It also quietly poses a question the whole ecosystem keeps dancing around: can you run a leverage market for long-tail tokens at all — and if so, who is on the other side of the trade?

I run a spot-margin protocol. We've thought about that question for a living. Here's where I've landed.

First, give perps their due

Perpetual futures won. For liquid majors, it isn't close. Derivatives run several times spot volume across the market — by some recent quarters close to ten to one — and on any token with a real perp, the large majority of leverage demand migrates to it. In our own look at Solana tokens that graduated to a Tier-1 perp listing, the share of leverage demand that moved to the perp clustered in the low-to-high 80s percent within a month — directional, but strikingly consistent. That's not an accident. Perps are capital-efficient, you don't have to source the underlying, shorting is clean, and funding keeps the price pinned. If you're trading SOL or BTC with leverage, you should probably be using a perp.

So this isn't a "perps are bad" argument. Perps are the right instrument for the assets they serve. The interesting question is what happens to everything else.

Who has to be short your long?

Start with a structural fact people gloss over. A perpetual is a synthetic contract: a long position doesn't exist until someone takes the matching short. Open interest is minted in long/short pairs, and it's zero-sum — every dollar a long makes, a short loses. So a perp market requires a short for every long.

Ideally, that short is just another trader. On a liquid major, it usually is — there's deep, two-sided flow: hedgers, basis traders, people with a genuine bearish view. Longs and shorts roughly net out, and a market maker only has to fill the small residual imbalance. That's exactly why perps work so well on majors. The market maker's job is to warehouse a little bit of net exposure and immediately hedge it somewhere else.

The long tail breaks this in two places at once.

First, the flow is one-directional. Nobody buys a brand-new memecoin to short it. Everyone who shows up wants to be long; almost no one wants to carry a short and pay to hold it. So the long/short mismatch isn't a small residual — it's large and permanent. Someone has to warehouse the entire net-long imbalance by being short against all of it.

Funding rates are designed to fix this — when longs are crowded, funding turns positive and pays shorts to step in. But funding only summons a short who can actually run a business being short, and that means a short who can hedge. Which brings us to the second break: on a thin token there's nothing to hedge with. No deep spot book to lay the risk into, no second venue to offset it. So the residual short can't be neutralized by anyone — it just sits there, naked and directional, on the most volatile assets in the market. The job falls to a market maker or a pooled vault holding risk no rational desk wants.

We've seen how that ends. A memecoin position on Hyperliquid once grew too large to liquidate into a thin book, fell onto the protocol's own liquidity vault as the last-resort counterparty, and — after the attacker simply bought spot to ramp the price — left that vault with roughly a 29% drawdown in a month, ultimately resolved by validators manually overriding the price. That's not a risk-parameter you can tune your way out of. It's the structural consequence of forcing a hedge-dependent counterparty to hold un-hedgeable risk. It's also why the pooled-vault models are honest about their limits: Jupiter's liquidity pool, which works beautifully for majors, holds SOL, BTC, ETH and stables and no memecoins, by design. That exclusion isn't a gap waiting to be filled. It's a solvency constraint.

Now look at spot. When you buy a token on a DEX, you are not minting a synthetic short anywhere. The token already exists; you're simply taking ownership from whoever is selling. No one has to hold the opposite bet. Spot margin is just that, plus a loan — you're still buying the real token, you've only borrowed part of the money to do it. So the "other side" of a spot-margin trade decomposes into two roles, and neither one is short your position:

• a seller of the token — already there in any market with real liquidity, taking no directional bet against you; and

• a lender of the capital — a secured creditor who earns interest and genuinely doesn't care which way the token goes.

That's the structure. Spot margin never needs the person a perp can't find. It splits the perp's one impossible role — the residual short who must also absorb un-hedgeable risk — into two roles that already exist and are easy to fill.

The same fault line, seen from the price

There's a price-side view of this exact problem, and it's worth being precise about it, because "illiquid tokens have manipulable prices" is true of every instrument and explains nothing on its own.

The difference is what the price is used for. A perp settles against a mark. The protocol pays out gains computed against that price — in real terms — whether or not the price is actually executable. So if you can push the mark, you extract a phantom payout: you get paid a number that no real liquidity stands behind. That's precisely what the vault attack above was.

Spot margin never settles against a mark. Value is only ever realized by actually transacting — you bought a real token, and if the position has to be unwound, the collateral is sold into real liquidity for whatever it actually fetches. A manipulated price doesn't mint a claim on anyone; at worst it triggers a liquidation that clears against the real book, and the loss is bounded by the gap between the trigger and what the sale recovers. Same illiquidity, completely different blast radius: a phantom obligation on one side, a bounded gap on the other.

Why a lender can supply where a short can't

It comes down to the shape of the risk.

The residual short on a long-tail perp holds risk that is unbounded, two-sided, and un-hedgeable. They lose if the token rips and they lose if it craters, and they can't lay either side off.

A lender behind a collateralized spot-margin position holds something else entirely. They lend the quote asset; the trader buys the real token and posts margin as a first-loss buffer. The lender's return is the interest rate — path-independent, the same whether the token moons or dumps. Their only real exposure is gap risk: the chance the collateral falls faster than it can be liquidated. That risk is bounded, one-sided, over-collateralized, and — crucially — it can be priced. You set a rate that compensates for it. You don't need a hedging desk, a second venue, or a quoting engine. You need underwriting.

That is what makes a long-tail leverage market feasible in the first place. Market-making the long tail demands a professional with infrastructure most participants don't have. Funding it, as a secured loan, is a fundamentally more accessible activity — the risk is legible enough that a far broader base of capital can in principle supply it.

I want to be careful here, because it's the spine of the argument and it's easy to overstate. The point is structural: lending makes long-tail liquidity procurable from a broader supplier base than market-making does, because the risk doesn't require a hedge. Proving that breadth out in practice is its own journey — the early phase of any lending market is funded by a handful of sophisticated participants before it opens up. But the direction is the opposite of the perp model's. The perp model gets narrower as the asset gets thinner, because fewer and fewer desks can hedge it. The lending model doesn't.

It's not theoretical

We see this in our own book. Almost none of our volume is in SOL or other majors — traders who want leverage on those go to perps, exactly as they should. What flows through us is the long tail: a recent month spanned well over a hundred distinct tokens, the kind of names that will never clear the bar for a Tier-1 perp. The demand for long-tail leverage is real, persistent, and replenishes faster than tokens graduate out of it. The question was never whether people want it. It's whether anyone can supply the other side. Lending is the answer that scales into the thin part of the market.

The honest edges

A few things I won't pretend away.

Lenders eat the same illiquidity a market maker refuses in a correlated crash — a sharp, fast move in a thin token can widen the gap between the liquidation trigger and what the collateral actually clears for. The difference is geometry: that gap is bounded and one-sided, where the short's exposure is unbounded and two-sided. Better geometry, not zero risk.

And token incentives can paper over bad economics for a long time. A subsidized perp venue can run a money-losing long-tail market on emissions — memecoin perps already do. So I'm not claiming perps structurally cannot serve the long tail. I'm claiming the lending model serves it durably, without a subsidy treadmill, because the counterparty is solvent by construction rather than by emission.

Finally, the boundary moves. As on-chain price infrastructure improves, more tokens become liquid enough to support a real perp, and the perp-eligible universe grows. I expect that to keep happening. But the long tail refills far faster than it drains, and the thinnest, newest, most reflexive end of it — where a lot of the actual leverage demand lives — stays a lending problem for the foreseeable future.

The shape of the thing

The mental model I've settled on isn't "perps versus spot margin." The on-chain market Solana wants to become needs more than one mechanism — the same way every top centralized exchange runs spot, spot margin, and perps side by side, because different instruments serve different assets and different intents. Perps are for the liquid core. Lending-backed spot margin is how leverage reaches the long tail that perps can't price and can't hedge.

Anatoly's experiment is a great piece of engineering pointed squarely at the liquid core, and it'll make that core better. The long tail was never the same problem. It was never even the same kind of problem.