Skip to content
MCS · Guides← all guides

Guide · any budget

Arc Mainnet: Circle's USDC-native L1, read as infrastructure

Public mainnet opened 16 September 2026. Gas is USDC, consensus is permissioned Malachite, execution is Reth, and the 10 billion ARC mint is not a token launch.

Job-to-be-done · Decide whether Circle's USDC-native L1 belongs on a builder stack, and what actually went live.

The job

You ship payments, agent workflows, or tokenised cash, and someone just asked whether Arc is "the chain to be on." The question underneath is narrower: what actually went live on 16 September 2026, what is still a slide, and which Ethereum assumptions break if you port a contract tomorrow.

This is not a GPU review. Arc has no proving ASIC, no DePIN fleet, and no miner-to-HPC pivot. It belongs on this site because Circle just stood up a dollar-denominated settlement computer next to the rest of the Ledger, and the silicon question is honest: there isn't one. The interesting layer is software — Malachite for ordering, Reth for state — running on commodity validators under a permissioned set.

This guide is not for you if you want a price, a pick, or a native-token farm. Gas is USDC. The 10 billion ARC mint this week is a genesis event Circle itself said is not a public launch.

The Ledger row is at /ledger/arc. The cash-engine reading lives on MadCoolMoney as The listed claim is CRCL. The rail is Arc..

What shipped

Circle Internet Group (NYSE: CRCL) opened Arc public mainnet on 16 September 2026. The pressroom post is the primary source, not a recap. Day-one claims that survive a read of that post:

  • An open Layer 1, EVM-compatible, with USDC as native gas and "more than" 74 billion USDC in circulation on Circle's network, not on Arc specifically.
  • More than 100 applications and more than 100 institutional and ecosystem builders named as live or exploring on private mainnet ahead of the public opening.
  • A founding validator cohort, alongside Circle, of 11 names: BlackRock, DTCC, Galaxy, ICE, Mastercard, MoneyGram, SBI Group, Standard Chartered, Sumitomo Corporation, Visa, and Worldpay (now Global Payments). Rollout is phased. Do not treat 11+1 as a measured live validator count.
  • Testnet processed more than 700 million transactions in under a year.
  • Post-quantum signatures are listed as supported today. Opt-in privacy is listed as in development for network-wide release, not as live.

Developer docs at docs.arc.io still described testnet as the public surface on launch morning. Treat the pressroom post and arc-node's named arc-mainnet chainspec as the launch evidence; treat the docs site as lagging.

Chain id 5042 is in the circlefin/arc-node changelog and in Arcscan's own network table. Arcscan measured a 506 ms mainnet block time on 10 August 2026, while the chain was already producing blocks in private mainnet. Finality is reported as instant: a committed block does not reorg.

DefiLlama's v2/chains row named Arc with chain id 5042 and about 308 million dollars of TVL during this session. That is a day-one snapshot, not product-market fit. Re-pull it before you cite it next week.

Two layers

Arc splits consensus from execution. Circle's system overview is the spec.

Malachite orders and finalizes. It is a Tendermint-family BFT engine, brought in when the Informal Systems team joined Circle in 2025. Two-phase voting (pre-vote, pre-commit). A block commits when more than two-thirds of the permissioned validator set pre-commits. Once committed, the block is final. Circle's own overview quotes a benchmark of 3,000+ TPS with 20 validators at under 350 ms finality. That is a benchmark with a stated validator count, not a sustained mainnet measurement. The Ledger does not plot it as tps_sustained.

Reth executes. Paradigm's Rust Ethereum client maintains accounts, balances, contracts and storage, then returns a state root for Malachite to finalize. Arc targets the Osaka EVM baseline, including EIP-7702 set-code transactions, and ships EIP-7708 native Transfer logs ahead of Ethereum Amsterdam.

The lifecycle is ordinary until it is not: JSON-RPC submit, mempool, proposer, pre-vote, pre-commit, commit, then EVM execution. Circle says the whole path completes in under one second. For a payments or agent flow, that is the load-bearing property: you can fire the webhook on inclusion.

The Fee Manager module is live. It pays fees in USDC and smooths the base fee with an EWMA of block utilization instead of EIP-1559's per-block step. Design target under normal load: about 0.001 dollars per ERC-20 transfer. Minimum base fee 20 Gwei; transactions below that are dropped from the mempool with no receipt. Base fee is credited to the block beneficiary, not burned. Arc Privacy Sector and Stablecoin Services are roadmap modules, not live.

USDC is the native asset, twice

On Ethereum, ETH is native and USDC is an ERC-20 with 6 decimals. On Arc, USDC is both.

Native view: 18 decimals, used for gas and msg.value. ERC-20 view: 6 decimals, used by applications. One balance. To display native units as USDC, divide by 10^12. Mixing msg.value with USDC.balanceOf() in pool math is off by that factor, and a liquidity pool that pairs native USDC against the ERC-20 interface is pairing an asset with itself.

A few rules that do not exist on mainnet Ethereum:

  • Value transfer to the zero address reverts (zero-value calls to 0x0 still succeed).
  • Burns are forbidden. A non-zero-value call to an account that already self-destructed in the same transaction reverts; on Ethereum that send succeeds.
  • The blocklist is enforced at runtime. An included transfer that hits it still consumes gas.
  • PREVRANDAO always returns 0. There is no onchain randomness. The beacon-roots contract Ethereum uses for the parent beacon root is omitted.
  • Blob (type-3) transactions are rejected. BLOBHASH returns 0.
  • Block timestamps are non-decreasing, not strictly increasing. Sub-second blocks may share a timestamp. Order on block number.

If you are porting a contract, start with Circle's EVM-differences page, not with "it's EVM-compatible." Compatibility here means Solidity and Foundry mostly work. It does not mean your ETH-gas mental model, your randomness, or your SELFDESTRUCT edge cases survive.

Foundry's stock anvil is a standard EVM. It will not reproduce Arc's native-USDC rules. Circle documents arc-anvil --network arc for local work.

What this is not

Arc is not a permissionless validator set. Circle's own disclaimer: launched by Arc Network Services LLC, operated by a permissioned set, not reviewed or approved by NYDFS or any other regulator. The trust model is "the institutions that clear the world's securities help produce the blocks," not "anyone can join." Proof of stake, and an ARC token as a coordination commodity with fees still paid in USDC, is a 2027 exploration. Circle said the 10 billion genesis mint is not a commitment to a public launch. A mint without transferable distribution is not a token launch under the Kalshi rules for KXTOKENLAUNCH-27JAN01-ARC either; that contract is a separate claim.

Arc is not a replacement for every chain USDC already lives on. Circle still runs CCTP and Gateway. The pitch is a home for dollar settlement, FX (StableFX), tokenised funds (USYC, BUIDL), and agent payments, not a TVL war with Ethereum.

Arc is not a silicon story. The Ledger's other three seams — ZK ASICs, DePIN GPUs, miner-to-HPC megawatts — do not apply. If a validator bill of materials is ever published, that is when this row earns a hardware note beyond "Reth + Malachite on commodity boxes."

Roadmap items to keep in the vapor column until they have a mainnet measurement: Network Sectors (Privacy / Payment targeting 100,000 TPS / Agent), the PoS transition, DTCC tokenization of DTC-custodied assets (second half of 2027, from the 5 August validator post), and network-wide confidential transactions.

Sources