Walrus vs Filecoin: Erasure Coding or Replication?

Filecoin gives you more storage per dollar and continuous on-chain proofs, at the cost of slow cold reads. Walrus gives you fast reads and programmable storage objects, at a higher price per gigabyte. Pick on read frequency first, then on cost.
We sell both, which is unusual enough to be worth stating up front — it means we have no reason to talk you out of either, and the comparison below is the one we actually use internally when advising teams.
The engineering difference
Filecoin stores sealed replicas. Each provider runs your data through an expensive encoding bound to their identity, then proves continuously that they still hold it. The sealing cost is what makes the proofs meaningful and what makes retrieval slow — sealed data must be unsealed before it can be served.
Walrus erasure-codes each blob into slivers spread across nodes, reconstructable from a subset. There is no sealing step and no unsealing step. A client pulls slivers in parallel and decodes, which is why reads are fast, and total overhead sits around 5x the original size rather than a multiple of full copies.
Everything else follows from that one choice.
Side by side
| | Filecoin | Walrus | |---|---|---| | Redundancy | Sealed replicas | Erasure-coded slivers | | Storage overhead | Per full replica | ~5x total | | Read performance | Cold; needs a hot layer | Fast, by design | | Retention model | Deal term, renewable | Epochs, renewable | | Coordination chain | Filecoin | Sui | | Storage proofs | PoRep and PoSt, continuous | Availability via encoding | | Programmability | Deal records | Blobs as Move objects | | Cost per GB | Lower | Higher | | Maturity | Longer track record | Newer |
Pricing, concretely
Through Lighthouse, August 2026:
| | Filecoin | Walrus | |---|---|---| | Free tier | 5 GB | 100 MB | | Entry paid | $12/month, 500 GB | $11/month, 250 GB | | Premium | $49/month, 2.5 TB | $79/month, 2.5 TiB |
At the entry tier the headline prices look similar, but Filecoin gives twice the capacity for a dollar more. At Premium the gap is stark: the same capacity costs $49 on Filecoin and $79 on Walrus.
The free tiers differ by a factor of fifty. If you are evaluating or prototyping, start on Filecoin for that reason alone.
Choosing on read frequency
The question that decides it is not cost. It is: how often will this data be read after it is written?
Rarely read → Filecoin. Archives, backups, compliance records, historical datasets, anything written once and consulted occasionally. You are paying for durability and verifiability, not for serving. The cold-read penalty is irrelevant when reads are rare, and the cost advantage compounds across terabytes.
Frequently read → Walrus. Application media, user-generated content, game assets, anything an audience actually loads. Here retrieval latency is the product experience, and the premium buys something your users perceive.
Continuously read and written → Walrus. Agent workloads are the clearest case. Autonomous systems do not have human upload patterns — they read and write constantly, in small operations, at machine pace. Storage optimised for occasional archival retrieval is the wrong shape for that.
Choosing on ecosystem
If you are building on Sui, Walrus has a structural advantage beyond performance: blobs are Move objects, so your contracts can hold, transfer and reason about storage directly. That is not a convenience — it is a capability Filecoin does not offer, and it enables patterns like automated renewal driven by contract logic. See blobs as Move objects.
If you are on an EVM chain, that advantage does not apply, and the decision reverts to read frequency and cost.
Choosing on proof requirements
If you need to hand a third party evidence that data was stored and continuously proved — a regulator, an auditor, a counterparty in a dispute — Filecoin's PoRep and PoSt records are the stronger artifact. They are public, continuous, and independently checkable. Proof of replication and proof of spacetime covers what they assert.
Walrus's guarantee is availability through encoding and committee coordination, which is a real and well-designed property but a different one, and less useful as evidence in an adversarial setting.
You do not have to choose permanently
Through Lighthouse both backends sit behind the same interface: same CIDs, same SDK calls, same gateways. Migration preserves identifiers, so moving a dataset from IPFS and Filecoin to Walrus does not break references already embedded in contracts or documents. See migrating from IPFS to Walrus.
That makes the decision reversible, which should lower the stakes considerably. A sensible default: start data on Filecoin, move the subset that turns out to be read-heavy to Walrus once you have real traffic data rather than a guess.
The honest summary
If cost per terabyte is the constraint and reads are rare, Filecoin, and it is not close. If retrieval speed is what your users experience, or you are building on Sui, or your workload is an agent rather than a person, Walrus earns its premium.
Most teams end up with both, split by access pattern rather than by preference.
Get Started
- Introduction to IPFS and Walrus
- Introduction to IPFS and Filecoin
- Migrate IPFS to Walrus
- Pricing
- Create an API key and start on the free tier
Stay in Touch
To learn more about Lighthouse, visit the official website, read the documentation, or jump in on GitHub. You can also join the community on Discord, X, Telegram, or LinkedIn.











































































