NFT Infrastructure: Storage, Metadata, and Standards That Last

Most NFT projects encounter infrastructure problems rather than contract problems, and the infrastructure decisions are made early and are hard to revisit. This guide covers them, drawing on FISTA Solutions' blockchain engineering work.

What does the stack look like?

Four layers, and only the first is on-chain.

Layer	Durability question
Token contract	On-chain; permanent
Metadata URI	Points where? Can it change?
Metadata document	Who serves it, and for how long?
Media file	Same question, larger file
Traits and rarity	Derived; usually off-chain
Marketplace display	Depends on all of the above

Why is storage the central issue?

Because the token is a pointer and the thing people bought is at the other end of it.

A token whose metadata URI returns nothing is still owned, still transferable, and entirely empty. Collections have effectively evaporated this way when a hosting account lapsed.

Content addressing helps: an identifier derived from the content means any node holding that content can serve it, and the identifier proves the content is unaltered. But somebody still has to hold it.

What does pinning involve?

Paying for content to remain available, continuously.

Content-addressed networks do not guarantee persistence by themselves. Data remains available because someone pins it, and pinning is a paid service or self-run infrastructure with the same obligations.

Budget for it in perpetuity or use a storage model with an endowment mechanism that pre-pays for a long period. Assuming the network will keep your data for free is how collections disappear.

When should metadata be mutable?

When the asset is meant to change, and stated as such.

Evolving assets, game items, and staged reveals all need mutability. That is legitimate. What is not legitimate is mutability by accident — a URI pointing at a server where anyone with access can change the image after sale.

Be explicit in documentation and, where the standard allows, on-chain. Buyers evaluating a collection should be able to determine whether what they see can change. See smart contract upgrade patterns.

How should minting work?

With gas cost, fairness, and reveal manipulation all considered.

Minting many tokens in one transaction is far cheaper per token than individual mints, which matters for large collections. Lazy minting defers creation until purchase, shifting the cost to the buyer.

Reveal design matters: if metadata is available before the reveal, or if the assignment is predictable, participants will identify and claim the valuable tokens. Commit to the assignment before it can be observed.

What is the reality of royalties?

They are marketplace policy, not a contract guarantee.

The standard signals a royalty preference that marketplaces may honour. Transfers can occur without any marketplace involvement, and marketplaces have made royalties optional.

Build the business model around what is enforceable — primary sales, utility, access rights tied to holding — rather than around a secondary revenue stream you cannot compel. This is general guidance, not legal advice.

What does the indexing layer need?

An index that resolves tokens to current owners and metadata, because marketplaces and your own interface both need it.

Querying ownership per token from a node does not scale to a collection view. An index of transfers gives you ownership, holdings per address, and history.

Cache resolved metadata rather than fetching from decentralised storage on every page view, which is slow and unreliable. See blockchain data indexing.

What are the common mistakes?

Metadata on a server you control. Unpaid pinning. Accidental mutability. Predictable reveals. Revenue models dependent on royalties. And fetching metadata from decentralised storage on every request.

How do you test it?

Verify that every token's metadata and media resolve, from a clean client, on a schedule. A broken pointer discovered by a holder is worse than one discovered by monitoring.

Test the mint under load, because a collection launch is a traffic event as much as a contract event.

What does it cost to operate?

Contract deployment and auditing, storage and pinning in perpetuity, and indexing infrastructure. The perpetual storage commitment is the one most often left out of the budget.

Gas costs at mint depend heavily on the mechanics chosen.

What should you measure?

Proportion of tokens whose metadata resolves, media availability over time, pinning cost trajectory, index lag, and gas cost per mint.

Where does AI fit?

In generation and in moderation. Generated artwork raises provenance and rights questions that should be settled before minting, not after.

Moderation matters for user-generated collections: automated screening before minting is far cheaper than dealing with content already on-chain, which cannot be removed. This is general guidance, not legal advice.

When is this the wrong approach?

If the goal is a record of ownership within a controlled system, a database achieves it with none of this complexity. Tokenisation earns its cost when transferability outside your system genuinely matters.

What should you do first?

Check whether your collection's metadata is content-addressed and pinned by someone paid to keep pinning it. If not, that is the durability problem to solve first.

How FISTA Solutions helps

FISTA Solutions builds and operates production systems through web and mobile, AI enablement, and staff augmentation: content-addressed metadata with pinning funded as an ongoing obligation, and mutability declared as an explicit property rather than left implicit, decisions documented with their reasoning, and handover that leaves your team able to maintain what was delivered. The record is 150+ projects for 50+ companies across 12+ countries.

To scope this work, message FISTA on WhatsApp, or read blockchain data indexing.