What Is the Across Intent Bridge, and Why Does It Land in Seconds in 2026?

I run a fixed test: bridge 100 USDC from Optimism to Base on desktop, no accelerator, no funny tricks. In 2024 that took me 4 to 7 minutes. Across compressed it to 14 seconds. By 2026, Across is one of the default routes recommended by virtually every DEX aggregator (LiFi, Bungee, Socket). The reason it is so fast is not exotic cryptography. It is a deceptively simple design choice: pay first, settle later.

That structure has a name: intent-based bridge, and Across was the first protocol to turn the model into a usable product. This post pulls it apart.

What makes an intent bridge different

Lay the two bridge models side by side.

A traditional lock-and-mint bridge: you send funds into a contract on the source chain, a cross-chain message fires, the validator set signs the proof, and the destination chain mints the corresponding asset. Funds cannot land in your destination wallet until the message confirms. If you are on a 7-minute confirmation window like classic LayerZero, you wait 7 minutes.

An intent bridge inverts it. You sign an “intent” — “I want to convert this Optimism USDC into Base USDC and I will pay $0.04 in fees.” The intent itself is a signed EIP-712 order. A set of off-chain actors called relayers then immediately send you USDC on Base from their own balance, and later claim “I served this user, release the locked funds to me” on the source.

User experience: the money lands in the first second. Verification can be slow because it no longer touches the user.

How the relayer makes money

Relayers are what makes Across actually work. They front capital for instant settlement and need to recover the cost plus a risk premium. Walk through a standard Optimism → Base 100 USDC bridge:

• User pays 100.04 USDC
• Relayer immediately delivers 100 USDC on Base
• Relayer locks 100 USDC on Optimism
• After a few minutes the relayer goes through UMA verification and claims 100.04 USDC from the Optimism contract
• Relayer profit: 0.04 USDC, minus about $0.01 of Base gas, leaving roughly $0.03

That sounds thin. But Across processes tens of millions of dollars a day in 2026, so relayers run a thin-margin high-volume game. The most profitable relayers are not the ones with the highest per-trade margin; they are the ones with the best inventory rebalancing efficiency, holding balances on many chains and using statistical arbitrage to avoid paying gas to move funds around.

This is why Across is cheap on the popular L2 corridors (Optimism, Arbitrum, Base, Polygon, mainnet) and either expensive or has no relayer at all on long-tail chains.

How UMA does final verification

If the relayer pays first, how do you prevent them from being cheated, and how do you stop fake intents from being filed?

The answer is UMA’s Optimistic Oracle. UMA is Across’s parent project; Across was not chosen from a menu of oracles, it is the largest deployment of UMA built by the UMA team. The flow:

1. Relayer delivers on the destination and submits a claim on the source contract
2. The claim enters a challenge window (currently 1 hour by default)
3. Anyone can dispute it during the window by posting a 5,000 USDC bond
4. If no dispute, funds release automatically; if disputed, UMA’s human voting arbitrates

The challenge window does not affect the user (the user already has their funds). But it locks up the relayer’s capital for an hour, which is priced into their quote. An hour of locked capital is roughly 5% annualized capital cost, which on a $100 bridge is about $0.0006, essentially negligible.

The hidden cost of rebalancing inventory

Across runs smoothly because of an invisible layer: inventory rebalancing.

Suppose everyone bridges Optimism to Base. The relayer’s Base inventory drains while Optimism inventory piles up. Eventually they run out and the system stalls. So relayers periodically ship USDC back through the official bridge (slow but cheap).

That step costs time (7-day exit), capital (a week of lockup), and gas. Real profit = bridge fee − UMA window cost − rebalancing cost. That is why Across has a floor: about $0.018 on a $100 bridge.

How it compares to other intent bridges

Intent bridges multiplied after 2024 and Across is not the only one. The 2026 main contenders:

Across is strongest on stability between mainstream L2s, Mayan wins on Solana, deBridge wins on size. If you only bridge between EVM L2s, Across is the default.

For a systematic intro to bridge models, start with the cross-chain bridge guide and the Layer 2 introduction; the intent bridge optimization is much clearer once you understand the baseline lock-and-mint flow.

What the ACX token does

Across has its own token ACX, with a narrower role than people assume. ACX is not relayer collateral (that uses UMA); it is not a gas token. ACX is mainly:

• Governance for the Across DAO
• LP incentives for USDC, ETH, WBTC, DAI pools
• “Across+” rewards (launched 2025) where active users complete quests for ACX

ACX market cap sits around $140M with high circulating ratio (~65%) and inflation mostly released. It is not a fee-share token. Holding it is a bet on the product, not a claim on protocol revenue.

What the experience numbers look like in 2026

My bridge logs from the past month:

• Between mainstream L2s: P50 6 seconds, P95 22 seconds
• $100 bridge average fee: $0.038
• $1,000 bridge average fee: $0.21
• Failure rate below 0.1% — no failures in 30 days
• Relayer quotes occasionally spike when destination gas surges; Stargate Hydra is cheaper in those moments

This is why Across quietly became the default in 2026. Not the coolest bridge (no ZK, no MPC), but it perfected the part the user notices: no waiting.

My take on the intent bridge model

The intent-bridge paradigm is still young. The biggest concern is economic: relayers are concentrated (about 40 active, top 5 take 70%). If the majors pause, the system stalls. The other concern is UMA — we have not yet seen a coordinated attack stress the 1-hour window. My discipline: single bridge size capped at $50k; larger amounts go through CCIP or the canonical bridge.

What intent bridges removed from the user is the wait. They did not make cross-chain safer; they made it feel no different from a same-chain swap.

Knowing where this model shines, where it stalls, and when to avoid it matters far more than memorizing relayer rankings.