Assets referencing off-chain content depend on that content remaining available, and the guarantees provided by different systems vary considerably.
Content addressing
Referencing data by a hash of its content rather than by location.
Which guarantees that retrieved data matches the reference.
It guarantees nothing about whether anyone is still storing it, which is the point most frequently misunderstood.
Pinning
Explicitly instructing a node to retain data.
Which is what actually keeps content available on peer-to-peer storage networks.
Unpinned content is garbage collected, and this has caused documented loss of referenced media.
Pinning services
Commercial operators that pin content for a fee.
Which reintroduces a paid service and a dependency.
Content pinned by a single service disappears if that service stops, which has happened.
Incentivised storage networks
Protocols paying storage providers with cryptographic proofs of continued storage.
Which provides an economic guarantee rather than a voluntary one.
Proofs of replication and proofs of spacetime are the mechanisms used.
Retrieval versus storage
Proving data is stored differs from making it available quickly.
Which is why retrieval markets developed separately.
Data can be provably stored and slow to retrieve, which matters for anything user-facing.
Deal duration
Storage agreements have terms and require renewal.
Which means permanence requires ongoing action or an arrangement that handles it.
Some services offer prepaid long-term storage funded from an endowment model.
On-chain storage
Encoding data directly in contract state or transaction data.
Which provides the strongest availability guarantee and is extremely expensive.
Fully on-chain collections exist and are necessarily small.
Practical assessment
Find where the metadata resolves, check whether the content is pinned and by whom, and consider what happens if that party disappears.
Collections whose media no longer loads are a documented and recurring outcome.
Gateways
HTTP endpoints serving content-addressed data to conventional browsers.
Which is how most users actually access it.
Gateway operators are a dependency, and gateway outages make content appear unavailable.
Mutability
Content addressing means changing the content changes the address.
Which is a feature for integrity and a problem for anything requiring updates.
Naming systems mapping stable names to changing content addresses exist for this reason.
Cost comparison
Incentivised networks are frequently cheaper than conventional cloud storage for cold data.
Which is a genuine finding and comes with different retrieval characteristics.
Total cost including retrieval is the meaningful comparison.
Verification
Anyone can verify that retrieved data matches its address.
Which is a real property that conventional storage does not provide.
Practical advice for issuers
Pin with multiple independent services, consider an incentivised network for durability, and document what happens if the project ends.
Redundancy
Storing multiple copies across independent providers.
Which is what actually produces durability.
Replication factors are configurable on incentivised networks and default settings vary.
Encryption
Data stored on public networks is readable unless encrypted before upload.
Which is the user's responsibility rather than the network's.
Key management for encrypted stored data is a separate problem entirely.
Legal and content considerations
Content-addressed networks make removal difficult by design.
Which creates genuine tension with legal removal obligations.
Gateway operators generally implement filtering as the practical response.
Use in practice
Metadata for tokens, application front ends and archival data are the main current uses.
Which suit content that is written once and read many times.
The core distinction
Content addressing gives integrity; storage incentives give availability. They are different guarantees and both are needed.
For anyone buying tokens with off-chain media
Check where the metadata resolves and whether it is pinned by more than one party.
Which takes one lookup and predicts whether the item will still display in five years.
The recurring failure
Collections whose media no longer resolves, because the project stopped paying a pinning service and nobody else was storing the content.
Which is entirely predictable from the design and continues to happen.
Permanence is a service somebody pays for rather than a property of content addressing.
Practical durability planning
Multiple pinning services, an incentivised storage deal with a long term, and a documented plan for renewal.
Which is what an issuer serious about permanence actually does.
Endowment models funding storage from returns on a deposited sum have been proposed and used in some projects.
For users assessing a collection
Resolve the metadata URI yourself and see what responds.
A centralised web server behind an asset marketed as permanent is a common and revealing finding.
A last note
Content addressing and storage are two separate guarantees, and the marketing for these systems has consistently blurred them. Knowing which one you actually have determines whether your data survives the year.
Summary
Verify where the content lives, who is paying to keep it there, and what happens when they stop.
Three questions, all answerable before you rely on the arrangement.
Assets that outlive their issuers are the ones where somebody answered all three at the outset.