Celestia Reveals the Engineering Behind Voyager, Its 21M TPS Payments Chain Design
Key Takeaways
- •Voyager's first full benchmark sustained an average of 21.2 million transactions per second, peaking at 21.6 million, across 100 million accounts over five hours.
- •The design runs all balance updates on a single CPU core with roughly 48 GB of node memory, staging data up to 32 transactions ahead to mitigate memory-fetch delays.
- •The benchmark was conducted on a single node and excluded state root computation, cross-machine replication, and large-scale real-time state queries, so it reflects isolated testing rather than live network throughput.
- •Voyager is not yet live, as Celestia confirmed the execution layer remains in active development and has not been deployed on mainnet.
- •Each Voyager transaction is published through Celestia's Fibre data availability protocol, which is already on mainnet, enabling outside parties to replay and verify payments as part of infrastructure aimed at agentic commerce.

Celestia ($TIA) has published the engineering details behind Voyager, its design for a high-speed payments chain. According to the company, the architecture was benchmarked at 21 million transactions per second (TPS) in testing, and the same design will later be opened to exchanges and margin trading.
One Core, No Contention
Many blockchains pursue speed by spreading work across multiple processor cores. Voyager takes a different route: it keeps every balance update on a single core, so no two tasks compete for the same data. That makes the headline number an engineering data point for a serialized design: it measures what one core can sustain when data is staged ahead of execution, before any of the multi-core paths Celestia describes come into play.
According to Celestia's engineering blog, Voyager's first full benchmark sustained 21 million transactions per second across 100 million accounts over five hours. Throughput averaged 21.2 million TPS and peaked at 21.6 million. ChainCatcher reported that execution ran on a single CPU core, with node memory of around 48 GB.
Data Ready Before Execution
The main bottleneck is fetching data from memory. Voyager addresses this by looking up to 32 transactions ahead and loading the data each one needs before it runs. Saving and publishing the data run as separate stages that do not block each other.
ChainCatcher also reported further design details. Celestia says Voyager splits payment processing into independently scalable threads, so the thread that updates balances is not blocked. Payment logic is built in rather than run through a general-purpose virtual machine, according to KuCoin News. State is stored in a flat table, and each transaction is published to Celestia's data availability protocol, Fibre, allowing outsiders to replay it and verify that payments were executed correctly.
Limits of the Test and What Comes Next
The benchmark figure comes from a controlled environment. 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. Phemex noted that the figure reflects isolated testing rather than live network throughput. Those exclusions also frame what to watch as development continues: whether Celestia reports figures that include state root computation, cross-machine replication, and large-scale state queries once Voyager moves from a single-node test toward a deployed network.
Voyager is also not yet live. Celestia confirmed that the execution layer remains in active development and has not been deployed on its mainnet, and KuCoin News quotes the company as saying it is "taking Voyager towards production." Celestia has additionally said the design extends to multiple cores and multiple Voyager instances, and that sharding storage across more disks would lift the current storage limit.
Payments for Agentic Commerce
Celestia positions Voyager as payment infrastructure for agentic commerce, where automated agents pay for content, data, and services on demand. Fibre, the data publication layer that Voyager relies on, has already launched on mainnet, and Voyager is the first execution-layer product following Celestia's acquisition of Sovereign Labs. The sequencing shows the layered approach: Celestia shipped the data publication layer first, and is now building payment execution on top of it, with every Voyager transaction published back through Fibre.
Sources
- Celestia Blog: Introducing Voyager, Hyperscale Payments for Agentic Commerce
- ChainCatcher: Celestia Launched the Payment Architecture Voyager
- KuCoin News: Celestia Unveils Voyager for AI Payments