Celestia has launched Voyager, a new payment chain built on a design that separates execution from data availability and uses a parallel state machine architecture. The network is aimed at providing native financial infrastructure for a high-frequency micropayment economy led by autonomous AI agents.
The piece was written by the Celestia Labs research team and translated by TechFlow. It argues that as artificial intelligence moves from generative chat toward autonomous agents, the internet’s main participants are shifting from humans to software agents capable of making decisions and acting on their own. Those agents may call APIs, buy compute, coordinate trades, and handle other tasks that require millisecond response times and very low-cost payments. Celestia says existing blockchain systems tend to run into congestion or rising fees when transaction demand reaches the millions per second.
Positioned as a payment rail for the agent internet
Celestia describes Voyager as a large-scale payment network built for the next phase of internet commerce. In laboratory benchmark tests, Voyager reached 21 million transactions per second. Based on the figures in the article, that level of throughput would support more than 1.8 trillion on-chain micropayment settlements in a single day.
The article frames this as part of a broader shift in how the internet operates. Instead of people sitting in front of screens and clicking through web pages, autonomous AI agents are expected to carry out tasks on behalf of users and institutions. The examples listed include planning travel, purchasing cloud computing resources, managing multisig treasury pools, and searching for arbitrage opportunities across decentralized markets.
Unlike occasional human interactions online, large numbers of agents working together can generate massive volumes of tiny requests. That creates a constant need for machine-to-machine micropayments. The article points to a scenario in which hundreds of millions of AI agents buy data fragments, pay for inference compute by token, or call paid search interfaces within milliseconds, putting pressure on both traditional financial rails and existing public blockchains.
Why traditional payment systems and general-purpose blockchains fall short
Celestia says traditional finance was built around human biological and business rhythms. Credit card settlement usually takes 2 to 3 business days to reach final clearing, while bank wires can take several days.
The article also highlights the economics of legacy payment gateways. It gives an example fee structure of $0.30 plus 2.9% per card transaction, arguing that this makes payments worth only a few cents economically impractical.
Celestia makes a similar case against most general-purpose Layer 1 and Layer 2 blockchains. In its view, they are constrained by single-threaded sequential execution models such as the Ethereum Virtual Machine, or EVM. When traffic spikes, gas fees rise sharply.
In the article, an ideal payment chain for agent-driven commerce needs three properties at the same time:
- near-unlimited horizontal scalability for throughput
- close-to-zero fees per transaction
- verifiable finality
How Voyager is built
Celestia says Voyager reaches eight-digit TPS by removing as much complexity as possible from execution. Traditional blockchains need to maintain large state trees because they are designed to support complex smart contract state and Turing-complete logic. In payment-heavy environments, the article says, most operations are simply additions to and deductions from account balances.
Voyager reduces execution to a lightweight state transition function and archives state on top of Celestia’s data availability, or DA, layer. The article describes this as a deep separation between a minimal transfer state machine and Celestia’s underlying DA layer.
At the consensus level, Voyager uses a parallel sharded state validation mechanism. Balance changes for different accounts are split across separate execution channels, which Celestia says removes lock conflicts associated with cross-contract calls.
The article also explains Celestia’s data availability sampling, or DAS, approach. With DAS, light clients do not need to download full blocks. Instead, they can download a proof sample of a few kilobytes and verify with high certainty that a payment has been safely confirmed by the network.
Benchmark results: 21M TPS and millisecond-level confirmation
To test the upper limit of the system under extreme conditions, the research team built a test network made up of hundreds of independent nodes running across globally distributed cloud server clusters.
According to the article, Voyager maintained a steady write rate of 21M TPS in that environment. End-to-end single-slot confirmation latency stayed at the millisecond level, and the network showed neither memory leaks nor fork delays.
Celestia converts that performance into daily settlement capacity of as many as 1.8 trillion real transactions on a single network. The article says that would be dozens of times larger than the combined processing capacity of all existing global credit card networks.
What this could enable for on-chain machine commerce
Celestia says high throughput is not just a headline metric. In the article, it is presented as a change to the economics of digital commerce. If the cost of a single transfer falls to one ten-millionth of a cent, business models that were previously hard to support become possible.
The examples listed include tipping writers by the number of characters read, paying for audio by the second, and charging automatically for GPU memory usage by the millisecond.
The article also says that a shared ledger built on open crypto protocols would allow AI agents trained by developers anywhere in the world to connect and trade from day one, without seeking permission and without opening bank accounts.
Next steps for the project
Celestia presents Voyager as a step toward application-specific scaling layers within its modular ecosystem. In the next development phase, the engineering team plans to focus on production deployment of a decentralized sequencer network and native integrations with major AI frameworks and agent orchestration tools.
The article closes by saying Voyager is intended to connect hundreds of millions of autonomous agents, and that the effort to build financial infrastructure for machine-driven commerce is only beginning.
