[Intermediate] — what the 2021 boom got wrong, what the technology genuinely provides, and where it’s actually being used.
What happened
2021–22 saw tens of billions of dollars trade in profile-picture collections. By 2023 most of it was worth a small fraction of peak, and trading volume fell by well over 90%.
Worth being precise about what failed. The technology worked exactly as designed: ownership was tracked, transfers settled, provenance was verifiable. What failed was the thesis — that a public ownership record of a JPEG creates durable value in the JPEG.
Three things the boom got wrong:
- Conflating the token with the artwork. The token is a row in a mapping pointing at a URL. It is not the image, and it conveys no copyright unless separately granted by licence → tokens said plainly
- Assuming scarcity creates value. Anyone can mint 10,000 of anything for a few dollars. Artificial scarcity of an infinitely reproducible thing is a coordination game, and coordination games end when the coordination does
- Treating a thin, reflexive market as a price. Floor prices were set by a handful of trades against near-zero liquidity, with substantial wash trading — much of it driven by marketplaces’ token incentives
What the technology actually provides
Strip out the speculation and the useful properties are real:
- Verifiable, permissionless ownership — anyone can check who owns what, without asking a company
- Transferability without an intermediary — no platform approval, no settlement window
- Provenance — the complete chain of custody, publicly auditable, forever
- Composability — other applications can read and build on your ownership record without permission
- Persistence beyond the issuer — the record survives the issuing company
That’s a genuinely useful primitive. It’s a registry, and it’s most valuable where the underlying thing is a right, a credential or an access grant — not where it’s an image.
Where it’s actually working
Ticketing. The strongest fit. Transferable, verifiable, with resale rules enforceable in the contract and a permanent record against counterfeiting. Adopted in production, quietly.
Domain names. ENS is an NFT, and it’s arguably the field’s most successful — because the token is the asset, not a pointer to one. vitalik.eth doesn’t reference a name elsewhere; the on-chain record is the whole thing → identity and naming.
In-game items. Genuine when the game is genuine. Items that outlive the game, are tradeable without the publisher’s marketplace, and are usable across games. The last one is mostly still aspiration — cross-game interoperability requires games to agree on meaning, which is a design problem, not a blockchain problem.
Credentials and memberships. Certifications, event attendance, DAO membership, club access. Often better as soulbound (non-transferable) tokens, where transferability would defeat the purpose.
Real-world asset tokenisation. Deeds, invoices, carbon credits. Legally uncertain and heavily dependent on off-chain enforcement — the token means what a legal system says it means, and no chain can make a court recognise it. Real work is happening here; treat claims carefully.
Art, honestly. A smaller, calmer market persists. Fully on-chain generative art (Art Blocks, Autoglyphs) has held up best, which is unsurprising: those pieces have no external dependency and the artwork genuinely is the token.
The design decisions that matter
If you’re building with NFTs, these are the ones that determine whether it ages well:
1. Where is the metadata? A plain HTTPS URL means the project’s server is a permanent dependency. Use IPFS with real pinning, Arweave, or fully on-chain → storage.
2. Is the metadata frozen? If the owner can change tokenURI, holders are trusting them. Freeze it, or say clearly that you haven’t.
3. What rights come with it? Nothing, by default. If you intend to grant licence rights, do it explicitly in terms you publish. CC0 and the Nouns/Loot approach — abandoning rights entirely — has produced more durable ecosystems than restrictive licensing.
4. Royalties are not enforceable on-chain. ERC-2981 is advisory; marketplaces choose whether to honour it, and most stopped when competition on fees intensified. Attempts to enforce via transfer blocklists fragment liquidity and mostly failed. Build a business model that doesn’t assume perpetual royalties.
5. Reentrancy in mints. safeMint calls onERC721Received, handing control to the minter mid-transaction — used repeatedly to bypass per-wallet limits. Apply Checks-Effects-Interactions and a guard → reentrancy.
6. Fair distribution. On-chain randomness is manipulable, so naive reveals get sniped for rares. Use Chainlink VRF, or reveal after minting closes → randomness.
Key insight
NFTs are a good ownership registry attached to a bad story about digital art. The registry — permissionless, verifiable, transferable, outliving the issuer — is genuinely useful for tickets, names, credentials and game items, where the token is the right rather than a pointer to a file on someone’s server. The 2021 boom tested the story, not the technology, and it’s the story that failed.
Related
- tokens, coins and NFTs — the data model
- decentralised storage — where the image lives
- identity and naming — ENS, the success case
- the honest assessment
Software engineering
Infrastructure & security
Computer science
Hardware & engineering