Celestia Reported 21M TPS for Voyager in a Single-Node Test

By Tom Whitfield

Celestia reported a benchmark throughput of 21 million transactions per second for its Voyager execution layer, according to the company’s “Introducing Voyager, Hyperscale Payments for Agentic Commerce” blog and coverage from ChainCatcher, KuCoin News and Phemex. Voyager remains under active development, and Celestia has not deployed it on its mainnet.

What does the 21 million TPS benchmark show?

Voyager’s design starts with what Celestia identifies as the main bottleneck: fetching data from memory. The execution layer looks ahead by as many as 32 transactions, then loads the data each transaction needs before that transaction runs. Moving the data fetch ahead of execution reduces the chance that processing will stall while waiting for required data.

Celestia also separates saving and publishing data into stages that do not block each other. The company says Voyager divides payment processing into independently scalable threads, allowing the thread that updates balances to proceed without blocking other payment work.

Payment logic runs inside Voyager rather than through a general-purpose virtual machine, KuCoin News reported. Celestia stores state in a flat table, and each transaction is published to Fibre, the company’s data availability protocol. Fibre makes the transaction available for replay, letting outsiders verify that the associated payment executed correctly.

Together, these choices create a specialized payment pipeline rather than a general-purpose chain environment. They also connect execution to a separate data publication layer, making verification possible outside the system that processes the transaction.

What did the controlled test leave out?

ChainCatcher reported that the test ran on a single node and did not include state root computation, cross-machine replication or real-time state queries at the same scale. The omissions matter because the 21 million TPS result does not represent the full set of functions required for a live, replicated network.

Phemex characterized the result as isolated testing rather than live network throughput. The number therefore describes Voyager’s performance under the controlled conditions specified by Celestia, not traffic handled by a deployed mainnet.

Celestia has said the design extends to multiple CPU cores and multiple Voyager instances. It also said sharding storage across more disks lifts the current storage limit. Those scaling paths address hardware and storage dimensions absent from the reported one-node configuration, although the source materials provide no revised throughput figure.

Why does Celestia need a dedicated payment layer?

Celestia positions Voyager as infrastructure for agentic commerce, where automated agents pay for content, data and services on demand. Its design is intended to let agents execute those payments while data saving and publication proceed independently. Separating those stages can prevent a data-retrieval or publication delay from stopping the full payment pipeline.

Fibre, the data publication layer Voyager relies on, has already launched on Celestia’s mainnet. Voyager is the first execution-layer product following Celestia’s acquisition of Sovereign Labs. The split deployment status is operationally important: Celestia has a live data layer for publishing transactions, but it does not yet have a live Voyager execution layer for processing them.

Is Voyager ready for mainnet?

Celestia confirmed that the execution layer remains in active development and has not been deployed on its mainnet. KuCoin News quoted the company as saying it is “taking Voyager towards production.” The cited materials do not provide a launch date, deployment timetable or mainnet throughput target.

That leaves the 21 million TPS figure as a controlled benchmark rather than evidence of current mainnet capacity. Businesses also cannot yet settle agentic-commerce payments through a live Voyager deployment, even though Fibre is operating as the underlying publication layer.

Without a disclosed launch window, the next verifiable milestone is a mainnet deployment with disclosed testing of state-root computation, cross-machine replication and real-time state queries at comparable scale. That live test, rather than another isolated TPS result, will show whether the design can support agentic-commerce payments in production.