Say you’re a lender with 100,000 USDC. You’re comfortable financing ETH-backed loans up to 65% LTV, and BTC-backed loans up to 60% LTV, on terms you’ve already worked out: acceptable oracles, acceptable liquidation rules, a minimum return you won’t go below. What you don’t know yet is whether tomorrow’s borrower will want to post ETH or BTC. Nobody does, until the borrower shows up.
Most isolated lending markets ask you to guess anyway. Before any borrower exists, you have to park your USDC in one market or the other, say 70,000 in the ETH pool and 30,000 in the BTC pool. If a BTC-backed borrower later asks for 30,000 and only 20,000 sits in that pool, the remaining 10,000 is stuck on the wrong side of a wall you built for reasons that have nothing to do with your actual appetite for risk. You were willing to make the loan. The capital was sitting uncommitted. It just wasn’t in the right room.
Isolation itself isn’t the villain. Keeping collateral configurations, oracles, and liquidation rules separate is genuinely useful: it contains risks that would otherwise spill across a shared pool of borrowers and lenders. The problem isn’t isolating risk. It’s forcing a lender to guess, in advance, which isolated market will need her money.
A lender can have a sharp view of which risks she’ll accept and no view at all of where borrowing demand will land next. Bundling those two decisions, risk appetite and demand forecasting, taxes the supply of credit. Under uncertain demand, that tax has a size you can write down.
The lender’s acceptance of credit risk is expressed through an earlier commitment to a market. The borrower’s selection happens when demand is already observable. That gap in timing is the whole asymmetry.
The decision a lender is required to make
Give that lender a name, Alice, and pin down exactly what she’s deciding. She has 100,000 USDC. She’ll lend against ETH at an initial LTV of at most 65%, or against BTC at an initial LTV of at most 60%, subject to oracles, liquidation mechanisms, and minimum returns she’s already specified. Those conditions describe the loans she would accept. Notice what they don’t describe: how much of the 100,000 should sit in either market.
A market-specific deposit forces that second, separate choice on her: perhaps 70,000 USDC for ETH-backed loans and 30,000 for BTC-backed loans. If that split reflects a deliberate portfolio limit, fine, it’s a risk decision. But if Alice would accept any mix of ETH and BTC exposure within her budget, the split isn’t a risk decision at all. It’s a forecast. It tells the protocol where she expects credit to be requested, not which credit she’s willing to supply.
What makes this a genuine asymmetry rather than an inconvenience is timing: borrowers and lenders meet the uncertainty at different moments. A borrower who wants USDC against BTC evaluates the terms when his need arises; he can see the market. Alice, who funded an isolated market weeks earlier, had to position her money before that request existed. The borrower can’t invent arbitrary terms, but he makes his financing decision holding information Alice simply did not have at deposit time.
Yes, Alice can watch the markets and shuffle unused funds as conditions move, and a manager or allocator can do it for her.¹ But a cleaner response is to stop requiring a market-specific commitment of unborrowed capital in the first place. The next section shows what the current requirement costs when she doesn’t.
¹ These reallocation strategies are real and widely used, but they come with their own execution rules, operational overhead, and latency. Latency is exactly what hurts when a borrower is waiting. They treat the symptom; the rest of this article is about the mechanism underneath it.
Thirty thousand dollars in the wrong market
Watch what happens to Alice’s split when borrowers actually show up. Take her 70,000/30,000 USDC allocation, and suppose borrowers have already drawn 60,000 USDC from the ETH market and 10,000 from the BTC market. Her book now looks like this:
| Market | Outstanding loans | Free liquidity |
|---|---|---|
| ETH/USDC | 60,000 USDC | 10,000 USDC |
| BTC/USDC | 10,000 USDC | 20,000 USDC |
| Total | 70,000 USDC | 30,000 USDC |
Now a BTC-backed borrower asks for 30,000 USDC on terms that satisfy Alice’s stated policy. She has exactly 30,000 uncommitted, but only 20,000 of it is in the market where he needs it. Without reallocation, she cannot fund his request in full, even though she has the money and accepts the risk.
The missing 10,000 USDC isn’t backing a loan. It’s backing a guess Alice never meant to make. If she has no reason to hold that reservation, it’s a restriction the allocation mechanism imposed, not one her tolerance for risk ever asked for.
Now replace her two market balances with a single uncommitted balance governed by the same policy. The 70,000 already lent stays committed, exactly as before. The other 30,000 is available for the BTC request, as long as funding it breaks no portfolio limit. The collateral and liquidation rules of the resulting loan stay every bit as isolated as they were.
So one feasible, fully-acceptable transaction was lost purely to advance placement. That raises the obvious objection: maybe Alice just chose a bad split. Could a lender who picks the best possible fixed allocation dodge these losses when demand is uncertain? The next section answers no, and says by how much.
A bound on the cost of advance commitment
Alice’s example generalizes, so let’s state the worst case precisely: her dilemma with the market count dialed up to .
Consider one allocation period with a single lender, capital , and isolated markets. Every market borrows the same asset, and the lender accepts all of their loan terms. There are no additional concentration limits.
Exactly one market receives demand during the period. That market requests up to units, with partial funding allowed. Let be the probability that market receives the request:
The lender knows these probabilities but not the outcome. Under fixed allocation, they must assign units to each market in advance, subject to:
When market becomes active, no more than can be lent. Expected executed volume is therefore:
For any permitted allocation,
The bound is attainable: allocate all capital to a market with the highest probability of receiving demand. Hence the optimal expected executed volume under fixed allocation is
Now allow the lender to keep the entire uncommitted balance available to any of the markets until the request arrives. Because the lender accepts every possible request, the active market can receive all units:
The value of deciding the allocation after demand becomes known is thus
If demand can arrive in more than one market with positive probability, . Under a uniform distribution, , so
Sit with those two numbers:
For four equally likely markets, the best advance allocation executes 25% of potential demand in expectation. Conditional allocation executes 100%.
The gap is three-quarters of the demand, and it comes entirely from when the allocation is decided, not from extra capital or looser admissibility rules. No cleverer fixed split closes it, because the bound already assumed the cleverest one.
This is a stylized single-period model;² the mechanism it isolates is not.
² The comparison assumes divisible borrowing demand, free and immediate conditional allocation, and no outstanding loans at the start, and it measures executed credit volume, not interest income or risk-adjusted return. Once a loan is funded, its capital is unavailable until the loan is released; the result never claims otherwise. The point of stripping the model this bare is to show the timing effect survives even when nothing else is in the way.
Different lenders, different constraints
Alice alone makes the mechanism visible, but a real market is full of lenders who don’t agree on anything. Some take BTC collateral, some refuse it; some wave through large exposures, some cap correlated positions hard. So it’s fair to ask whether the result survives once every lender carries her own constraints. It does. The general version follows.
Let denote a market state, including current demand and existing debt. For a given advance allocation , define as the set of new loan allocations that respect both the lender’s risk policy and the capital reserved in each market. Let enforce the same risk policy and total available capital, but without those advance market reservations.
If the conditional mechanism is capable of reproducing any permitted fixed allocation, then
For the objective of maximizing feasible executed volume, this implies
Read that inclusion carefully, because it holds only if you preserve all the relevant constraints. Accepting ETH and BTC as collateral types does not license unlimited exposure to either. If Alice caps BTC-backed loans at 40% of her portfolio, lifting the market reservation must not quietly lift that cap too. The conditional mechanism is allowed to be more flexible about where her capital goes, never more permissive about how much risk it takes.
The general statement is only a weak inequality: conditional is never worse. The earlier demand model supplies the concrete case where it’s strictly better. Neither claims anything about whether a greedy loan-selection rule maximizes returns over a long horizon. This is a statement about the feasible set, not about optimal play within it.
Express willingness to lend, not a list of destinations
If the cost comes from making Alice name a destination before demand exists, the fix is to let her stop naming destinations and state her terms instead.
An isolated market is one specific bundle: a borrowing asset, a collateral type, pricing, an oracle, liquidation rules. Alice’s acceptable conditions almost certainly span many such bundles at once, and she can say so directly, without wiring a separate pile of capital to each one.
Make that precise. Let be the set of every possible loan contract: every combination of asset, collateral, LTV, oracle, and liquidation rule. Alice’s policy carves out
the contracts she would accept. A borrower, call him Bob, arrives with his own acceptable set , the terms he needs. The two of them can trade only when
That condition is about terms, not funding. Actually executing the loan still needs available capital, a supported loan mechanism, and room under Alice’s portfolio limits. Agreement on terms is necessary, not sufficient.
Her standing policy might read: USDC loans backed by ETH below one LTV limit and BTC below another, these oracle and liquidation implementations, a minimum return that scales with maturity. Leave those conditions fixed and the same policy keeps applying no matter how the mix of incoming requests shifts. It even applies to a market that didn’t exist when she wrote the policy, as long as the new market’s parameters satisfy it; she never has to enumerate it in advance. That’s the real difference between a parameter-based offer and a list of offers nailed to specific market identifiers: a list can still share one budget and skip advance funding, but only a policy can describe which future configurations qualify.³
None of this means the policy can be a checklist you run once at origination and forget.⁴ Some risks live in implementation details, or shift over a loan’s life. And portfolio limits couple loans that look fine alone: funding two acceptable requests together can still breach a cap neither one breaches by itself. An enforceable policy has to reason about both the contract and the positions it creates.
³ This is also where a policy-based offer stays useful as a protocol adds markets over time: the offer grows to cover them for free, instead of needing Alice to return and opt into each new market by hand.
⁴ And none of it makes the matching problem trivial. Capital is finite, so compatible borrowers compete for it; Alice wants her minimum return, Bob wants the cheapest offer that clears. A protocol might rank offers by price and split proportionally at ties, but those are market-design choices, not consequences of the bound proved above.
Isolated risk, conditional liquidity
None of this requires a common collateral pool or shared liquidation losses. A market keeps its own credit rules and simply draws funds from lenders whose standing policies already admit those rules. Capital commits the moment a loan executes; repayments and losses then flow back to whoever financed that specific obligation. The isolation you wanted stays exactly where you put it. Credit-risk isolation and advance liquidity reservation are two different things, and the first can be worth keeping even when the second is worth throwing out.
On Sui, you’d express market configurations and lending commitments as Move objects and execute funding atomically with loan creation. The object layout, the matching service, and the accounting of lender claims are all real design work, and there’s a concrete trap to design around.⁵
⁵ Sui shared-object execution serializes conflicting writes to the same object, so a popular shared funding object can turn into a throughput bottleneck, the same “hot object” congestion that shapes a lot of Sui design. It’s an implementation concern, not an economic one; none of the architecture choices change the question this article is about.
There are limits to the benefit. A market that can already draw from a shared balance atomically, without material restrictions or costs, has removed the advance-reservation constraint for that capital. Conditional supply cannot free funds committed to live loans, eliminate price and liquidation risk, or predict which loan will turn out to be most profitable. If the lender’s original allocation reflects a genuine concentration limit, the new mechanism must honor it.
What it can remove is a particular reason for leaving an acceptable borrower unfunded: the lender had the money, and accepted the loan’s risk, but had reserved the money for another market before learning where demand would appear.
Conclusion
Isolating lending markets was never supposed to mean preassigning every dollar to one of them in advance.
A lender can know precisely which risks she’ll take without knowing which collateral market will need her money next, and forcing her to predict the second has a cost you can write down: positive in a simple uncertain-demand model, even measured against the best fixed allocation she could have chosen. The general result is more modest but more durable: drop the advance reservation and the feasible allocation set only grows, as long as her actual risk and capital limits are held fixed.
So let a lending policy describe the risk a lender is willing to take, and let allocation happen when compatible demand actually arrives. The collateral risk stays bounded by her chosen constraints; the forecast of where demand will appear doesn’t have to be smuggled into the deposit decision anymore. Alice’s 100,000 USDC doesn’t need to guess its own future; it just needs a policy good enough to recognize the right borrower when he walks in.
Related implementation: Morpho Midnight supports multi-market offers backed by a shared budget, and the Morpho Public Allocator supports on-demand reallocation within configured limits. These address aspects of the allocation problem discussed here; the argument above concerns the market-design constraint rather than any single protocol.