Introducing the ZK Stack's Atlas upgrade: beyond 15K TPS with one-second ZK finality.

Today, the ZKsync team is excited to share details of the Atlas upgrade for the ZK Stack, which provides a major step forward towards the vision of a network of sovereign chains, secured by cryptography and powering the global economy.

This release delivers a high-performance, low-latency sequencer, support for multiple VM configurations (including full EVM equivalence), and integration with Airbender, the world’s most performant open source RISC-V proof system, for one-second finality with ZK proofs.

Global markets are moving onchain

Increasingly, businesses, institutions, and nations are embracing blockchains as a significant component of their financial infrastructure. Just this year we have seen:

  • Regulatory clarity is unlocking payment use cases. With clearer rulesets arriving in major markets (such as the GENIUS Act in the US) and improved banking connectivity, we’re seeing significant momentum for payments and stablecoins.

  • Major growth in onchain RWAs. rwa.xyz is tracking >$30B in tokenized real-world assets on public chains (excluding stablecoins), with sustained month-over-month growth across issuers and holders. Tradable has facilitated 38 private credit deals on ZKsync representing an active loan book of $2.1B.

  • Tokenized funds at scale. BlackRock’s tokenized U.S. Treasury fund (BUIDL) crossed $1B AUM on Ethereum in March 2025.

  • Institutional private chains on Ethereum. Deutsche Bank is working with Memento on tokenized asset infrastructure - an example of regulated, privacy-preserving domains that still anchor to public liquidity.

  • Mainstream fintechs are moving onchain. Robinhood and Stripe have both recently announced plans to launch their own blockchains as they look to add support for tokenization and stablecoins

We believe this trajectory will continue, and that we’re heading towards a world in which business and institutions routinely participate in the digital asset economy via systems they control, deployed as part of their internal infrastructure. These systems will be configured differently to support specific use-cases, but will still need to integrate across a single network, able to exchange value with cryptographic guarantees of correctness.

ZKsync, and the ZK Stack, are designed for this future. ZKsync connects public and private chains into one cryptographically secure network, made incorruptible by ZK, and connected to the billions in onchain liquidity via Ethereum. The ZK Stack is a toolkit for deploying these high performance chains, providing builders with control over options related to privacy, economics, and access.


The ZK Stack’s Atlas upgrade

The Atlas upgrade going live is the culmination of several major projects by the ZKsync team.

A high-performance sequencer, capable of >15K TPS

Firstly, we are introducing a new transaction sequencer, designed from the ground up around three dimensions: improving throughput (including resilience to traffic bursts), reducing latency (for rapid tx inclusion), and system simplicity.

Some of the most impactful decisions included:

  • We stripped all unnecessary functionality out of the sequencer, with a specific focus on minimizing synchronous persistence. The sequencer does not know about batches or L1, and is supported instead by separate asynchronous components.

  • We enforced strong separation between aspects of the system’s state. This included the state needed to execute the VM, the state required by the APIs, and the state required to generate the ZK proof or to verify it on L1, reducing unnecessary overhead on components.

  • We made it easy to replay blocks. The sequencer is fully idempotent, and other components have been optimized to reduce unnecessary computational work. This allows for a more resilient system with better developer experience.

The first results are already very exciting!

Before we get into the numbers, it is important to remember that transactions per second (TPS) is a flawed metric for profiling these systems; the computational resources required to execute a transaction vary depending on what it is doing. Gas/second has been proposed as a better metric, but for some systems this has a similar issue. The EVM’s gas prices are intended to directly price the computational work of Ethereum, but this strong coupling isn’t preserved when you also prove the transaction.

To help contextualize the sequencer capabilities we developed several demonstration scenarios, keeping in mind the limitations just described.

  1. High frequency price updates. First, we model the situation in which a team wants to bring high frequency price feeds onchain, deploying an oracle contract and testing with generated traffic.

    In this scenario the system was able to sustain ~23K TPS, corresponding to ~0.66 gigagas per second, and an average tx inclusion time of 250ms.

  2. Stablecoin transfers. Next, we model a payments focused system in which 50,000 simulated users are transacting between each other via ERC20 token transfers.

    In this scenario we saw a sustained performance of ~15K TPS, corresponding to ~0.5 gigagas per second, and an average tx inclusion time of 500ms.

  3. ETH transfers. Finally, we tested the system with native ETH transfers, again with a pool of simulated users. In the other experiments we used a blocktime of 100ms, but here tested raising this to 500ms.

    This is where we saw the most significant performance, with the system processing ~43K TPS, corresponding to ~0.9 gigagas per second, and an average tx inclusion time of 450ms.

We’re excited to share these early numbers, and while we have designed this sequencer for performance and are excited by the early results, it is nowhere near optimized and many more improvements will be made as the system matures.

Unleashing Airbender: one-second block proofs and $0.0001/transfer

With this upgrade, we’re delighted to see Airbender stepping into the spotlight. Airbender, introduced in an earlier blog post, is the world’s fastest open-source RISC-V zkVM.

As a reminder, it offers:

  • Fast proofs → fast finality: Airbender provides one-second block proofs and minutes-to-Ethereum finality.

  • General-purpose: Airbender can prove any program that compiles to RISC-V.

  • Best-in-class hardware utilization: Airbender is an incredibly efficient system, requiring a substantially smaller hardware footprint than competitor systems.

Efficient, near-real time ZK proofs are a breakthrough technology, enabling functionality that simply cannot be supported without it.

  • Applications such as onchain order books, perps, exchanges, and AMMs depend on fast finality to reduce risk. Airbender allows systems to verify and settle extremely quickly.

  • Public and private systems (such as ZKsync’s Prividiums) can interoperate without requiring third parties, as fast ZK proof generation allows for a network supporting near real time incorruptible connectivity. It will be possible to send assets of any size, anywhere in the network, in one second.

  • Tokenized securities and FX trades cannot wait minutes or hours to finalize. Airbender will shorten settlement cycles and allow teams to meet compliance-driven SLAs.

One-second ZK finality

Airbender provides a critical property: one-second ZK finality.

At the network edge, blocks from ZK Stack chains are proved in ~1s and verified by the ZKsync Gateway (the coordination layer). Counterparties get cryptographic assurance of execution almost immediately, and the proofs are then settled to Ethereum within minutes for full L1 finality.

This is a fundamental change to how systems can integrate. In a world of many chains, the naïve way to trust another chain is to re-execute it yourself, meaning running full nodes on often powerful hardware for every system you connect to. That is prohibitively expensive and incompatible with privacy (you can’t re-execute what you can’t see inside a Prividium).

Proofs, not intermediaries, carry trust across domains. With one-second proofs, you don’t re-execute, you verify. Anyone (chain operators, exchanges, even a user’s mobile device) can quickly verify a succinct proof and act with confidence. This is also how private chains can keep user data private while still composing with public liquidity, revealing only the ZK proof of correctness.

The combination of a sequencer providing low latency tx inclusion, Airbender generating a proof within a second, and then Gateway verifying and coordinating cross-chain messages, is how this architecture is able to realize the vision of a world transacting onchain.

Native EVM support via a more flexible system architecture

Additionally, the Atlas upgrade brings significant improvements to how ZK Stack chains will operate. We have redesigned the system-level component that is responsible for executing transactions and managing the state. This component is implemented as a Rust program and is compiled to both the x86 and RISC-V instruction sets. x86 is used for running the sequencer and RISC-V is fed to Airbender, and used to compute the ZK proof attesting to the correctness of the system.

This design has several compelling properties, which serve as the foundation for the entire upgrade:

  • One program, two targets. Compiling the same code to x86 for the sequencer and to RISC-V for proving means what you execute is what you prove, reducing the systemic risk and audit surface relative to previous proof system designs.

  • Multi-VM by design: any VM that compiles to RISC-V can be proven by the same pipeline, allowing for more optionality for chain builders. The EVM is first, and native support is already live. Support for other execution environments, such as the EraVM and WASM, can be added later.

  • Improved performance: proving an ERC20 transfer will cost ~$0.0001, with significant improvement in the throughput that can be supported by the sequencer.

Mainnets incoming

In our 2025 Roadmap, we committed to bytecode-level EVM equivalence and performance at industry scale. This upgrade advances those dimensions significantly, and further expands the toolkit available to ZK Stack builders. The first teams to be taking advantage of the Atlas upgrade already have testnets live, and we expect many other chains to follow in the coming weeks, on both testnet and mainnet.

High performance systems alone will not bring the world onchain because the endgame isn’t one big chain. It’s many sovereign systems, connected as part of a secure network. We expect banks, enterprises, public agencies, hospitals etc will all operate their own blockchain infrastructure (private where required) and participate an incorruptible network, anchored to Ethereum’s security and liquidity.

This vision requires both:

  • High-performance, reliable systems, for builders to develop the applications their users want.

  • Real-time ZK proofs so those systems can settle and interoperate without trusted intermediaries.

Alone, neither is sufficient.

With the Atlas upgrade, blocks are included quickly and proved quickly, so chains can finalize to Ethereum on minute timescales while keeping user-visible latency near real-time. Payments will clear with card-like UX but cryptographic finality. Price feeds will arrive at the speed finance applications need, and settle deterministically

This upgrade is another major step forward for the ZK Stack, which is ready today as a platform for teams looking to issue and settle real-world assets, run global payments with near-instant settlement, and tap into billions in onchain liquidity via Ethereum.

Find out more on the ZK Stack page.

Subscribe to ZKsync
Receive the latest updates directly to your inbox.
Mint this entry as an NFT to add it to your collection.
Verification
This entry has been permanently stored onchain and signed by its creator.