Hyperliquid's Custom L1 Architecture: How HyperCore and HyperBFT Redefine On‑Chain Trading

In the evolution of decentralized finance (DeFi), perpetual futures trading has long faced a seemingly intractable trilemma: latency, liquidity, and cost. Traditional DEXs either suffer from high gas fees and slow execution, or rely on centralized components that sacrifice transparency. Hyperliquid offers a radically different answer—not by building an application on top of an existing blockchain, but by creating a purpose‑built Layer 1 from the ground up, designed exclusively for trading.

Why a “Trading‑Specific” L1?

General‑purpose blockchains (e.g., Ethereum, Solana) must handle a vast array of transactions—NFT mints, DAO votes, social protocols, DeFi lending, and more. This “one‑size‑fits‑all” design means the execution engine must accommodate every use case, making it impossible to achieve extreme optimization for any single one.

Hyperliquid’s core philosophy is: the order book cannot merely live inside smart contracts—the order book must be the blockchain itself. Only by embedding order‑book logic directly into the base L1, coupled with a specially optimized consensus mechanism, can we achieve hundreds of thousands of orders per second without sacrificing decentralization. This conviction led to the creation of Hyperliquid’s custom L1 architecture, composed of HyperBFT (consensus layer) and HyperCore (execution layer).

HyperBFT: A Consensus Engine Built for Trading

Consensus is the blockchain’s “referee”—it decides how validators agree on transaction order and block production. Instead of adopting off‑the‑shelf solutions like Tendermint, Hyperliquid developed HyperBFT from scratch—a custom Byzantine Fault Tolerant (BFT) consensus algorithm inspired by the HotStuff protocol.

Performance Leap

The performance gains from HyperBFT are orders of magnitude. The original Tendermint‑based consensus used by Hyperliquid topped out at roughly 20,000 orders per second. HyperBFT multiplies that number by 100×—the mainnet currently supports about 200,000 orders per second, with the consensus and network stacks theoretically scalable to 2 million orders per second.

On latency, HyperBFT is fine‑tuned for end‑to‑end responsiveness. For validators geographically close to users, the median latency from order placement to finality is 0.2 seconds, and the 99th percentile latency stays below 0.9 seconds. In practice, this means a trader’s click is confirmed almost instantly.

Technical Highlights

HyperBFT integrates several cutting‑edge optimizations:

Optimistic Execution: Allows transactions to be executed before finalisation, drastically reducing block times.

Optimistic Responsiveness: Consensus dynamically adapts to network conditions—once a quorum of validators is reached, blocks are produced without unnecessary delays.

Non‑Blocking Consensus: Processes transactions without waiting for prior blocks to be finalised, improving throughput.

On the security front, HyperBFT offers deterministic finality—once a transaction is included in a block, it is irrevocably confirmed, with no forks or reorgs. As a BFT‑style protocol, it tolerates up to 1/3 of validators being malicious or faulty, ensuring continuous network operation.

HyperCore: A Trading Engine That Eliminates “Noise”

If HyperBFT is the referee, HyperCore is the athlete on the field—Hyperliquid’s native high‑performance trading engine.

Extreme Focus by Design

Unlike general‑purpose execution environments, HyperCore is a specialised state machine whose sole purpose is to manage order books. Written in Rust, it handles only the most critical tasks: order matching, margin calculations, and position liquidations.

This “noise‑eliminating” design pays off significantly: HyperCore never waits for an NFT mint or a complex DAO vote to finish, enabling it to deliver centralised‑exchange‑level performance. It maintains a fully on‑chain Central Limit Order Book (CLOB) where every order placement, cancellation, modification, fill, and liquidation is recorded on‑chain.

Performance Metrics

HyperCore’s block time is below 100 milliseconds. In live operation, average throughput has grown from roughly 5,900 orders per second in Q2 2025 to about 23,700 orders per second in Q2 2026, with peak capacity reaching 200,000 orders per second. The platform records billions of transactions and operations daily.

Dual Execution Environments Under a Unified Consensus

One of the most elegant aspects of Hyperliquid’s architecture is that HyperCore and HyperEVM share the same HyperBFT consensus and run on the same chain.

HyperCore handles native trading (spot and perpetual order books, margin, liquidations).

HyperEVM provides an EVM‑compatible environment for general‑purpose smart contracts.

Both share the same state root and consensus security, but their execution is isolated—DeFi applications on HyperEVM do not slow down HyperCore’s matching engine. This design allows Hyperliquid to maintain extreme trading performance while fostering a developer ecosystem through HyperEVM, achieving