The scaling mechanics of Celestia’s data availability
Celestia scales through data availability sampling where light nodes achieve 99% confidence levels using minimal hardware. The upcoming Matcha Upgrade increases block size to 128 MB to target 20+ MB/s performance for modular finance.
Light node sampling and hardware
Celestia scales through data availability sampling (DAS). These light nodes perform multiple rounds of random sampling for small portions of block data. Once a light node reaches a predetermined confidence level, such as 99%, it considers the block data available. This mechanism enables light nodes to contribute to network security and throughput using hardware significantly cheaper than full nodes. Full nodes must download the entire block to verify data availability. Light nodes do not have this capability and only download the block header, which includes the Merkle root of transactions, the hash of the previous block, the block height, and the timestamp.
Light nodes do not handle consensus or block production like full nodes or validators. They request transactions and query blocks using RPC endpoints. You should note that light nodes do not require the constant synchronization or security monitoring that full nodes demand. Celestia uses the Tendermint consensus algorithm, which is a set of rules that nodes use to produce new blocks and agree to finalize them. This consensus works within the framework of the Cosmos SDK, a development kit that provides a framework with which the business logic of new L1 blockchains can be created by separating software components into distinct modules.
The Mamaki Testnet requires specific software for node operation, including golang for source code builds. A user running a light node on an Ubuntu 20.04 LTS system requires the following minimum hardware:
| Component | Minimum Requirement |
|---|---|
| Memory | 2 GB RAM |
| CPU | Single Core |
| Disk | 5 GB SSD |
| Bandwidth | 56 Kbps |
Expanding throughput with Matcha
The 2025 Matcha Upgrade expands Celestia’s capacity for modular finance. This expansion targets high-frequency trading and massive volume requirements. The Matcha Upgrade increases block size by 16x to 128 MB, which provides a validated permissionless data availability layer throughput of 21.33 MB on testnets to support high-volume transaction activity for single-sequencer rollups in the modular ecosystem.
| Metric | Ethereum (Blobs) | Celestia |
|---|---|---|
| Throughput | ~0.38 MB | ~8 MB |
| Block Size | N/A | 128 MB (post-Matcha) |
| Relative Cost | High | 1-2 orders of magnitude lower |
Celestia’s data availability costs remain 1-2 orders of magnitude cheaper than Ethereum. Ethereum data costs for rollups historically range from 5-90% of operational expenses. Through its unique modular architecture, the Matcha Upgrade targets 20+ MB/s of performance to enable exchange-grade settlement.
Market growth and consensus assumptions
Celestia’s data posted KPI grew 2,559% over the last six months. Eclipse, an Ethereum rollup running the Solana Virtual Machine, provides more than half of the data currently posted to Celestia. This surge shows the ability of Celestia to facilitate interaction between different virtual machines. NOWNodes provides reliable Celestia RPC nodes and APIs, including BlockBook, Archive, and WebSocket access, to help developers avoid the burden of maintaining their own servers and infrastructure while maintaining constant access to the blockchain. These services provide 99.95% uptime and access to real-time WebSocket streams for new blocks, transactions, and events, while providing clear pricing and shared infrastructure with no hidden limits. NOWNodes allows developers to access full blockchain history on supported networks and query historical states and logs beyond pruned node limits.
Light nodes make an honest majority assumption about state validity because they do not verify transactions. If a majority of validators act dishonestly, they can create invalid blocks that light clients cannot identify. This creates a distinct trust assumption for cross-domain communication. How will the network handle increasing validator centralization?
Join the discussion