DEX Perps: GMX GM Pools, Virtual AMMs, and Counterparty Risk

Every perpetual futures trade needs someone on the other side. On a centralized exchange that someone is another trader or a professional market maker, matched by an order book. On most decentralized perpetual exchanges it is a smart contract holding pooled capital, and the people who supplied that capital are the counterparty whether they think of themselves that way or not.

GMX's documentation states it without euphemism: "The GLV / GM pool is the counterparty to traders. If traders profit, that profit comes from the value of the GLV / GM pool."

That sentence is the whole article. Providing liquidity to a perpetual DEX is not a fee-farming activity with some price risk attached. It is underwriting a book of leveraged directional bets, compensated by fees. This piece works through the three liquidity architectures in use, quantifies the exposure an LP actually carries, and explains what the protocol's balancing mechanisms do and do not solve.

1. Three ways a decentralized perp finds a counterparty

The first three are the designs where someone is passively underwriting. The distinction that matters most is between a shared pool and an isolated one, because it determines whether you can choose your exposure.

2. The multi-asset pool and why GMX moved on from GLP

GMX's original design used GLP, a single index-like pool holding a basket of assets that backed every market on the exchange. Depositors minted GLP, received a share of trading fees, and served as the counterparty to all traders across all markets simultaneously.

The design solved a real problem. It let a perpetual exchange bootstrap deep liquidity without recruiting professional market makers, and it made passive liquidity provision accessible. It also had a structural flaw: a GLP holder could not choose their exposure. Underwriting a major asset's perpetual market meant also underwriting every other market in the basket and holding every asset in it, whether or not they wanted either.

GMX's current documentation no longer covers GLP. The provisioning liquidity pages describe GM pools and GLV vaults, and the V1 liquidity documentation no longer resolves. Treat GLP as a retired design that is important for understanding why the current architecture looks the way it does, not as a live product to allocate to.

3. Isolated GM pools and GLV vaults

GMX's current architecture breaks the single basket into per-market pools.

A GM pool is defined by three things: an index price feed, a long token, and a short token. In an ETH/USD market, WETH backs long positions and USDC backs short positions. Some markets support only swaps, while others enable leverage trading. Because each pool is market-specific, a depositor picks exactly which market they are willing to underwrite and which two assets they are willing to hold.

A GLV vault sits one layer above. It bundles multiple supported markets and shifts liquidity between them automatically based on utilization, with allocation recommendations from a risk advisor. It trades the precision of choosing a single market for the convenience of automated allocation across several.

On compensation, GMX's fee documentation states that of collected fees, 63% goes to liquidity providers and 37% to the protocol on Arbitrum and Avalanche.

What this architecture actually changes

It converts an unavoidable exposure into a chosen one. The GLP-era LP underwrote everything. A GM pool LP underwrites one market and holds two named tokens. That is a genuine improvement in risk control and it is the main reason the architecture changed.

It does not reduce the exposure within a chosen market. Isolation is about which risks you take, not how large they are. A concentrated ETH/USD GM position can carry more risk than a diversified basket did, because there is no offsetting market absorbing the loss when ETH traders are right.

Fragmentation is the cost. Splitting one deep pool into many shallow ones means each market has less depth, which raises price impact for traders and can make individual pools more sensitive to a single large position. GLV vaults exist to partly reclaim the pooling benefit.

4. What a perpetual LP is actually long

This is the section that changes how the position should be sized, and it is the one most LP dashboards do not make visible.

A worked example

Take an ETH/USD GM pool holding $5M of WETH and $5M of USDC, for $10M of total pool value. Open interest on the market is $3M long and $1M short.

The traders are net long by $2M. Because the pool is the counterparty to all of them, the pool is net short $2M of ETH exposure, on top of the $5M of WETH it holds outright.

Now move ETH up 20%:

What this example actually tells you

You hold two exposures, not one. The first is the pool's asset composition, which behaves like an index fund. The second is a short position equal to the traders' net open interest skew. LPs almost always notice the first and almost never size the second, yet the second is what determines whether the position beats simply holding the assets.

The break-even is a volume hurdle, and it is calculable. For the LP to beat holding the basket, fees must exceed the traders' net profit. At a 0.04% to 0.06% position fee with 63% flowing to LPs, roughly 0.03% of traded notional reaches the pool per side. Offsetting the $400,000 loss above therefore requires on the order of $1.3 billion in trading volume, before borrow fees and funding are counted. That is the number to compare against a market's actual throughput.

Skew is the variable that decides everything. With balanced open interest the pool is delta-flat against traders and collects fees cleanly. With a large one-sided skew the pool is running a substantial directional position it did not choose. Depositing into a heavily skewed pool immediately before a move in the skewed direction is the dominant way perpetual LPs lose money, and the skew is public information available before you deposit.

5. How the protocol pushes open interest back toward balance

Because skew is the LP's main risk, GMX applies four mechanisms that all point the same direction: making it cheaper to trade toward balance and more expensive to trade away from it.

What this toolkit actually tells you

These mechanisms price the LP's skew exposure. They do not remove it. Every one of them makes skew more expensive to create and more profitable to correct, which reduces how often and how far skew develops. None of them cap the loss if skew exists when a large move happens. The LP is being compensated for carrying skew risk, not insured against it.

Persistent skew despite rising costs is a warning, not an opportunity. Funding and borrow fees rise while an imbalance lasts. If traders remain heavily one-sided while paying steadily increasing costs to do so, that is evidence of conviction rather than inattention. The LP yield looks best at exactly the moment the LP's short-skew leg is most likely to be run over. Treat a high fee yield driven by extreme skew as compensation for a risk that is currently elevated.

Fee yield and profit and loss are different numbers. A dashboard showing an annualized fee yield is reporting only the income leg. The pool's realized return is that income minus trader net profit and loss and minus the pool's own asset price movement. Judge the position on total return against the alternative of holding the two tokens directly.

6. Virtual AMMs: pricing without a pool

A different design removes the pooled counterparty altogether. A virtual AMM, pioneered by Perpetual Protocol, borrows the constant product formula from spot AMMs but strips out the assets behind it.

Where a spot AMM holds real token reserves supplied by liquidity providers, a virtual AMM's reserves are accounting entries that exist only to generate a price. A market might be initialized with a virtual quantity of the base asset against a virtual quantity of the quote asset. Opening a long records a purchase against the virtual reserve, which moves the quoted price up along the curve. Traders post real collateral into a clearing vault, and profit and loss settles against that vault. Only the pricing layer is synthetic.

The appeal is that there are no liquidity providers to recruit and no impermanent loss to compensate, because there is no real pool. The weakness follows immediately from the same fact. With no pooled counterparty, a one-sided market has nobody structurally taking the other side, and the solvency gap falls on the protocol's insurance fund. A shared pool absorbs trader profits from capital that was explicitly committed for that purpose. A virtual AMM absorbs them from a fund that was not sized for it.

For an LP evaluating where to deploy, the practical implication is simple: a virtual AMM design offers no LP position at all. If a protocol using this model advertises a yield, look carefully at what is actually generating it, because it is not counterparty compensation from a pool.

7. Synthetix v3: modular collateral and credit delegation

Synthetix took a different path to the same problem of letting underwriters choose their exposure.

Synthetix v2x used a single shared debt pool. Every staker backed every market, and a staker's debt moved with the aggregate performance of all synths in the system. It was the same structural issue GLP had: no way to opt out of an exposure you did not want.

SIP-300 describes v3 as "a redesign of the Synthetix protocol built on top of novel smart contract architecture" featuring "multi-collateral staking, customizable debt positions, and permissionless synths." The design goal is stated directly: v3 "gives stakers flexibility to select which Synthetix markets to underwrite while preserving fungibility of snxUSD across markets."

The mechanism is credit delegation through pools. Per SIP-302, "Pools accept collateral in vaults, enable the creation of collateralized debt positions, and delegate credit to markets." Pool owners configure two parameters: weights, which set the proportional allocation of credit across markets, and a maximum debt per dollar of collateral, which caps exposure. When a backed market generates losses, participating pools assume that debt in proportion to their delegated credit.

What this design actually tells you

Multi-collateral and market selection are the same idea applied to two dimensions. V3 lets an underwriter choose what they post as collateral and which markets that collateral backs. Both were fixed in v2x, and both were sources of unwanted exposure.

Credit delegation makes the loss path explicit. A pool's maximum debt per dollar of collateral is a stated ceiling, and losses distribute by delegated proportion. That is a more legible risk contract than a shared debt pool where a staker's obligation moved with system-wide performance they could not influence.

Choosing a pool is choosing a risk manager. Stakers do not pick markets directly. They deposit into pools whose owners set the weights and caps, with governance maintaining preferred and approved pool lists. The underwriting decision therefore includes an assessment of the pool operator's parameters, not only the markets themselves.

8. An underwriting checklist for perpetual LPs

9. Conclusion

Providing liquidity to a decentralized perpetual exchange is an underwriting business dressed as a yield product. The pool is the counterparty, trader profits come out of pool value, and the fee stream is the premium paid for accepting that role.

The architectural history is a single idea being refined. Multi-asset pools like GLP made passive underwriting possible but gave depositors no control over what they underwrote. Isolated GM pools and Synthetix v3's credit delegation both answer the same complaint from different directions: let the underwriter choose the exposure. Virtual AMMs sidestep the problem by having no underwriter at all, which relocates the solvency risk rather than resolving it.

For anyone actually deploying capital, the analysis reduces to three numbers that are all available before entry. The net open interest skew tells you the size of the directional position you are taking on. The volume hurdle tells you how much trading activity is required for fees to pay for a normal adverse move. And the return of simply holding the pool's two tokens tells you what you must beat for the position to have been worth taking.

For how perpetual DEX liquidity provision compares with other advanced on-chain strategies, see advanced DeFi yield and hedging strategies. For the trader's side of the same funding mechanics, see funding rates explained.

Frequently asked questions

Sources and further reading

Primary protocol documentation:

• GMX Docs — Providing liquidity
• GMX Docs — Fees
• Synthetix Improvement Proposals — SIP-300: Synthetix V3
• Synthetix Improvement Proposals — SIP-302: Pools (V3)

Related CoinBeaver articles:

• Advanced DeFi yield and hedging strategies
• Trading funding rates on margin with Boros
• Perpetual futures explained
• Funding rates explained
• Open interest explained
• Impermanent loss explained

This article is educational and is not financial advice. All pool sizes, open interest figures, and volume calculations above are illustrative examples rather than live market data. Protocol architectures, fee splits, supported markets, and available products change frequently, and retired designs remain widely described in secondary sources. Verify current products and parameters in each protocol's own documentation before providing liquidity.