Chains also supplied capital through ecosystem wealth effects: as SOL and ETH prices rose, users had more to deploy in apps on those chains, and difficult bridging plus chain loyalty kept capital in place. Cross-chain movement had become easier, and capital could arrive as stablecoins, tokenized assets, and institutional funds, so a new protocol’s dependence on any one token’s wealth effect was declining. Issuers of real-world assets wanted deployment on every chain; Securitize, Superstate, and Ondo are named as already showing that pattern. Price appreciation is therefore not used here as an adoption metric. It is described as a historical funding channel that was weakening.

On data, the essay says chains never had much of a moat, because nearly all onchain financial information is public. AI was also wearing down a different historical advantage: the developer ecosystem. Ethereum’s edge over newer chains such as Solana or Aptos had been mature tools, audited code, and integrations. Porting an application across virtual machines was getting easier, so that edge mattered less. General-purpose chains, in this reading, were losing distribution and capital advantages they once had and had never built a data moat. They would compete mainly on execution and settlement, where block space is ample and fees were still falling. That is not a claim that blockchains are worthless. Applebaum cites a summit talk by Tushar a few years earlier: an L1 token can behave as a capital asset, a commodity asset, and a store-of-value asset at once. The illustration is conditional, not a forecast in the essay—if Solana eventually processed 1 billion payment transactions a day at $0.01 each, annual revenue would be $3.65 billion. The essay adds only that transaction volume was still growing, citing Blockworks Research. No current daily count is stated in the text. The conclusion he draws is that L1 and L2 value may come from somewhere other than what investors first expected.

For applications, the essay says the businesses that capture the most value need at least one of the three: distribution in the form of a direct customer relationship that is hard to exit (Robinhood or Hyperliquid are the examples); a balance sheet that can absorb losses; or the ability to underwrite with data others cannot get. Those remain harder to copy than software, regardless of model quality. Multicoin says it does not much matter whether the claim is an equity or a token; both can capture value. Recent SEC guidance and rule proposals, as described in the essay, had opened compliant paths for tokens to capture value, and the firm planned to invest in that set of opportunities. What matters, in their words, is whether the business controls a scarce resource, has pricing power because of it, and can turn that into investor returns. As more activity moves onchain, the gap between a protocol with one of these moats and a protocol that only has software would widen. They describe themselves as spending substantial time, as one of the larger investors in the sector, on which projects actually hold those moats and can keep them.

Incumbents already hold the scarce inputs; startups hold a temporary willingness to take risk

The essay grants the obvious reading: incumbents look favored. Robinhood has the customers. Goldman Sachs may have more capital than any DeFi prime broker. Moody’s has data and licenses. If software is becoming close to free and the real barriers are expensive resources, the firms that already own those resources start ahead.

The remaining startup advantage, as stated, is a willingness to take risks incumbents will not. AI cuts both ways: the same tools that could let Robinhood ship a DEX in a weekend could let a three-person company do it. As the cost of a competitive product falls, the opening is markets and customers the giants will not serve.

The cases named, all as of the November 2026 essay:

  • Ethena built a synthetic dollar around a delta-neutral hedge. Traditional finance already knew the basis trade but had been unable or unwilling to bring it onto crypto exchanges. Ethena’s assets under management were $5.4 billion. That is an AUM stock, not a statement about how many of those dollars would remain if the basis changed.
  • Pump.fun and fomo built new trading experiences around meme coins, a then-new asset category, and used that as a distribution wedge. The essay calls them among the most successful projects of that market cycle. No revenue, user, or retention figures are given.
  • Hyperliquid entered a derivatives market led by Binance, Bybit, and OKX and took share quickly. It also listed perpetuals on pre-IPO companies such as OpenAI and Anthropic, markets traditional exchanges would not list, which drew new traders and liquidity. Wallet creation on the day of the SpaceX IPO is cited, via HyperTracker, as an illustration. The essay does not state the wallet count.
  • On capital, Project 0 financed DeFi collateral that traditional prime brokers would not accept.
  • On data, USD.ai underwrote loans against GPUs as collateral, an asset banks had almost no experience with.

The common pattern in the text is entry into markets the incumbents were not serving, for reasons that might include regulation, operational complexity, reputation, or a prize that looked too small. While those gaps last, Applebaum expects asymmetric entry to remain available, as it has repeatedly in crypto’s history. AI plus DeFi is attractive, in this framing, because AI lowers product cost and DeFi lets a startup reach customers and capital without rebuilding a full financial system. Risk tolerance itself is not a lasting moat. Regulation catches up, and incumbents enter after a market is proven; Bitcoin spot markets are the example given. A startup does not need the moat on day one. It needs an opening the incumbent will not take, and enough time to turn that opening into distribution, capital, or proprietary data, and then to compound it.

What the essay treats as unchanged

Applebaum’s close, still as of November 11, 2026, is that trillions of dollars of financial activity could move onchain over the next decade, and that AI would make the products serving that activity easier to build than before. Software was being commoditized. Customers, capital, and proprietary data were not, and whoever held them would have pricing power. That is why Multicoin said it was constructive on the next generation of DeFi firms. Wider access to AI tools would let startups contest incumbents faster and more cheaply. A firm that then established one of the three moats would have a chance at a durable advantage. Code can be copied or regenerated by a newer model. The moats, in their view, cannot be generated, and that gap is the investment opportunity they describe.

None of those forward statements is a measured adoption series. The durable-use question the essay actually supports is narrower: activity tied to funded, hard-to-move accounts, to capital that is paid to bear losses, or to data outsiders cannot see, is argued to be stickier than activity tied to a clonable interface, a public risk model, or a token incentive. Where the text gives a number, the number is a stock, a fee, a historical revenue total, or a relative share—not a retention rate.

Source: panewslab.com (originally published 2026-11-11 10:29 UTC+8)

Further reading