How Does Irys Store Data?

Irys stores data by distributing ten full replicas of each piece of data across 16 TB storage partitions, with each partition assigned to ten staked miners. Miners pre-fill their drives with cryptographic randomness keyed to their mining address (a technique called Matrix Packing), which encodes a unique replica per miner and lets data be written at disk speed. Once stored, miners produce continuous storage proofs through Useful Proof of Work, the storage-tied component of Irys's hybrid consensus. Failing a storage challenge results in slashing and reassignment of the partition.

Irys is a datachain, which means storage is incentivized at the consensus level. Miners earn rewards by storing data and proving they are still storing it, the same way miners on a generic blockchain earn rewards for block production. The mechanism below is what that incentive looks like in practice on Irys specifically.

How Irys storage differs from other datachains

Different datachains implement storage incentives in different ways. The shape of the mechanism shapes the tradeoffs.

Full replication versus erasure coding. Walrus uses erasure coding (Red Stuff), expanding each piece of data 4–5x into smaller fragments distributed across many nodes. Reconstruction requires assembling fragments from multiple nodes. Irys takes the opposite approach: ten full copies, each held by one staked miner. Recovery does not need fragment assembly; any single replica can serve the data immediately.

Pre-filled drives versus encode-on-write. Walrus encodes data into Red Stuff fragments when it is uploaded, which adds CPU and bookkeeping cost per write. Irys miners pre-fill their drives with cryptographic randomness derived from their mining address before any data arrives. When data is written, the miner only has to combine it with the pre-filled randomness, which happens at disk speed. The Irys whitepaper describes this as "Matrix packing encodes a miner's address into each 256 KiB chunk."

Deal-based versus consensus-incentivized. Filecoin runs a marketplace where storage providers and clients enter deals, with proofs (Proof-of-Replication, Proof-of-Spacetime) verifying the deal. Irys does not use a marketplace. Storage incentives are paid through block rewards and fees in a single chain-level fee market, with no per-deal coordination required.

Permanence-by-endowment versus term flexibility. Arweave uses a one-time payment funded into an endowment designed to pay for permanent storage. Irys supports both flexible-duration storage and permanent storage, with an endowment model designed to fund storage over a 200+ year horizon as storage costs decline.

The unifying property across all four is consensus-level storage incentives. The differences shape what each datachain is good at.

How data gets stored on Irys

The lifecycle of a data item from upload to retrieval has a few distinct stages.

Upload through the bundler. An application sends data to Irys through a bundler service, which groups multiple data items into a single bundle and submits it to the network as a data transaction. The bundle has a transaction ID (the onchain identifier of the data) and is broadcast to the network.

Initial validation in the Submit Ledger. The data first arrives in the Submit Ledger, where miners storing the data generate ingress proofs confirming they have received and stored their assigned replica. The Submit Ledger handles initial validation and the first round of replication.

Promotion to the Publish Ledger. Once promotion requirements are met, the data moves to the Publish Ledger, and its Merkle root becomes part of network state. From this point, the data is part of the long-term storage commitment of the chain.

Replication across partitions. Storage is organized into 16 TB partitions. Each partition is assigned to ten staked miners. Each miner holds a full replica of the partition, with the miner's address encoded into the storage representation through Matrix Packing. The result is ten unique, miner-specific copies of every piece of data in that partition.

Ongoing storage proofs. Once data is in the Publish Ledger and replicated across miners, the storage commitment is enforced continuously through Useful Proof of Work. Miners produce storage proofs derived from the data they hold, and these proofs are verified at the consensus level on every block.

Retrieval. Applications retrieve data through the Irys gateway, which serves stored data by its transaction ID. Bundling infrastructure supports sub-second retrieval after upload by distributing data across the network quickly, before all ten replica miners have finalized their copies.

How Irys verifies that data is being stored

The verification mechanism is the engine that holds the storage commitment together. The full description is in; the summary follows.

Storage proofs are not submitted periodically as separate transactions. They are produced continuously by miners as part of block production. Each miner reads from random locations in the partitions they hold and generates a cryptographic hash combining the read data with the current network randomness. If the hash meets the network's difficulty threshold, the miner has produced a valid block solution.

The economic property is "hard to produce, easy to verify." A miner has to be reading from genuinely held data to have any chance of producing a winning proof, and verifying a winning proof is cheap. The unsuccessful reads do not need to be recorded anywhere. The chain only stores the winning proof per block, which makes the storage proof system scalable: one proof per block, regardless of how much data the network holds in total.

Storage challenges run continuously. A miner that cannot produce proofs for the data they claim to hold is slashed, and the partition they failed is reassigned to another staked miner. The stake plus the partition pledge are at risk on every block, which makes the cost of pretending to store data higher than the cost of actually storing it.

The mechanism is called Useful Proof of Work because the work miners do (reading and proving stored data) is both the work that produces blocks and the work that proves the storage commitment. Traditional Proof of Work uses hash puzzles disconnected from network data: the puzzles secure the chain but do not also serve the storage layer. Useful Proof of Work redirects the work that secures the chain toward proving storage, so block production and storage verification are the same activity.

The economics of storage

Irys's three-part fee market is what lets the protocol price storage close to the underlying cost of storage hardware. Miners earn revenue from three sources:

1. Storage fees when data is written to Irys.
2. Execution fees when IrysVM smart contracts run.
3. Programmable data fees when smart contracts read stored data through the precompile.

Because miners earn from all three, the protocol does not need to price storage at a premium to keep miners profitable. Storage can be priced close to hardware cost, with miners earning the rest of their revenue from execution and programmable data usage.

Both storage and execution fees are paid in the network's native token, IRYS, in a single fee market. There is no separate token for storage and execution, and no per-deal coordination required.

For data that should be stored permanently, Irys uses an endowment model. The upload fee is funded into a storage endowment designed to pay miners for storage over a 200+ year horizon, with declining storage hardware costs extending the term as they fall. Permanent data on Irys is funded by the original upload payment alone, with no recurring fees required from the uploader.

Storage mechanism compared

What this means for applications storing data on Irys

The practical implications for a developer storing data on Irys:

Data is durable by protocol commitment. Once data is in the Publish Ledger, ten staked miners are economically committed to keeping it. A miner that loses or refuses to serve the data loses stake and pledge, and the partition is reassigned. There is no separate pinning service to pay or trust.

Retrieval is fast. Bundling infrastructure makes uploaded data available within seconds, before the full ten-miner replication has finalized. After replication, ten copies serve the same data, so retrieval scales with the network.

Onchain identifiers are stable. Each piece of data has a transaction ID that is part of network state after promotion to the Publish Ledger. Applications can reference data by this identifier and rely on it being resolvable.

Storage cost tracks underlying hardware. Because miners earn from execution and programmable data fees as well, storage on Irys can be priced close to the underlying cost of storage hardware, with the difference covered by the other fee markets.

FAQ

How many replicas does Irys store of each piece of data?

Ten full replicas. Each 16 TB partition in the network is assigned to ten staked miners, and each miner holds a complete copy of the partition encoded with their mining address through Matrix Packing. The strategy is full replication: any single replica can serve the data without needing to assemble fragments from other miners.

How does Irys verify that miners are actually storing the data?

Miners produce continuous storage proofs as part of block production. Each proof is derived from random reads of the data the miner holds, combined with current network randomness. Producing a valid proof requires actually reading from held data; verifying a proof is cheap. Miners that cannot produce proofs are slashed, and the partition they failed is reassigned to another staked miner. The system is called Useful Proof of Work because the work that produces blocks is the same work that proves storage.

Can data stored on Irys be lost?

Data on Irys is protected by ten independent staked replicas. A single miner failure triggers automatic partition reassignment before the storage commitment is at risk. Each miner has an economic stake plus a partition pledge at risk on every block, so the cost of failing to store the data is higher than the cost of storing it. The system is designed to keep data available through individual miner failures, network turnover, and adversarial attempts to break the storage commitment.

How fast is data retrieval from Irys?

Bundling infrastructure supports sub-second retrieval after upload by distributing data across the network quickly. After full replication, ten miners hold the data, so retrieval scales with the number of replicas. Applications retrieve data through the Irys gateway by transaction ID.

What is Matrix Packing?

Matrix Packing is the storage encoding scheme Irys miners use to prepare their drives. A miner pre-fills their partition with cryptographic randomness derived from their mining address before any data arrives. When data is written, the miner combines it with the pre-filled randomness, which happens at disk speed and produces a unique miner-specific replica. The whitepaper describes this as "Matrix packing encodes a miner's address into each 256 KiB chunk."

How Irys stores data, in one paragraph

Irys stores data by distributing ten full replicas of each piece of data across 16 TB storage partitions, with each partition assigned to ten staked miners. Miners pre-fill their drives with cryptographic randomness keyed to their mining address (Matrix Packing), which lets data be written at disk speed and encodes a unique replica per miner. Once stored, miners produce continuous storage proofs through Useful Proof of Work, the storage-tied component of Irys's hybrid consensus. Failed challenges result in slashing and partition reassignment. The result is consensus-level economic commitment to keep stored data available, with retrieval handled through the Irys gateway and pricing supported by a three-part fee market (storage, execution, programmable data) that lets storage be priced close to underlying hardware cost.

For implementation details, see the Irys docs and the Irys whitepaper.