Stablecoin settlement at million-transaction-per-day scale is no longer theoretical, but almost nobody in the chain publishes the number you want. Chains document block or slot times: Solana Foundation puts Solana at a 300ms mainnet slot time, and the Circle CCTP docs put a Fast Transfer at roughly 8 to 20 seconds. Settlement providers publish neither sustained throughput nor error rates. Throughput is still the battleground, and the answer depends on three things at once: which chain you settle on, how your provider batches, and how it recovers from errors.
​

This benchmark walks through the real numbers for six leading providers (Circle CCTP, Bridge.xyz, BVNK, Coinbase Institutional, Anchorage Digital, Fireblocks), the block and finality figures each chain documents, and the batching and retry behavior that determines whether a 1M-transaction day actually clears.
​

Stablecoin throughput is the sustained rate of successful settlements (in transactions per second, or TPS) a provider can deliver end-to-end, including signing, broadcast, inclusion, and finality. It differs from raw chain TPS because providers add batching, queueing, and compliance checks. A useful figure combines effective TPS with a 99th-percentile latency target.
​

Three layers stack to produce the observable number:

Chain throughput sets the upper bound, and the figure each chain actually documents is its block or slot time, not a sustained TPS number. ethereum.org gives Ethereum 12-second slots, Arbitrum Nitro docs give Arbitrum One a 250ms default block time, Base transaction-ordering spec describes 2-second blocks with 200ms Flashblock preconfirmations, and Solana Foundation has Solana at 300ms. L2 rollups sit at the fast end because they batch calldata back to Ethereum.

Two of these numbers moved recently and are worth checking against anything you read elsewhere. Solana's slot time went to 350ms on August 19, 2026 and to 300ms on August 25, 2026, per Solana Foundation, so the 400ms figure that still circulates is stale. Polygon's finality changed too: Polygon Heimdall-v2 docs describe milestones delivering deterministic finality within 2 to 5 seconds, replacing the old multi-minute checkpoint wait.
​

Provider throughput is not published in TPS by any of the six providers below. Rather than infer a number from volume figures and batch sizes, the table records only what each provider states on a page that can be opened and checked.

Every figure above comes from a page published by the provider itself, checked in September 2026. Where a cell says nothing is published, that is the finding: the provider does not state it. Internal ledger settlements (Coinbase and Anchorage's omnibus model) are bounded by the provider's own systems rather than by a chain, because they never hit chain. Onchain settlement is what matters when a counterparty demands proof.
​

Batching is the single biggest lever. The three dominant patterns are real-time per-transaction, micro-batched (sub-second windows), and daily-window net settlement. Each has different throughput and error semantics.

The trade is universal. More batching equals higher headline throughput and lower fees, but slower individual confirmation and bigger blast radius if a single batch fails.
​

At 1M transactions per day, even a 0.1% failure rate produces 1,000 errors. Provider retry logic determines whether those become support tickets or self-heal. Three failure modes dominate: nonce collisions, attestation timeouts, and destination-chain reorgs.

Retry behavior also varies. Exponential backoff with a bounded attempt count is the common pattern, and MPC providers can route a retry through an alternate signer. The specific attempt counts are integration details rather than published policy. Coinbase Institutional uses internal ledger reservations so client-facing balances never reflect a failed onchain leg.
​

One million daily transactions is 1,000,000 divided by 86,400 seconds, or about 11.6 per second averaged over 24 hours. Real traffic is not flat, so the system has to absorb a multiple of that during peak hours without queue backup. Size the peak from your own traffic history, not from a rule of thumb. Three patterns achieve this in production:

Orchestration platforms like Eco's intent-based routing, LI.FI, and Squid coordinate across providers and chains so a single payment intent can land on whichever chain has capacity at that millisecond. This is increasingly the design pattern for treasuries that refuse to be locked to a single rail. See how bridging-based routes fail and what reduces error rates for the failure-mode deep dive.
​

Enterprise contracts increasingly specify TPS floors, p99 latency ceilings, and uptime credits. None of these providers publishes its SLA terms on its public site, so the numbers land in a contract rather than a docs page. Ask for the signing-latency commitment, the settlement-window definition, and whether uptime covers onchain inclusion or only the provider's own systems. The contract language differs by provider, and so does the financial remedy. What enterprise settlement providers actually promise in their SLAs is the companion read.
​

For treasury desks moving size, throughput is also an execution-quality question. RFQ flows (request-for-quote) handle large blocks via a small number of high-touch transactions, while OTC desk integrations batch differently. Stablecoin OTC execution vs RFQ compares the two for treasury operators.
​

Every chain figure in this piece comes from that chain's own documentation, checked in September 2026 and linked inline. Provider rows record only what the provider itself publishes; where nothing is published, the table says so rather than estimating. No throughput figure here is a measurement taken by Eco, and no provider TPS estimate has been carried over from earlier versions of this page.
​

Sources: Circle CCTP docs, ethereum.org on blocks and on proof-of-stake finality, Arbitrum Nitro docs, Base transaction-ordering spec, Solana Foundation on reduced slot times, Polygon Heimdall-v2 docs, TRON developer docs, Stripe newsroom, and BVNK's Series B post.
​