zkSync TPS: Real Throughput Numbers vs Marketing Claims
If you have spent any time reading rollup pitch decks, you have probably seen zkSync TPS quoted at numbers ranging from 2,000 to 100,000. Those figures are not lies, exactly, but they are not what you will experience when you swap a token at peak hours either. This guide separates the theoretical zkSync TPS ceiling from what onchain data actually shows, and why the gap matters when you are deciding which Layer 2 to trust with real money.
We are writing this because retail users keep getting sold “Visa-killer” throughput stories that fall apart the moment a popular mint goes live and gas spikes. You deserve the unvarnished version.
What “TPS” actually measures on a zk-rollup
Transactions per second sounds like a single number. It is not. On a zero-knowledge rollup like zkSync Era, throughput depends on:
- Sequencer capacity — how fast the centralized sequencer (currently operated by Matter Labs) can order transactions.
- Prover capacity — how fast validity proofs can be generated for each batch.
- L1 data availability — how much calldata or blob space Ethereum mainnet can absorb per block.
- Transaction complexity — a simple ETH transfer is cheap; a Uniswap swap with 3 hops is not.
A marketing TPS number usually measures only the first bottleneck, using the simplest possible transaction. Real applications hit all four limits at once.
What the data actually shows
According to L2Beat’s live activity tracker, zkSync Era’s 30-day average throughput has consistently sat in the 10-25 TPS range through 2024 and into 2025, with short bursts above 50 TPS during airdrop activity and meme cycles. Peak observed throughput has touched roughly 100-150 TPS for short windows.
Compare that to Matter Labs’ theoretical claim of “up to 100,000 TPS once full proof parallelization ships.” That figure assumes:
- Every transaction is a basic ETH transfer (~21k gas equivalent).
- The prover network has unlimited horizontal scaling.
- Ethereum L1 has unlimited blob throughput (it does not — EIP-4844 caps blobs per block).
Strip those assumptions and the realistic ceiling for the next 12-24 months is closer to a few hundred sustained TPS, not tens of thousands. That is still a meaningful improvement over Ethereum’s ~15 TPS L1 baseline, but it is not the science fiction number on the homepage.
Why this matters for presale buyers
A lot of new tokens launching in 2025-2026 are choosing zkSync Era specifically because the gas fees are cheap and the sequencer is fast. That is fine — until the token actually launches and 200,000 wallets all try to claim, swap, or bridge at the same minute. We have seen this pattern repeatedly on every L2:
- The sequencer queues build up.
- Effective TPS drops because complex contract calls are clogging the prover.
- Gas fees, while cheaper than L1, can spike 20-50x for a few hours.
- Bridges back to L1 slow because batch finalization is delayed.
If you are evaluating a presale that lives on zkSync Era, treat the chain’s TPS marketing the same way you treat the project’s roadmap: assume the realistic version is roughly 10-20% of the headline number. Our broader checklist for evaluating Layer 2 launches lives in our Ethereum L2 presales guide, which covers sequencer centralization risk, withdrawal delays, and bridge security on top of throughput.
How zkSync compares to other rollups
Quick reality check on average observed TPS as of early 2026, based on L2Beat data:
- Arbitrum One: 20-40 TPS sustained, peaks around 60.
- Optimism: 5-15 TPS sustained.
- Base: 30-80 TPS sustained, frequently the busiest L2.
- zkSync Era: 10-25 TPS sustained.
- Starknet: 3-10 TPS sustained.
The validity-proof rollups (zkSync, Starknet) generally show lower sustained TPS than the optimistic rollups (Arbitrum, Base) right now. This is not because the technology is worse — it is because proof generation is currently the bottleneck and decentralized provers are still being rolled out. That gap is expected to close, but “expected to” is doing a lot of work in that sentence.
The 7-day finality problem nobody mentions
Here is a detail the TPS conversation usually skips: zkSync’s validity proofs are fast in theory, but withdrawing assets from zkSync Era back to Ethereum L1 still takes around 24 hours under normal conditions, sometimes longer when prover capacity is constrained. Optimistic rollups take 7 days for trustless withdrawal. zkSync’s faster finality is real and it matters — but only if you use the native bridge, not a third-party fast bridge that adds its own counterparty risk.
If you are bridging large sums for a presale or claim, read our self-custody safety guide before you click anything. And if your wallet does not handle zkSync Era natively, check our wallet shortlist reviews — not every popular wallet has working zkSync support yet.
What we could not verify
A few things we tried to confirm and could not get straight answers on:
- Decentralized sequencer timeline: Matter Labs has discussed sequencer decentralization repeatedly but has not published a firm shipping date as of May 2026.
- Prover network economics: who pays provers, how rewards are distributed, and how this scales with TPS — the official documentation is light on specifics.
- Maximum sustained TPS under adversarial load: no public stress test data we could independently verify.
When a project’s own documentation does not commit to specific numbers, treat external “100,000 TPS” claims with extra skepticism.
Honest summary
zkSync TPS in real usage is roughly 10-25 transactions per second, occasionally peaking higher, with a realistic 12-24 month ceiling of a few hundred TPS once prover decentralization improves. The five-figure marketing numbers describe a theoretical limit under ideal conditions that no production workload has ever sustained. That does not make zkSync a bad chain — it is genuinely cheaper and faster than Ethereum L1 — but if you are buying a presale token because the chain is “infinitely scalable,” you are buying the slide deck, not the network.