Author: Ponyo
Compiled by: Sui Network
Key Summary
? Architecture: Irys is a fully functional integrated Layer 1 'data chain' that provides native blob access for contracts but requires a completely new set of validation nodes. Walrus builds an erasure-coded storage layer on Sui, which is easier to integrate but requires cross-layer coordination.
? Economic Model: Irys uses a single token IRYS to unify payment of fees and rewards, providing a simple user experience but with higher price volatility risk. Walrus separates functions into two tokens: WAL (for storage) and SUI (for gas), effectively isolating costs but requiring maintenance of two incentive systems.
? Durability and Computational Power: Irys maintains 10 full replicas and streams data directly into its virtual machine; Walrus uses a method of approximately 5-fold redundancy with erasure coding and hash verification, resulting in lower storage costs per GB, but a more complex protocol implementation.
? Adaptability: Irys offers a 'pay once, permanent storage' donation model, which is very suitable for preserving immutable data but comes with high upfront costs. Walrus adopts an 'on-demand payment, automatic renewal' leasing mechanism that facilitates cost control and can quickly integrate with Sui.
? Adoption: Walrus is still in its early stages but is developing rapidly, with PB-level storage, over 100 node operators, and has been adopted by several NFT and gaming brands. In contrast, Irys is still in the pre-expansion phase, with data volumes not reaching PB level, and its node network is still growing.
Both Walrus and Irys are committed to solving the same problem: providing reliable, incentivized on-chain data storage. However, their design philosophies are entirely different: Irys is a Layer 1 blockchain specifically built for data storage, integrating storage, execution, and consensus into a vertical architecture; while Walrus is a modular storage network, relying on Sui for coordination and settlement while operating an independent off-chain storage layer.
Although the Irys team initially portrayed it as a superior 'built-in' option and defined Walrus as a limited 'external' system, in reality, both have their pros and cons, with different trade-offs. This article objectively compares Walrus and Irys across six dimensions from a technical perspective, refuting one-sided assertions, and providing developers with a clear choice guide to help them determine the most suitable path based on cost, complexity, and development experience.
1. Protocol Architecture
1.1 Irys: Vertically Integrated L1
Irys embodies the classic concept of 'self-sufficiency.' It comes with a consensus mechanism, staking model, and execution virtual machine (IrysVM), all closely integrated with its storage subsystem.
Validation nodes simultaneously serve three roles:
Storing user data in full replicas;
Executing smart contract logic in IrysVM;
Protecting network security through a hybrid mechanism of PoW + staking.
Since these functions coexist within the same protocol, every layer from the block header to data retrieval rules can be optimized for large-volume data processing. Smart contracts can directly reference on-chain files, and proof of storage will follow the consensus path of sorting ordinary transactions. Its advantage lies in the high consistency of the architecture: developers only face a single trust boundary, a single fee asset (IRYS), and the experience of reading data in contract code feels native.
However, the cost is a high startup investment. A brand new layer network must recruit hardware operators from scratch, build indexers, launch block explorers, strengthen clients, and develop developer tools. In the early stages when validation nodes have not yet grown robustly, block time assurance and economic security lag behind established chains. Therefore, Irys's architecture opts for deeper data integration while sacrificing ecosystem launch speed.
1.2 Walrus: Modular Overlay Layer
Walrus takes a vastly different path. Its storage nodes operate off-chain, while Sui's high-throughput L1 is responsible for processing sorting, payments, and metadata through Move smart contracts. When a user uploads a blob (data block), Walrus shards and disperses it across nodes, then records an on-chain object on Sui that contains content hashes, shard allocation, and lease terms. Renewals, penalties, and rewards are executed as ordinary Sui transactions, with SUI paying for gas, but using WAL tokens as the settlement unit for storage economy.
Leveraging Sui, Walrus immediately gains the following advantages:
Verified Byzantine fault-tolerant consensus mechanism;
Robust development infrastructure;
Powerful programmability;
Liquidity-based token economy;
Many existing Move developers can directly integrate without protocol migration.
But the cost is the need for cross-layer coordination. Each lifecycle event (upload, renewal, deletion) must be coordinated between two semi-independent networks. Storage nodes must trust Sui's finality while maintaining performance even during Sui congestion; while Sui validation nodes do not verify whether actual disks are storing data, they must rely on Walrus's cryptographic proof system to ensure accountability. Compared to a unified design, this architecture inevitably incurs higher latency, and some fees (SUI gas) will flow to roles that do not actually store data.
1.3 Design Summary
Irys adopts a vertically integrated monolithic architecture, while Walrus is a horizontally layered modular solution. Irys has greater architectural freedom and a unified trust boundary but must overcome the challenges of ecosystem building due to cold starts. Walrus significantly lowers the entry barrier for developers within existing ecosystems by leveraging Sui's mature consensus system but must address the coordination complexity of two economic domains and operator systems. Neither model has absolute superiority; they simply have different optimization directions: one pursues coherence, and the other pursues composability.
When protocol choice relies on developer familiarity, ecosystem attractiveness, or speed to market, Walrus's layered model may be more realistic. When the bottleneck lies in deep data and computation coupling or requires custom consensus logic, Irys, designed specifically for data, has sufficient justification to bear a heavier architectural burden.
2. Token Economics and Incentive Mechanisms
2.1 Irys: A token drives the entire protocol stack
Irys's native token IRYS encompasses the economic model of the entire platform:
Storage fees: Users prepay IRYS to store data;
Execution gas: All smart contract calls are also priced in IRYS;
Miner rewards: Block subsidies, storage proofs, and transaction fees are all paid in IRYS.
Since miners are responsible for both data storage and contract execution, computation income can compensate for the inadequacy of storage earnings. Theoretically, when DeFi activities on Irys are booming, computation earnings can subsidize data storage, achieving services close to cost price; if contract traffic is low, then the subsidy mechanism adjusts inversely. This cross-subsidy mechanism helps balance miner rewards and aligns incentives across roles in the protocol. For developers, a unified asset means fewer custodial processes and a simplified user experience, especially suitable for scenarios where users do not wish to interact with multiple tokens.
However, the downside is the risk interdependence of a single asset: once the price of IRYS drops, rewards for computation and storage will also decrease simultaneously, putting miners under double pressure. Thus, the economic security of the protocol is tied to the same price volatility curve as data persistence.
2.2 Walrus: Dual-token Economic Model
Walrus splits functional responsibilities into two tokens:
$WAL: Economic unit of the storage layer. Users pay for rental space using WAL, and node operators earn WAL rewards by staking and storing data fragments, with rewards also linked to their delegated staking weight.
$SUI: Gas token used for on-chain coordination transactions. Any transactions on Sui for uploads, renewals, penalties, etc., require SUI and reward Sui validation nodes, not Walrus storage nodes.
This separation keeps the storage economy clear: the value of WAL is only influenced by data storage demand and lease periods, unaffected by DEX trading or NFT booms on Sui. At the same time, Walrus can also inherit Sui's liquidity, cross-chain bridges, and fiat entry points—most Sui builders already hold SUI, thus the marginal cost of introducing WAL is low.
But the dual-token model also has incentive fragmentation issues. Walrus nodes cannot participate in SUI fee income, so the price of WAL must be sufficient to independently support hardware, bandwidth, and return expectations. If the price of WAL stagnates while SUI gas skyrockets, user costs will rise, but storage providers will have no direct gains. Conversely, a DeFi explosion on Sui boosts verification node earnings but is unrelated to Walrus nodes. Therefore, to maintain long-term balance, active optimization of the economic model is required: storage prices need to flexibly fluctuate based on hardware costs, demand cycles, and WAL market depth.
2.3 Design Summary
In short, Irys offers a unified and simple user experience but bears concentrated risks; Walrus delineates boundaries at the token level, providing more refined economic accounting but needing to address two market systems and fee diversion issues. Builders should weigh whether they prefer a seamless experience or a preference for the separation of economic risks to align with their product planning and funding strategies.
3. Data Durability and Redundancy Strategy
3.1 Walrus: Achieving lightweight high reliability with erasure coding
Walrus splits each data block (blob) into k data shards and adds m redundant parity shards (using RedStuff encoding algorithm). This technique is similar to RAID or Reed-Solomon encoding but optimized for decentralized and highly variable node environments. Only k out of k + m shards are needed to reconstruct the original file, providing two advantages:
High space efficiency: Under typical parameters (approximately 5x scaling), the required storage space is reduced by half compared to traditional 10x replica copying schemes. In simple terms, storing 1GB of data on Walrus requires about 5GB of overall network capacity (sharded across multiple nodes), while a traditional full replica system might need 10GB to achieve similar security.
Strong on-demand repair capabilities: Walrus's encoding method not only saves space but also bandwidth. When a node goes offline, the network only reconstructs the missing shards rather than the entire file, significantly reducing bandwidth overhead. This self-healing mechanism only requires downloading data approximately equal to the size of the lost shard (i.e., O(blob_size/ shard_count)), whereas traditional replica systems typically need O(blob_size) of data.
Each shard and node allocation will exist in the form of an object on Sui. Walrus rotates the staking committee every epoch, challenges node availability through cryptographic proofs, and automatically re-encodes when node loss exceeds safety thresholds. This mechanism, while complex (involving two networks, multiple shards, and frequent validations), achieves maximum durability with minimal capacity.
3.2 Irys: Conservative but Robust Multi-Replica Mechanism
Irys intentionally chose a more primitive and direct durability method: every 16TB of data partition is fully stored by 10 staking miners, each holding a copy. The protocol introduces a specific miner's 'salt' (Matrix Packing technology) to prevent double counting on the same hard disk. The system continuously reads and verifies the hard disks of nodes through 'proof-of-useful-work' to ensure each byte truly exists, otherwise miners will be penalized and lose their staked assets.
In practice, data availability depends on: whether at least one of the 10 miners responds to the query? If a miner fails verification, the system will immediately initiate re-replication to maintain the standard of 10 copies. The cost of this strategy is a data storage redundancy of up to 10 times, but the logic is clear, and all states are concentrated on a single chain.
3.3 Design Summary
Walrus focuses on addressing the frequent replacement of nodes through efficient encoding strategies and Sui’s object model, thereby ensuring data durability without increasing costs. Irys believes that as hardware costs decrease rapidly, a more direct and heavier multi-replica mechanism is actually more reliable and worry-free in practical engineering.
If you need to store PB-level archival data and can accept a more complex protocol, Walrus's erasure coding has an economic advantage on a per-byte basis. However, if you value operational simplicity (one chain, one proof, ample redundancy) and believe hardware expenditure relative to product delivery speed is negligible, Irys's 10-replica mechanism can provide durability assurance with minimal thought.
4. Programmable Data and On-chain Computation
4.1 Irys: Natively supporting data with smart contracts
Since storage, consensus mechanism, and the Irys virtual machine (IrysVM) share the same ledger, contracts can easily call the read_blob(id, offset, length) method as if reading their own state. During block execution, miners stream the requested data segments directly into the virtual machine, perform deterministic checks, and continue processing the results in the same transaction. No oracles, no user parameter passing, no off-chain intermediaries are needed.
This programmable data structure can implement the following use cases:
Media NFTs: Putting metadata, high-resolution images, and royalty logic all on-chain, ensuring enforcement at the byte level.
On-chain AI: Performing inference tasks directly on model weights stored in partitions.
Big Data Analysis: Contracts can scan large datasets such as logs, genomic files, etc., without the need for external bridging.
Although gas costs increase with the number of bytes read, the user experience remains a transaction priced in IRYS.
4.2 Walrus: 'Verify before Compute' Model
Since Walrus cannot stream large files directly into the Move virtual machine, it adopts the design pattern of 'hash commitment + witness':
When users store blobs, Walrus records their content hashes on Sui;
Subsequently, any caller can submit the corresponding data shard along with a lightweight proof that the shard is correct (such as a Merkle path or complete hash);
Sui contracts will recalculate hashes and compare them against Walrus metadata. If verification succeeds, trust that data and execute subsequent logic.
Advantages:
Immediately usable, with no modifications needed for the L1 protocol;
Sui validation nodes do not need to be aware of GB-level big data content.
Limitations:
Data must be manually retrieved: callers must pull data from the Walrus gateway or nodes and package limited-length data segments in transactions (restricted by Sui's transaction size);
Sharding processing overhead: For large data processing tasks, multiple microtransactions or off-chain preprocessing + on-chain verification are required;
Dual gas costs: Users need to pay SUI gas (for transaction verification) and WAL (indirectly paying for underlying storage costs).
4.3 Design Summary
If your application requires processing several MB of data per block (such as on-chain AI, immersive media dApps, verifiable scientific computing processes, etc.), the embedded data API offered by Irys is more attractive.
If your scenario emphasizes data integrity proof, small media display, or heavy computation occurs off-chain with only result verification on-chain, Walrus is already capable.
Thus, the choice is not about 'whether it can be implemented,' but where you want to place the complexity: at the protocol layer (Irys) or the middleware application layer (Walrus)?
5. Storage Duration and Permanence
5.1 Walrus: On-demand rental model
Walrus adopts a fixed-period leasing model. When uploading data, users pay with $WAL to purchase a fixed storage period (billed per epoch of 14 days, with a maximum one-time purchase of about 2 years). After the lease expires, if not renewed, nodes can choose to delete the data. Applications can write auto-renew scripts via Sui smart contracts, turning 'lease' into de facto 'permanent storage,' but the responsibility for renewal always lies with the uploader.
The advantage is that users do not have to prepay for capacity that might be abandoned, and pricing can track real-time hardware costs. Additionally, by setting data lease expiration times, the network can perform garbage collection on data that is no longer paid for, preventing the accumulation of 'perpetual garbage.' The downside is that missing renewals or running out of funds will lead to data disappearance; long-running dApps must run their own 'keep-alive' bots.
5.2 Irys: Protocol-level guaranteed permanent storage
Irys offers a 'permanent storage' option similar to Arweave. Users only need to make a one-time payment of $IRYS to fund miners for future storage services over hundreds of years in the form of an on-chain fund (assuming storage costs continue to decline, covering about 200 years). After completing this transaction, the responsibility for storage renewal shifts to the protocol itself, and users no longer need to manage it.
The result is a user experience of 'store once, available forever,' ideally suited for: NFTs, digital archives, and datasets that require immutability (such as AI models). However, its downside is the high initial cost, and this model highly depends on the price health of $IRYS for the coming decades, making it unsuitable for frequently updated data or temporary files.
5.3 Design Summary
If you want to control the data lifecycle and pay based on actual usage, please choose Walrus; if you need unwavering long-term data durability and are willing to pay a premium for it, please choose Irys.
6. Network Maturity and Usage
6.1 Walrus: Capable of production-level scale
The Walrus mainnet has only been live for 7 epochs but is already running 103 storage operators and 121 storage nodes, with a total of 1.01 billion WAL staked. The network has currently stored 14.5 million blobs, triggered 31.5 million blob events, with an average object size of 2.16MB, totaling a storage data volume of 1.11PB (about 26% of its 4.16PB physical capacity). The upload throughput rate is approximately 1.75KB/s, with the sharding map covering 1000 parallel shards.
The economic aspect also shows strong momentum:
Market capitalization of about $600 million, FDV (fully diluted valuation) reaches $2.23 billion;
Storage price: approximately 55K Frost per MB (equivalent to about 0.055 WAL);
Writing price: approximately 20K Frost per MB
Current subsidy ratio up to 80% to accelerate early growth
Several high-traffic brands have adopted Walrus, including Pudgy Penguins, Unchained, and Claynosaurs, all building asset pipelines or data archiving backends on it. The network currently has 105,000 accounts, with 67 projects in integration, supporting PB-level data transmission for real-world NFT and gaming scenarios.
6.2 Irys: Still in Early Stages
According to the Irys public data panel (as of June 2025):
Contract execution TPS ≈ 13.9, storage TPS ≈ 0
Total storage data volume ≈ 199GB (officially claiming 280TB of space)
Number of data transactions: 53.7 million (of which June accounted for 13 million)
Number of active addresses: 1.64 million
Storage cost: $2.50 / TB / month (temporary storage), or $2.50 / GB (permanent storage)
Miner system 'coming soon' (uPoW mining mechanism not yet enabled)
Programmable data call costs $0.02 per chunk (data block), but since the permanent storage fund has not yet been in place, the actual data writing volume remains very limited. Currently, contract execution throughput performs well, but bulk storage capacity is still basically zero, reflecting its current focus on virtual machine functionality and developer tools rather than data carrying capacity.
6.3 Significance of the Numbers
Walrus has reached PB-level scale, generating revenue and passing stringent tests by consumer NFT brands. In contrast, Irys is still in its early guidance phase, feature-rich but requiring miners to join and meet data volume requirements.
For customers assessing production readiness, Walrus currently performs as follows:
Higher actual usage: Over 14 million blobs uploaded, PB-level data storage;
Wider operational scale: Over 100 operators, 1000 shards, over $100 million in staked amount;
Stronger ecological attractiveness: Leading Web3 projects have already integrated and used it;
Clearer pricing system: WAL/Frost fees are clear and transparent, with visible on-chain subsidy mechanisms.
While Irys's integrated vision may play to its strengths in the future (such as miner onboarding, permanent storage fund implementation, and TPS improvement), based on current quantifiable throughput, capacity, and customer usage, Walrus has a more practical lead.
7. Future Outlook
Walrus and Irys represent opposite ends of the spectrum of on-chain storage design:
Irys centralizes storage, execution, and economic models into a single IRYS token and a dedicated L1 blockchain for data, providing developers with a frictionless on-chain big data access experience and a protocol-level commitment to 'permanent storage.' Correspondingly, development teams need to migrate to a still-young ecosystem and accept higher hardware resource consumption.
Walrus, on the other hand, builds its erasure-encoded data storage layer on Sui, reusing mature consensus mechanisms, liquidity infrastructure, and development toolchains to achieve highly cost-effective per-byte storage costs. However, its modular architecture also brings additional coordination complexity, a dual-token experience, and ongoing attention to 'lease renewal.'
Choosing which one is not a 'right or wrong' issue, but depends on which bottleneck you care about most:
If you need deep data and computation combination capabilities, or a protocol-level 'permanent preservation' commitment, then Irys's integrated design will be more suitable.
If you value capital efficiency, rapid onboarding on Sui, or highly customized control over the data lifecycle, Walrus's modular solution is a more pragmatic choice.
In the future, the two are likely to coexist in parallel as the on-chain data economy continues to expand, serving different types of developers and application scenarios.
