Open your wallet. Pick any NFT you own — the one you'd never sell. Now answer one question honestly, without opening a block explorer: can you name the server that holds the actual picture?
Most people can't. And that is the entire story.
On April 23, Nifty Gateway — the curated NFT marketplace that Gemini built out and ran — closes its withdrawal window. Before it does, it is offering users a batch extraction tool. It is also, according to commentary published on CryptoSlate, promising to keep hosting metadata "indefinitely." That word — indefinitely — is doing an enormous amount of work in that sentence. It is a promise made by a company that has already decided to stop doing this.
The telling detail isn't the shutdown. Marketplaces come and go. The telling detail is that a shutdown requires this much explanation in the first place. Nobody had to write a thousand-word explainer about what happens to a game library when a platform exits a business line, because nobody was ever under the illusion that a storefront inventory was a self-contained object they carried around with them. NFT buyers were. That illusion was the product.
We don't panic here. We verify. So let's verify what "owning" an NFT actually consists of — and why the bill for it is arriving now, quietly, on a schedule nobody put in the mint price.
The Rise Was Real, Which Is Why the Fall Matters
Nifty Gateway arrived with an unusual pedigree for an NFT marketplace. It sat under Gemini, the New York–regulated exchange founded by Cameron and Tyler Winklevoss, which meant it carried something rare in this sector in 2020 and 2021: a compliance department. Nifty's pitch was curation. It handled the messy parts for artists — the minting, the hosting, the drop mechanics — and in exchange it took a cut and kept the infrastructure under its own roof.
That model worked spectacularly for about eighteen months.
The numbers from that window still read like a fever chart. In March 2021, Christie's sold Beeple's Everydays: The First 5000 Days for $69.35 million. In September 2021, Sotheby's moved 101 Bored Ape Yacht Club tokens for $24.4 million. These weren't crypto-native auctions in a Discord server. These were the two most storied auction houses on earth, applying the same catalog treatment to digital JPEGs that they apply to Fabergé eggs and Old Master paintings. The commentary is blunt about what that represented: the peak of collective mania, dressed in institutional clothing.
Then the music stopped. Trading volumes collapsed. The speculative bid evaporated. And at some point in that decline, Gemini made a rational business decision — focus on core financial applications, stop running a curated art marketplace.
What interests me is not that the decision was made. It's how it was framed. The CryptoSlate piece describes the ending as "sad and anticlimactic," and that's exactly right. There was no explosion. No hack, no insolvency headline, no founder fleeing a jurisdiction. There was a wind-down notice, a batch export tool, a deadline, and a paragraph of reassurance about metadata hosting.
Here's the part that should worry you — and the part that shouldn't. The part that shouldn't: Nifty did not steal anything. It gave users a window to extract their assets and a migration path for the content it had been hosting. By the standards of this industry, that is an orderly retreat, and it deserves to be described as one.
The part that should: an orderly retreat is not the same thing as an obligation discharged. "We will host it indefinitely" is a sentence about intent, not a technical guarantee. It has no smart contract behind it. It has no third-party audit. It has no escrow. It is a company that has already announced it is leaving this business, promising to keep doing a thing it is leaving the business of doing.
That gap — between what a platform says it will maintain and what is actually enforced — is where the whole story lives.
Three Layers, One JPEG, and a Magic Trick
To understand why this matters, you need to understand that your NFT is not one thing. It never was. It's three things stacked on top of each other, and they are not equally durable.
The first layer is the token itself. This is an ERC-721 entry on Ethereum — or an equivalent standard on another chain. It records one fact and records it beautifully: who owns this token ID right now. That record is replicated across thousands of nodes. It will survive the collapse of any single company, any single server, any single jurisdiction. It is, genuinely, permanent.
The second layer is the metadata — typically a small JSON file describing the asset. It carries the name, the description, the attributes, and critically, a pointer to the image. This file lives off-chain in almost every case. It is not on the blockchain. It is reachable through a function called tokenURI.
The third layer is the media file itself — the actual JPEG, PNG, MP4, whatever it is. Also off-chain. Also reached by a pointer, this time from inside the metadata.
Your wallet or your marketplace takes these three layers and stitches them into a single, seamless rectangle on your screen. You look at it and your brain says: I own that picture.
Your brain is being flattered.
I've done this by hand, and so can you. Four years ago, during the EOS airdrop verification blitz, I spent weeks manually auditing wallet addresses across Telegram groups to separate genuine community holders from sybil attackers. The lesson I carried out of that sprint wasn't about EOS specifically. It was that the thing users believe they hold and the thing the chain actually records are frequently two different objects, separated by a layer of interface that nobody audits because it looks like it's working.
Putting a large media file on-chain is expensive — prohibitively so for most art. Choosing to keep it off-chain was a reasonable engineering decision in 2017, and it remains a reasonable one. The cost of that decision is a dependency. And dependencies have to be paid for, maintained, and eventually explained to the person who thought they bought an object but actually bought a claim on a service.
The commentary captures this in one line that I keep coming back to: the buyer was supposed to get something outside the reach of any platform, and instead "inherited a durable relationship with a hosting department."
Read that again. Not a file. A relationship. With a department. At a company.
One Line of Code Decides Whether Your Art Survives
Here's where it gets concrete, and where you can start doing something useful tonight.
Every ERC-721 contract has a tokenURI function. It returns a string — usually a URL — that tells the world where to find the metadata. That single string determines your asset's fate more than any rarity trait ever will.
There are broadly four places that string can point.
A centralized server, often the company's own domain. If your tokenURI points to a company-owned host, then your art survives exactly as long as that company keeps paying the hosting bill. When the company winds down, the link dies. If the link is also immutable in the contract, you cannot fix it. Not you, not the artist, not a DAO, not a court. The chain will faithfully record that you own a token whose image will never load again.
An HTTP URL on a third-party host. Same failure mode, plus an extra party. Whoever runs the domain has to keep renewing it. Registrars do not care that your JPEG is culturally important.
An IPFS content identifier. This is where most people think they're safe, and where a great many of them are not.
An Arweave transaction ID. The strongest of the common options, with caveats we'll get to.
The key insight — and the one the CryptoSlate piece articulates with unusual precision — is that these three layers can survive for entirely different lengths of time. The token is permanent. The metadata might last a decade. The image might vanish next Tuesday. Nothing in the wallet interface will tell you which of those clocks is running on which asset.
We don't panic here. We verify. And the verification is genuinely straightforward: open a block explorer, find your contract, call tokenURI, and read the string it returns. I have now done this for my own small collection, and I found exactly what I expected to find — a mix. Two assets pointed at IPFS hashes. One pointed at a URL I did not recognize. One pointed at an endpoint that resolved, but slowly, to a server in a jurisdiction I had to look up.
That's the audit. It takes about ninety seconds per asset. Almost nobody has done it.
IPFS and the Trap Hidden in Its Own Name
The InterPlanetary File System is one of the genuinely elegant ideas to come out of the last decade of infrastructure work. Instead of identifying a file by where it lives, IPFS identifies it by what it is — a cryptographic hash of its contents. Ask the network for that hash and any node that has a copy can serve it to you. Location becomes irrelevant. Redundancy becomes natural.
That architecture is real, and it's good. What it is not is a storage service.
This is the sentence the commentary nails and the industry has spent five years avoiding: joining the network is not the same as the network saving your file.
IPFS has no built-in persistence guarantee. Nodes cache content they've recently served. When a node needs to reclaim space, it runs garbage collection, and it will cheerfully garbage-collect your file — because from the node operator's perspective, your file is just cache. It's not personal. It's a memory management decision.
The only thing that prevents this is pinning. Pinning is an explicit instruction to a node: keep this, forever, do not collect. Someone has to issue that instruction, and someone has to pay for the disk, the bandwidth, and the electricity behind it.
So when you see an NFT that says "stored on IPFS," you've learned something about the addressing scheme. You have learned nothing about whether anyone is currently pinning it.
There's a second layer of confusion stacked on top. Most people don't query IPFS directly — they go through a gateway, which is a normal web server that fetches content from the IPFS network on your behalf and serves it to your browser. Gateways are convenient and they are a single point of failure. When a gateway goes down or rate-limits you, your image doesn't load. The data is still there. The pipe is broken.
This produces the single most damaging form of confusion in the entire asset class: users cannot tell the difference between "my art is gone" and "my art is temporarily unreachable." Those are completely different problems with completely different remedies, and the standard NFT interface gives you no way to distinguish them. The commentary flags this diagnostic problem explicitly, and it deserves more attention than it has received. One leads to an unnecessary panic sale. The other leads to a necessary one.
Arweave, the Endowment, and What "Permanent" Really Costs
Arweave takes a fundamentally different approach, and it's the one Nifty Gateway chose for the assets it migrated.
Arweave is a network built around a single, aggressive promise: pay once, store forever. You pay a fee today, and that fee is split — part goes to miners who store the data now, and part goes into a storage endowment, a pool whose investment yield is meant to cover the cost of storing your data indefinitely into the future.
This is a clever model, and it's a much better fit for how people think they bought an NFT than anything else on the market. It matches the psychological shape of the purchase. You paid once. You expect to own it forever. Arweave is the only major storage layer that structures its economics around that intuition.
Which is precisely why I want to be careful here, because a lot of people are about to treat Arweave as a synonym for "problem solved," and it isn't.
The endowment model depends on a set of assumptions: that the storage cost per byte continues to fall as fast as it historically has, that the price of the AR token doesn't collapse in a way that erodes the endowment's purchasing power, and that the network continues to attract enough miners to keep serving data. The commentary is appropriately cautious on this point — it notes that "permanent" rests on everything from the chain's continued operation to incentives keeping storage providers engaged.
Those are reasonable assumptions. They are not guarantees. And converting a dependency on one company into a dependency on a token economy is a trade, not an escape. It is usually a good trade — a distributed network of self-interested miners is a sturdier custodian than a wind-down notice — but it is a trade.
Here's the part that should worry you — and the part that shouldn't. What shouldn't worry you: Arweave being imperfect is not a reason to prefer a centralized server. That's strictly worse. What should worry you: anyone telling you a storage layer has removed the problem entirely is selling you a second illusion to replace the first one.
The Migration That Cannot Run Backwards
Now the detail that turns a general principle into a specific, dated exposure.
When Nifty Gateway wound down, it migrated newly hosted metadata and media over to Arweave. That's the right call, and it's the reason the wind-down deserves credit rather than contempt.
But migration is prospective, not retrospective. The commentary is clear on this: early NFTs in the collection have their metadata permanently linked to Nifty's own servers. Those links were written into contracts at mint, and contracts do not have an undo button. You cannot migrate a link that is already frozen into immutable code.
So what happens to those assets? They stay alive through one thing only: the company's stated commitment to host indefinitely, extended past the point where the company has any commercial reason to exist in this market.
Let's keep the community calm and the record straight. This is not a reason to panic-sell. It is a reason to know. There are, right now, a large number of tokens sitting on Ethereum whose ownership records are as permanent as the chain itself, and whose content depends on a promise from a division of a company that is closing that division. The chain will always tell you that you own it. It will never tell you that you can see it.
I have started calling these zombie assets, and I think the term is going to spread. They are not dead — the token is very much alive and transferable. They are not fully alive either, because the thing that makes them legible as art is on life support administered by a third party. And the crux of it, as the commentary puts it, is that a promise about migration describes what a company says it intends to do. It does not prove that any individual migration succeeded.
The Rent That Was Never In the Price
The title of the CryptoSlate piece is "The NFT party is over and everybody now owes storage rent." That framing is sharper than it looks, because it identifies a genuine economic mismatch at the center of the entire asset class.
Look at how the money actually flows.
A marketplace earns a one-time fee when an NFT is minted or sold. The artist earns a one-time payment, plus whatever royalties the standard allows. The buyer pays once. Every revenue event in this stack is a spike — a discrete transaction that completes and is done.
Now look at the cost side. Media files need to be stored continuously. Someone pays for a disk, a bandwidth allocation, and electricity every single month, forever, for as long as the asset is expected to remain viewable. That's not a spike. That's an annuity running in the opposite direction.
In a bull market, this is invisible. Trading fees are fat. Storage is a rounding error absorbed by the platform as a cost of doing business, and nobody ever puts it on a line item. In a bear market, trading volume evaporates and the storage bill stays exactly the same. Now it's a pure liability attached to a product line that isn't producing revenue.
Which is exactly the moment a rational company decides to focus on core financial applications.
Here is the cost-bearing structure in plain terms, and it's worth internalizing:
The centralized platform pays server costs for as long as it remains in business and interested. This is the weakest link in the chain, because it's the only one tied to a company's strategic mood.
The IPFS pinning provider charges a per-gigabyte monthly subscription. The data persists while the subscription is paid, and stops when it isn't. This is honest, at least — it's a service with a bill, like any other.
The self-hosted node costs hardware, electricity, and maintenance, borne by the holder. This is the most durable option and the one with the highest technical barrier.
Arweave takes a one-time payment and converts it into an endowment-backed perpetual storage claim, which is the closest thing the market has to a genuine buy-once-own-forever model.
I want to be precise about one thing, because precision matters when you're talking about other people's money. This structure is not a Ponzi scheme. There is no promise of fixed returns, no mechanism where new buyers pay interest to old ones, no compounding obligation that must eventually break. The commentary's framing is accurate and worth respecting: what exists here is deferred cost, not fraud.
Early collectors received storage that was effectively subsidized by platform economics. That subsidy was never priced into what they paid. When the platform exits, the cost isn't destroyed — it becomes visible, and it lands on the holder. It's a quiet transfer from the future to the present, and the present has just arrived.
The 2021 Cohort Is the Exposure
If you're trying to figure out whether this touches you, start with a date.
The NFTs most at risk were minted in the window when the market was hottest and the infrastructure was least considered. That's 2021 through early 2022 — the Beeple era, the Sotheby's BAYC auction, the period when drops sold out in seconds and nobody in the room was asking where the file would live in five years.
Those are the assets most likely to have metadata hardcoded to a specific company's domain, because that was the default behavior of most minting tooling. Those are the assets most likely to have been bought by people who have never opened a block explorer. And those are the assets whose original platforms are, statistically, least likely to still be operating at full strength.
Combine that with what the commentary identifies as the deeper risk — that collectors widely believe a token is a complete asset — and you get a large population of holders who may be carrying content that has already, silently, become unreachable, and who have no framework for checking.
I've done this by hand, and so can you. During the Terra collapse in 2022, I coordinated a community truth effort that ended up aggregating verified loss stories and debunking viral misinformation in Discord channels. I personally worked through more than a thousand user queries. The single most common pattern I saw was not ignorance. It was confidence — people who were certain they understood their exposure and were wrong in a specific, structural way. That's the same failure mode I expect here, and it's why I'd rather write this now than after the complaints start.
The blue chips are not automatically exempt. A Bored Ape's rarity trait — the gold fur, the laser eyes, the crown — contributes enormously to its price and contributes exactly nothing to answering the question of where its metadata lives. The commentary makes this point with a line I'd like framed: the ape's rare hat contributes very little to diagnosing the problem, even though it contributes a great deal to the price. Rarity is a pricing mechanism. It is not a durability mechanism. Those are different axes, and the entire market has been collapsing them into one.
Your NFT Might Be Collateral, and the Collateral Might Be Hollow
There's a downstream consequence that hasn't been priced at all, and it sits in DeFi.
NFTs have spent three years being positioned as collateral. Lending protocols accept them against stablecoin loans. Fractionalization platforms carve them into tranches. The entire premise is that an NFT is a bearer asset — something you hold, something that can be seized and liquidated, something whose value can be marked to a floor price.
But a bear token that can't render is a different kind of asset than a bear token that can.
I spent a chunk of 2020 inside Compound's cToken interest rate models, decoding why rates were spiking and why retail was panic-selling into a mechanism they didn't understand. The lesson that stuck wasn't about interest rate curves. It was that a collateral system is only as sound as the weakest thing it treats as uniform. Compound treated every cToken as equivalent, and they weren't. The market learned that the expensive way.
Apply the same lens to NFT collateral. Two tokens with identical floor prices can have radically different content-availability profiles. One points at a well-funded Arweave transaction. One points at a domain that expired nine months ago. Every lending protocol I'm aware of prices both identically, because none of them are querying tokenURI and checking whether the string resolves.
That's an unpriced risk sitting inside a lending market. The commentary flags it as an implicit shock to the NFT-collateral thesis, and I think that undersells it slightly — this isn't just a valuation question, it's a liquidation question. What does a forced sale of an unrenderable NFT actually recover?
The Lazy Read, and the Read That Actually Matters
Let me push against the obvious conclusion now, because the obvious conclusion is wrong in an interesting way.
The obvious conclusion is: "This proves NFTs are worthless." You can find that take in a thousand quote-tweets within an hour of any story like this. It's satisfying. It's also lazy, and it's lazy in a way that misses the genuinely useful signal.
Here's the contrarian read. This is not the story of NFTs collapsing. This is the story of NFTs growing up, and growth is expensive.
Every asset class goes through a phase where the exciting narrative runs far ahead of the boring infrastructure, and then a phase where the boring infrastructure has to be built before the narrative can survive a second act. Aviation had it. Early internet commerce had it. Container shipping had a decade of bankruptcy before it had a decade of dominance, because the ships were ready long before the ports, the labor agreements, and the insurance markets were.
What's happening right now is that NFTs are being forced to answer a question that was always there and was never urgent: what are you actually made of?
The answer, it turns out, is: a permanent ownership record and a fragile content layer. The ownership record is genuinely novel and genuinely valuable. The content layer is an unsolved dependency problem. And the market, until this week, had been pricing the first and assuming the second.
That's not a collapse. That's a correction in what the thing is. And assets that survive a redefinition like that tend to be worth more afterward, not less, because the buyers who remain know what they bought.
Here's the part that should worry you — and the part that shouldn't. What shouldn't: the fact that NFTs have a storage dependency. Every digital object has one. What should: that the dependency was never disclosed at the point of sale, which means an entire generation of buyers made a purchase decision on false premises. That's a disclosure failure, and disclosure failures get regulated.
The RWA Crowd Should Be Reading This Closely
Now let me point this at the people who will be loudest to dismiss it — the tokenized-real-world-asset crowd.
For three years, I've watched RWA get sold as the bridge between traditional finance and public blockchains. The pitch deck is always the same: bring treasuries on-chain, bring real estate on-chain, bring private credit on-chain. And for three years, the actual institutional adoption has been thin, slow, and mostly confined to permissioned environments that the retail public chain crowd never actually gets to touch.
The reason is visible in exactly what Nifty Gateway just demonstrated, and it's the reason I've been skeptical of this narrative since the first version of it circulated. Traditional institutions do not need a public chain. They already have settlement systems that clear. What they need, and what they've always needed, is a legally enforceable custody arrangement with audited counterparties and clear liability allocation.
Look at the NFT structure again. Layer one is a permanent, censorship-resistant ownership record. Layers two and three are a custody relationship with a third party that can exit at will. Now replace "JPEG" with "treasury bond" or "title deed to a property." The custody problem gets dramatically harder, the regulatory exposure gets dramatically larger, and the consequences of a broken link are no longer aesthetic.
The Nifty Gateway wind-down is a free case study for anyone pitching tokenized assets. It shows, at small scale and low stakes, exactly what happens when the on-chain layer and the off-chain layer drift apart. Multiply that by the value of real-world assets and you're not looking at a sad story about digital art. You're looking at a systemic design flaw that the entire industry has agreed not to discuss.
And there's a parallel worth naming while we're here. The stablecoin market has a similar structure: an enormous promise of redemption, resting on reserve attestations that have never been subject to a truly independent audit, with roughly seventy percent of the market concentrated in a single issuer. Everyone in this industry knows that. Everyone writes around it. We are, collectively, extremely good at not looking at the load-bearing wall.
There's a further irony for anyone watching from Asia. Hong Kong's virtual asset licensing regime is being pitched as a regional embrace of innovation, and it's worth being honest about what's driving it — the same thing driving every financial-center policy decision in that region for forty years. This isn't a philosophical conversion. It's a competitive repositioning, mostly aimed at a neighbor that spent the previous cycle winning the family offices. The NFT storage crisis has nothing to do with that contest, and everything to do with the same underlying dynamic: institutions showing up to extract value from infrastructure they did not build and have no particular interest in maintaining.
Arweave Is Not an Exit, It's a Different Door
I want to close the contrarian section on the recommendation everyone is about to make, because it's being made too cleanly.
The reflex advice after a story like this is: "Store your NFT on Arweave and stop worrying." That's better advice than most. It is not complete advice, and incomplete advice delivered confidently is how the next version of this problem gets built.
Arweave's permanence model depends on an endowment that has never been through a genuine multi-decade stress test. It depends on storage costs continuing to decline along their historical curve. It depends on AR maintaining enough value that the endowment's yield covers real storage costs. And it depends on a mining ecosystem that stays sufficiently incentivized to keep serving data even when the fee revenue from new uploads slows down.
Every one of those dependencies is plausible. None of them is proven. And the honest framing is not "Arweave solves it" — it's "Arweave is a better class of dependency than a company, and you should still understand what you're depending on."
The larger point is that every storage solution is a dependency. The decentralized ones are simply dependencies with better-aligned incentives and no single point of failure. That's a real improvement, and it's not the same thing as removal.
We don't panic here. We verify. And verifying means understanding what your specific asset is actually standing on.
What Happens Next, and What to Watch
The Nifty Gateway story will be over in the news cycle within a week. The structural problem it exposed will still be there in 2030. So here's what I'm tracking, and here's what I think you should track with me.
The withdrawal window aftermath. April 23 is the date that matters. What I want to see is how many users report content failures in the weeks after it closes, and whether those failures are gateway problems or genuine losses. The volume of complaints will determine whether this stays a footnote or becomes a template.
The fate of the "indefinite hosting" commitment. This is directly checkable by anyone. If the metadata servers stay up and resolving a year from now, the commitment was honored and the model deserves credit. If they quietly go dark, we've learned something important about the enforceability of informal custody promises — and about how much weight to give similar language from other platforms going forward.
The Arweave migration success rate. Migration is a process, and processes have failure rates. Chain-level tracking of how many of Nifty's assets actually landed and remained retrievable will tell us whether the redeem path is as solid as it looks on paper.
Whether wallets ship content-availability checks. This is the highest-leverage fix available, and it costs almost nothing. A wallet that runs tokenURI, attempts to resolve it, and shows a green, amber, or red indicator next to each asset would eliminate the entire diagnostic gap overnight. No protocol change required. I'll be watching for the first wallet to ship it, and I'll be watching how long the rest take to copy it.
Whether marketplaces standardize availability disclosure. The equivalent fix on the trading side is a mandatory field — something like a content-availability attestation — displayed before purchase. Standards like this get adopted the same way they always do: one platform does it, users notice, competitors follow.
Whether blue-chip collections audit their own metadata. This is the one I'd most like to be wrong about. If major collections come forward with public, verifiable audits of their metadata architecture, the market gets a floor to stand on. If they stay silent, the silence is informative.
And for those of you holding assets today, the practical sequence is short enough to do this week:
Pull your contract addresses. Query tokenURI on each one. Note whether it returns an IPFS hash, an Arweave transaction ID, or a URL. For the URLs, check the domain — who owns it, does it resolve, how long has it existed. For the IPFS hashes, figure out whether anyone is actually pinning them. Distinguish carefully between failure modes: open the raw content in a separate tab before you conclude anything is lost, because a dead gateway and a dead file look identical in a wallet and are completely different problems.
Then write it down. A simple list, asset by asset, of what each one is standing on. It takes an afternoon.
The uncomfortable truth of the last four years is that the market spent a long time asking the wrong question. It asked whether a JPEG could be worth a million dollars, and it got an answer. The question it never asked — whether the JPEG would still be there in five years — is the one that just became load-bearing.
When the storage question becomes the thing people ask before they bid, and not the thing they discover after the platform closes, this asset class will finally have something it has never had: a durable floor. Not a floor price. A floor of legibility — a shared, verifiable understanding of what every token actually contains and what every token actually depends on.
Everything on-chain is a promise about records. Everything off-chain is a relationship. The next cycle belongs to whoever builds the layer that makes the second one legible to the first.
***
Glossary
ERC-721 — The Ethereum standard that records NFT ownership. It proves each token is unique and defines how it transfers. It does not store the media file.
Metadata — The off-chain description file for an NFT, containing the name, attributes, and a pointer to the media. It is the bridge between the token and the art, and it is the most common point of silent failure.
tokenURI — The ERC-721 function that returns the location of the metadata, usually as a URL. Where this string points determines your asset's long-term survival far more than any trait does.
IPFS — A content-addressed storage protocol that identifies files by cryptographic hash rather than server location. Elegant, location-independent, and carrying no persistence guarantee on its own.
Pinning — An explicit instruction to an IPFS node to retain a specific piece of content and exempt it from garbage collection. Without pinning, files are cache, and cache gets cleared.
Garbage collection — The process by which a storage node reclaims space by discarding unneeded data. In IPFS, unpinned content is eligible for collection at any time.
Gateway — A web server that fetches IPFS content on a user's behalf and serves it over HTTP. Convenient, and a single point of failure. A dead gateway looks identical to lost data.
Arweave — A decentralized storage network built on a one-time-payment, permanent-storage model, funded by an endowment that is meant to cover indefinite future storage costs.
Content addressing vs. location addressing — Content addressing identifies a file by what it is (IPFS, durable against server changes). Location addressing identifies it by where it lives (an HTTP URL, dead the moment the server or domain is).
Custody — The holding of an asset by a third party on a user's behalf. Any custody arrangement creates a continuing dependency on the custodian's continued operation and good faith.
Endowment — Arweave's storage fund, built from a portion of each upload fee, whose investment yield is intended to pay for storage indefinitely into the future.
Zombie asset — My term for a token with a permanent, fully functional on-chain ownership record and an unreachable or soon-to-be-unreachable content layer. Legally alive, practically hollow.
***