Key takeaways
- Reporting indicates a marketplace involved in the NFT boom is now addressing storage costs with holders — a consequence of the fact that NFT images were almost never stored on a blockchain.
- A token typically contains a pointer to data held elsewhere. If that pointer is an ordinary web address, the file persists only while somebody continues hosting and paying for it.
- Content-addressed storage proves which file was intended but still requires somebody to store it. Fully on-chain tokens are genuinely permanent and are rare because the cost is high.
- Separately, crypto token buybacks have reached a record of roughly $638 million through late August 2026, with two projects accounting for close to 90% of the total.
In 2021 a Beeple NFT sold at Christie's for $69.35 million, and Sotheby's sold a group of Bored Ape tokens for $24.40 million. Five years later, per reporting, one of the marketplaces that helped build that market is explaining storage costs to the people who hold those tokens.
This is not a scandal and it is not a failure of the technology. It is a technical detail that was disclosed, was documented, and was almost universally skipped — and it is worth understanding properly, because the same detail applies to a great deal of what is being tokenised right now.
The image was never on the blockchain
Here is the fact that explains everything else.
Blockchain storage is extraordinarily expensive per byte, because every full node stores every byte forever. Putting a multi-megabyte image on a major chain would cost a great deal and impose that cost on the entire network permanently.
So almost nobody did it. What the token actually contains is a pointer — a reference to data held somewhere else. The blockchain records, immutably and permanently, that you own token number so-and-so, and that this token points to a location.
What it does not do is guarantee that anything is still at that location.
Three ways that pointer can be built
An ordinary web address. The weakest version. The token points at a URL on a server. The file exists while somebody pays for that server, keeps the domain registered, and chooses to continue. Many early tokens point at infrastructure that no longer exists. The token remains perfectly valid and it points at nothing.
A content identifier. Better. Here the address is derived mathematically from the file itself, so it proves which exact file was intended — any alteration produces a different identifier. This solves authenticity completely and does not solve persistence, because somebody still has to store the bytes. If every party hosting that file stops, the identifier remains correct and unresolvable.
Fully on-chain. The strongest and the rarest. The image data itself lives in the contract, usually as generated graphics rather than photographs, because the cost scales with size. These are genuinely permanent for as long as the chain exists. Very few projects chose this, and the ones that did were making a deliberate and expensive decision.
Why this was predictable in 2021
None of this was hidden. The contracts were public and the metadata structure was readable by anybody who looked. The information was available to every purchaser.
What happened was more ordinary than deception. During a rapid price rise, almost nobody performs technical due diligence on the thing they are buying, because the thing being bought was not really the file. It was the expectation of resale. And for that purpose, where the image is hosted is genuinely irrelevant.
The question only becomes material when the resale expectation disappears and the holder is left with what they actually own. Which is the point at which somebody has to answer who is paying the hosting bill.
The part that matters far beyond NFTs
This is why the story is worth your attention even if you never owned one.
The same structure appears in essentially every tokenisation of anything. A token is a record of a claim. The thing claimed is somewhere else. And the durability of the whole arrangement is determined by the weakest link in that chain, not by the blockchain component.
A tokenised bond depends on the legal structure holding the bond. A tokenised property depends on whether a court recognises the claim. A stablecoin depends on the reserves and the issuer's willingness to redeem. A wrapped asset depends on the bridge's lock remaining intact.
In every case the blockchain does one job extremely well — recording who holds the claim, immutably and publicly. And in every case that job is the easy part. The hard part is off-chain and it is where the failures happen.
The question to ask of anything tokenised
One question, and it works universally.
If every company involved in this ceased to exist tomorrow, what would I still have?
For a fully on-chain token, you would still have the artwork, because it is in the contract. For a token pointing at a company's server, you would have a valid record of ownership of a dead link. For a tokenised real-world asset, you would have whatever the legal structure survives with — which is either robust or nonexistent depending on documents that have nothing to do with the token.
That single question cuts through most of the marketing in this industry, and it can usually be answered from public information in a few minutes.
Elsewhere: a record year for buybacks
Separately, and more structurally interesting than it sounds: crypto token buybacks reached roughly $638 million through late August 2026, per Allium Labs data — a record, up from about $545 million over the same period in 2025. Two projects accounted for close to 90% of it, with roughly $370 million from one and about $200 million from another.
Buybacks are worth understanding rather than cheering. A protocol using revenue to purchase its own token is doing something comparable to a share repurchase: returning value to holders by reducing supply rather than by distributing cash. Whether that is a good use of capital depends entirely on what else the money could have done, and on whether the revenue funding it is durable.
The detail to check is always the same one we raised about burns: compare the buyback against ongoing issuance over the same period. A protocol buying back a large-sounding amount while issuing more than that is net expanding supply, and the buyback figure alone conceals it.
The market
Bitcoin is around $84,570, up modestly on the day and about 5% over the week. Ether is near $2,711, also up around 5% on the week. Tether is holding approximately parity.
Also circulating: Michael Saylor has published a proposed "bill of digital rights" covering individuals and companies using digital assets, framed around enabling companies to raise capital. It is an advocacy document rather than a policy development, and worth reading as a statement of one industry position rather than as a regulatory signal.
What to take from this
The blockchain part of an NFT worked exactly as designed for five years. The ownership records are intact and verifiable. What is failing is a hosting arrangement that nobody examined, because during a price rise nobody examines anything.
The lesson generalises. In almost every application of this technology, the on-chain component is the reliable part and the off-chain dependency is where the risk lives. Find the off-chain dependency first, and you will understand the thing you are holding.
Frequently asked questions
Was my NFT image really never on the blockchain?
For the large majority of tokens, correct. Blockchain storage is extremely expensive per byte because every node stores it permanently, so the token almost always contains a pointer to data held elsewhere. You can verify this yourself: read the token's metadata URI on a block explorer. If it is an ordinary web address, the file's persistence depends on whoever operates that server. A small number of projects did store image data fully on-chain, and those are genuinely permanent.
Does the token become worthless if the file disappears?
The ownership record remains intact and verifiable forever — that part of the system continues working exactly as designed. What may be lost is the ability to display the associated artwork from the original source. Whether that makes the token worthless is a question about what people valued in the first place, and the honest answer is that it depends entirely on the specific project and its community.
Does content-addressed storage solve the problem?
Partly. Because the identifier is derived from the file itself, it proves exactly which file was intended and detects any alteration. That solves authenticity completely. It does not solve persistence — somebody must still store the bytes. If every party hosting that file stops, the identifier stays mathematically correct and cannot be resolved to anything.
How do I check what I actually own?
Look up the contract on a block explorer and read the metadata URI for your token. An ordinary URL means dependence on that host. An IPFS or similar content identifier means dependence on somebody pinning the data. Data embedded directly in the contract means genuinely permanent. Then ask the general question: if every company involved ceased to exist tomorrow, what would remain?
Why does this matter if I do not own NFTs?
Because the structure is identical across tokenisation generally. A token records a claim; the thing claimed is elsewhere. A tokenised bond depends on its legal structure, a stablecoin on its reserves and redemption, a wrapped asset on a bridge's lock. In each case the blockchain handles ownership records well and the off-chain dependency is where failures occur. Identifying that dependency is the core skill.
Not financial advice. Crypto assets are volatile and unregulated in many jurisdictions. In India, gains are taxed at 30% with 1% TDS on transfers. Do your own research and never invest money you cannot afford to lose.
Editorial note: Crypto Shakti uses an AI-assisted research and drafting workflow. Every article is grounded in the linked primary sources and live market data captured at publication time.