Non-Custodial Order Books: How Hyperliquid, dYdX and Lighter Differ
Three credible non-custodial order-book exchanges run on three different architectures. Here is how their performance, security and decentralisation trade off.
The case that 'centralised exchanges always have better UX and decentralised exchanges always have weaker UX' was decisively broken in 2024-2025 by a wave of high-performance non-custodial order-book venues. Three of them — Hyperliquid, dYdX v4, and Lighter — reached meaningful volume on three different architectural bets. Comparing them tells you a lot about where on-chain perpetuals are heading.
Hyperliquid
- Architecture: custom L1 with HyperBFT consensus — built specifically for an on-chain order book
- Block time: ~70ms — the order book matches and updates faster than most CEXs
- Throughput: 100k+ orders per second sustained, sub-second user-perceived latency
- Self-custody model: user funds live in a margin account on the L1; user signs orders with their key
- Fee model: zero gas for order placement (paid by the protocol from trading fees), maker-rebate trading fees
- Liquidity: as of 2026 the largest non-custodial perps venue by volume — frequently exceeds $10B daily
dYdX v4
- Architecture: Cosmos SDK app-chain with a CometBFT validator set
- Block time: ~1 second — slower than Hyperliquid but still well below Ethereum
- Order matching: validator-side off-chain matching with on-chain settlement — orders are not posted to the chain individually
- Self-custody model: user signs orders with a session key derived from their main wallet
- Decentralisation: full validator set; governance via DYDX token
- Trade-off: lower order-book latency than Hyperliquid in exchange for broader validator decentralisation
Lighter
- Architecture: zk-rollup-based order book — orders settle on Ethereum L1 with succinct zk proofs
- Block time: variable — depends on proof generation cadence
- Decentralisation: inherits Ethereum L1 security via the zk proof verification
- Strength: highest end-of-day verifiability, no validator-set trust assumption
- Trade-off: higher latency than Hyperliquid for orders to be irrevocably settled
Choosing Between Them
- Maximum performance, deepest liquidity, lowest latency: Hyperliquid
- Cosmos-native ecosystem, governance participation, mature validator set: dYdX v4
- Maximum verifiability, Ethereum-rooted security guarantees: Lighter
- Stable mature trading volumes for new entrants: Hyperliquid still leads in 2026
- Long-tail markets and niche perpetuals: increasingly fragmented across all three
Steyble Perps Routing
Steyble Perps routes user orders to whichever non-custodial venue has the best combination of price, depth, and latency for the requested instrument — typically Hyperliquid for majors, with Lighter and dYdX picked for specific instruments where they offer better liquidity. The user sees a single self-custodial perps surface that hides the venue selection complexity, which is the right separation between user experience and the underlying infrastructure layer.
What These Three Venues Share
- Self-custody throughout — user funds never leave a margin account they control
- Maker-rebate fee models that incentivise tight quoting — translates to better fill prices for takers
- Insurance funds well above 5% of open interest in normal regimes
- Open API access for algorithmic and AI-agent trading
- Oracle-based mark prices that resist single-venue manipulation
What to Watch Going Forward
Two trends will shape this category through 2027. First: order-book matching latency continues to compress, with sub-50ms becoming the new baseline; this favours venues with custom L1 architecture (Hyperliquid) over those running on shared chains. Second: regulatory scrutiny on perpetuals — particularly in the US — will likely force venue redomiciling, and the venues that adopted clean self-custody postures earliest will face the lightest friction. Watch the venue's geographic footprint and licensing posture as carefully as you watch its trading metrics.