BNB Chain Is Building a New Network for High-Frequency Trading and AI Agents

BNB Chain, the blockchain platform tied to Binance, is building a separate layer-1 network for high-frequency trading and autonomous AI agents. The team wants to process over 100,000 transactions per second by sending them straight through, without a public order queue.

The team plans to launch a public testnet by the end of 2026 and the mainnet in early 2027. The new chain will run alongside the existing BNB Smart Chain, which already holds about $5 billion in locked value, rather than replacing it. The idea is to give the team room to experiment with an architecture built specifically for fast trading, without risking the stability of the chain that's already live.

Exchange-grade speed without giving up custody

The project is meant to give self-custodied traders execution speeds close to centralized exchanges. That gap is a big reason active traders keep moving funds onto CEXs in the first place: onchain execution simply loses on speed and predictability. Developers are removing the public mempool and streaming transactions straight to the block leader. That approach lowers latency and makes front-running harder, since pending orders are no longer visible in an open queue before execution, cutting off the data MEV bots typically rely on.

That streaming model has a tradeoff, though. The fewer participants that see a transaction before it executes, the more trust rests on how the block leader itself gets chosen. BNB Chain says it will detail that part of the design later, separately from the performance numbers. In the end, this isn't about replacing centralized exchanges, it's about making sure self-custody stops meaning slower order execution.

The target: 50 milliseconds and over 100,000 transactions per second

Those figures put the network closer to the parameters of a major exchange's central order book than to typical layer-1 throughput, where tens of thousands of transactions per second already count as a strong result. According to the team, the next phase is about execution rather than consensus: blockchains already sped up recordkeeping and storage, while the smart contract engine itself stayed the bottleneck.

In practice, that's a different engineering focus. Instead of speeding up the network of participants agreeing on block order, the team is optimizing the code that runs inside each block. That work looks closer to what exchange engineers do on matching engines than to typical blockchain protocol development. The largest centralized exchanges spent decades building infrastructure around millisecond-level demands, and that's the bar public blockchains are now chasing: the team is setting its target carefully, narrowing the gap just enough that self-custody stops being a tradeoff for speed.

BNB Smart Chain's progress in the first half of 2026

Even before announcing the separate network, BNB Chain had already improved the main chain's performance. Over the first half of 2026, block times fell from 750 to 450 milliseconds, while benchmark throughput rose from roughly 2,800 to about 5,200 transactions per second. Those changes touched a network that already carries protocols holding billions in locked assets, so the team had to validate each update against live traffic rather than test conditions alone. For everyday users, that mostly shows up as lower fees and fewer moments when a transaction gets stuck in a queue during a spike in activity.

"All smart contracts today need optimization directly at the execution layer. The chains got fast at consensus and storage, but the execution engine itself still works like it's translating sentence by sentence."

- David Z, chief technical officer at BNB Chain, in comments to CoinDesk

JIT compilation and reserved space for bridges and oracles

The speedup rests on execution-layer techniques. Just-in-time compilation compiles code as it runs rather than ahead of time, so repeated smart contract calls get faster with each subsequent execution. Each of these features used to require a separate third-party service bolted on top of the blockchain, so moving them into the protocol itself should make it easier to build applications specifically for AI agents. Strength reduction swaps expensive math operations for simpler equivalents wherever accuracy doesn't suffer. The network sets aside part of its throughput specifically for oracles, liquidations, and cross-chain bridges, so peak load from one service doesn't crowd out another. That matters most during sharp price swings, when liquidations and oracle updates need priority processing.

The roadmap adds a set of infrastructure features on top of that: native privacy, account abstraction, gas sponsorship, transaction batching, scheduled execution, and passkey signing.

• Account abstraction: wallets will be able to run more complex signing rules without a dedicated smart contract for each case.
• Gas sponsorship will let apps cover transaction fees on behalf of users, removing a barrier for new wallets that don't yet hold the network's native token.
• Transaction batching: several actions can settle in a single confirmation, cutting the combined fee for complex operations.
• Scheduled execution and passkey signing are built more for automated agents than for a person typing at a keyboard.

AI agents and stablecoins as fuel for the next trading wave

BNB Chain's plans line up with a broader industry push toward infrastructure for autonomous AI agents that can trade, pay, and execute transactions without constant human oversight. Last month Coinbase rolled out accounts built specifically for such agents, and stablecoins like USDT are increasingly used as a payment rail for autonomous software, since they don't depend on swings in a network's native token price. Accounts built for AI agents look like a logical extension of that trend: bots and algorithms already generate a meaningful share of onchain activity on popular networks, and dedicated infrastructure lets that traffic run separately from human transactions instead of competing with them for the same block space.

That changes the load profile infrastructure has to handle. Instead of sharp spikes during high-profile releases, the network needs to sustain a steady stream of small transactions from bots and algorithms running around the clock. That kind of traffic is harder to forecast in advance, so the team is building headroom for peak load rather than average load.

Quantum defenses and a cautious outlook

Separately, the BNB Chain team is researching quantum-resistant security. The goal is to let users adopt quantum-safe protection without changing wallet addresses or breaking existing applications, though this track is still at the research stage and the team hasn't set a concrete timeline for it. Roadmaps of this scale in crypto often slip by a quarter or more, so the testnet and mainnet dates are best read as a target rather than a hard deadline. Even so, a team running a network worth billions of dollars publicly committing to specific millisecond-level numbers already raises the bar for competitors.

Competition for traders with millisecond-level demands is already heating up. Ethereum has spent years wrestling with scalability on its own layer, and Robinhood launched its own equities blockchain this month. If BNB Chain sticks to the timeline, the first real test of the promised 100,000 transactions per second will happen on the testnet, long before the live traders and bots this network is actually being built for start using it.