The Future of Digital Art Storage, Ethereum or IPFS?

On August 24, 2026, Interplanetary Shipyard, the collective that has maintained the core of the IPFS ecosystem since spinning out of Protocol Labs in 2024, announced that Protocol Labs will not renew its funding. Its final day of IPFS work is September 30, 2026.¹ Nothing gets deleted on that date. What ends is the paid maintenance behind the software and public infrastructure most of the network runs on. The risk is not a cliff but a slow fade, and this piece is about why that fade keeps happening and what actually escapes it.

A pattern, not an accident

If this feels familiar, it should. The middle layer of Web3 has been dying in public for years, one funding decision at a time.

August 2022. Infura deprecated its public IPFS gateway. Thousands of projects had hardcoded Infura's hostname into their token metadata, and those tokens broke permanently: the address written into the chain pointed at a company, not at content.²

May 2024. Cloudflare shut down its public IPFS gateways and redirected all traffic to ipfs.io and dweb.link, praising the move as healthy stewardship.³

June 2024. NFT.Storage decommissioned uploads, telling users that retrieval would keep working but that latency and availability "may degrade over time."⁴ That phrase is the epitaph of an entire architectural era.

April 2025. RTFKT's CloneX, the Nike owned collection created with Takashi Murakami, went black across every marketplace when Cloudflare restricted the project's account. Nearly 19,800 avatars, a collection that once floored above US$60,000 per token, displayed nothing but a restriction notice.⁵⁶ The images returned days later, and the collection fled to Arweave. But a collection backed by one of the largest companies on Earth was, for a moment, one lapsed hosting plan from invisibility.

January 2026. Nifty Gateway announced its shutdown and gave collectors 90 days to withdraw roughly 650,000 NFTs. A subset of tokens minted in 2021 or earlier have metadata bound to Nifty's own servers, which the closed company has promised to keep hosting out of goodwill, forever.⁷

And the oldest wound of all: XCOPY's early works on Ascribe. When that platform fell into what the artist himself called the cryptoart platform graveyard,⁸ editions were lost. Of the ten editions of Death Wannabe, only three remain accessible, and Disaster Suit lost its metadata entirely.⁹ Some of the most historically significant cryptoart in existence survives as rumor.

The rot is measurable. A 2022 study that sampled roughly 498,000 NFTs found about 20 percent already broken: unresolvable metadata, dead image paths, IPFS links that no longer resolved.¹⁰ That was during the bull market, when everyone was still getting paid.

Every one of these failures has the same shape. The token was on the chain. The art was somewhere else. And "somewhere else" turned out to mean "wherever someone was still willing to pay for."

The migration is visible inside a single catalogue. XCOPY's own tokens read as a timeline: the early 1/1s point at IPFS, the DAMAGE CONTROL editions moved to Arweave, and the recent work lives fully onchain, from FRESH HELL through SERFIE, etched inside Ethereum three days before the Shipyard announcement.

Take the first row: Right-click and Save As guy, arguably the most iconic piece of digital art ever minted, and a perfect case study in how IPFS actually works. The 2018 piece that named the whole right-click argument stores both its metadata and its image at URLs on ipfs.pixura.io, SuperRare's gateway hostname, frozen into the token by a contract with no function to ever change it. The artwork below is embedded through that same gateway URL, live. If it renders, a company is still paying for it to render.

This one token lives on three clocks, and each one teaches a different layer of the stack.

Rescuable forever. The protocol layer, and IPFS at its best. The CID inside those URLs is a cryptographic fingerprint of the artwork's own bytes: any copy, on any node, in anyone's backup, can be verified against it and re-pinned by anyone, no permission needed. As long as one copy exists anywhere, the work can be brought back.

Visible for now. The infrastructure layer, where the bills live. What a marketplace actually displays comes through a gateway, a server somebody operates and pays for. The image renders today because SuperRare still runs one. The moment that bill stops, the art goes dark on every site that points at it, even if the bytes survive somewhere else.

Permanent never. The layer that does not exist. Nothing in the token and nothing in the protocol pays anyone to keep these bytes alive. No endowment, no incentive, no guarantee: survival is goodwill, renewed for as long as someone chooses to renew it.

One honest caveat: the image above will probably never go blank. Fame is its own pinning service, and for a work this iconic somebody will always pick up the bill. The fade comes for the undiscovered instead, the artists with a dozen collectors, the work nobody is watching. Art history is the field where value arrives late, and some of today's unknowns are tomorrow's XCOPYs. His own early editions on Ascribe were lost before the world decided he mattered.

What IPFS actually promised

None of this is IPFS's fault. The protocol is good engineering: a CID is a fingerprint of the work itself, any surviving copy can be verified against it, and that is why so much of the graveyard above is rescuable at all. The failure is in what the market asked IPFS to be: a distribution protocol, sold to collectors as a permanence guarantee. Nothing in the protocol pays anyone to keep your bytes, so keeping them is a favor, a promise with a payroll. That system never removed trust, it relocated it: you stopped trusting a URL and started trusting a maintenance budget.

Why Ethereum

There is exactly one storage layer in this industry where keeping your bytes is nobody's favor, and it was not built for art at all. Ethereum cannot function without every node agreeing on its entire state, so every full node holds every byte as a condition of participating in the first place. Storage there is not a service somebody offers. It is a property of consensus itself.

Ethereum runs on human effort too, and plenty of money. The difference is who pays, and why. Nobody has to choose to keep your art alive there: the validators are paid by the protocol itself, block by block, to preserve the state your art lives in. A pin is a favor someone renews. The chain is a machine that preserves your work because it cannot preserve anyone's money without it.

That coupling is the entire argument. There is no unpin button, because no per-file decision exists: your bytes cannot be dropped without breaking the state, and the state cannot break without breaking the money. Art stored fully onchain, the image encoded in the contract, the metadata generated by the contract, zero external references, renders and verifies byte for byte from any node on Earth, with no company, no foundation, no last man standing between it and its future. It fails only if Ethereum fails, and an entire economy is staked on that not happening. It costs more to write, once. Then there is nothing left to pay and no one left to trust.

And that one-time cost keeps falling. In the 2021 mania, gas routinely traded above 100 gwei and a single mint could cost hundreds of dollars; writing megabytes into the chain was unthinkable. Today mainnet gas sits around a tenth of a gwei, roughly a thousandfold cheaper, which is why the table above prices whole artworks in tens of dollars. And the direction of travel is set: the block gas limit has already doubled to 60 million, the Fusaka upgrade pushed it higher while lightening the load on nodes, and Vitalik's stated target for 2026 is another fivefold increase.¹¹ More room per block means cheaper bytes. Forever is getting cheaper to write.

Built for generational assets

This is the entire reason we are building ETCH'D.

THE ONCHAIN MARKETPLACEETCH'DA CYPHERPUNKS PRODUCT

ETCH'D is infrastructure for artists: the platform handles the contracts, the encoding, and the etching, and the artist walks away with an immutable, artist owned contract holding every byte of the work. One door policy applies: if a work needs a server, it does not get in. Every piece minted through ETCH'D is self contained, no IPFS pointers, no gateways, no trust in anyone's uptime, because ETCH'D is built for generational assets. Not assets for a market cycle. Not assets for a roadmap. Assets built to be inherited.

ETCH'D exists to empower artists: real tools, usable today, valuable for as long as the chain runs. We are building the future of digital art together, one etched work at a time.

The IPFS era taught collectors to ask where the image is hosted. The better question is what has to stay alive for your art to survive. For fully onchain work, the answer is one thing: the most economically secured, most replicated database humanity has ever run.

Development on ETCH'D is well underway, with a public beta launching soon. If you are an artist that is interested in working onchain, we want to hear from you: reach out to @cypherpunksart on X.

---

References

1. Wood, C. and Schmahmann, A., "The end of IPFS at Shipyard," Interplanetary Shipyard, August 24, 2026. ipshipyard.com ↩
2. Saifi, S., "Why Every NFT Marketplace Uses the Same 3 IPFS Gateways (And What Happens When They Fail)," Medium, July 3, 2025. medium.com ↩
3. "Cloudflare's public IPFS gateways and supporting Interplanetary Shipyard," Cloudflare Blog, May 14, 2024. blog.cloudflare.com ↩
4. NFT.Storage, "Notice: Uploads Decommissioned," GitHub repository notice, June 30, 2024. github.com ↩
5. "RTFKT's CloneX avatars reappear after issue blacks out NFTs," Cointelegraph, April 25, 2025. cointelegraph.com ↩
6. "Cloudflare removes RTFKT CloneX NFT art due to Terms of Service violation," Blockworks, April 24, 2025. blockworks.com ↩
7. "Nifty Gateway promises permanent NFT hosting in shutdown update," The Block, January 27, 2026. theblock.co ↩
8. XCOPY (@XCOPYART) on X, March 17, 2021: "Story behind this: I minted 7 variations of this piece on SR and gifted them to collectors who had supported my work on ascribe (which fell into the cryptoart platform graveyard)." x.com ↩
9. "NFT Marketplace Collapse: Nifty Gateway, Foundation Lead Wave of Major Platform Shutdowns," Blockonomi, February 2026. blockonomi.com ↩
10. Harrison, G., "When good NFTs go bad: We sampled almost 500,000 NFTs and found about 20% of them were broken," alwaysNFT via Medium, May 2022. medium.com ↩
11. "Vitalik Buterin's 2026 roadmap: inside Ethereum's 5x gas limit increase, 'targeted' upgrades," AMBCrypto, November 28, 2025. ambcrypto.com ↩

---

Collector FAQ

Will my IPFS artworks disappear after September 30?

No. Nothing is deleted on that date. IPFS content lives on whichever nodes choose to store it, and September 30 removes the maintainers and Shipyard's public infrastructure, not the network. Your artwork survives as long as at least one node keeps a copy pinned. The realistic risk is not a cliff but a slow fade: fewer gateways, fewer pinners, more broken images with every passing year. The Nifty Gateway and NFT.Storage stories above are what that fade looks like.

How do I check whether a piece I own is at risk?

Look at what the token actually points to: read the tokenURI on the contract, on Etherscan or any explorer. Roughly four tiers, best to worst:

• Fully onchain (a data: URI, generated by the contract): unaffected by any of this.
• ipfs:// with a CID: portable. The address is a fingerprint of the content itself, so anyone can re-pin it and any copy can be verified. Survives the death of any single provider.
• A gateway URL baked in (https://ipfs.io/ipfs/..., a Pinata or Infura hostname): at risk. If that hostname dies, the link breaks even if the content survives somewhere. This is exactly what broke tokens in the Infura shutdown. One saving grace: if the URL contains /ipfs/ followed by a CID, the CID can be extracted and re-pinned, so the art is rescuable even after the hostname dies.
• A plain server URL (https://api.someproject.com/...): fully dependent on that company paying its hosting bill.

What should I actually do before the 30th?

Three moves, in order of effort. First, download and back up the raw image and metadata files of anything you care about; your own copy beats every promise in this article. Second, if the token uses ipfs:// addresses, pin those CIDs yourself with a paid pinning service or your own node; pinning is permissionless and you do not need the artist's involvement. Third, if you are an artist or control a contract with updatable metadata, migrate to Arweave, or properly, to the chain itself.

The image already fails to load. Is the art gone?

Maybe not. If the token carries a CID, any surviving copy, on any node, in anyone's backup, can be verified against it byte for byte and re-pinned. That is the genuinely good part of IPFS's design. If the token points at a dead hostname, recovery depends on someone, somewhere, having kept the file.

Are Arweave-backed works affected?

Not by this announcement. Arweave is a separate network with an endowment model designed to fund storage indefinitely. It is a materially stronger promise than a pin, and it is still a pointer to a system whose economics have to keep working for your art to exist.

What does "fully onchain" mean, precisely?

The tokenURI is generated by the contract from data stored on the chain, image included, typically as a base64 encoded data URI. You can render it, verify it, and inherit it from any Ethereum node, with no other party involved.

I'm an artist. How do I publish my work on ETCH'D?

ETCH'D is in development, with a public beta launching soon. The best preparation is the file itself: the work must be fully self contained and render from its own bytes, with no external fonts, scripts, or references. Our guide The Weight of Forever covers formats, optimization, and what etching costs by size. If you are an artist that is interested in working onchain, we want to hear from you: reach out to @cypherpunksart on X.

Doesn't onchain storage cost a fortune?

It costs more than a pin, once, and nothing ever again. Small works cost a few dollars, multi megabyte works are real money on mainnet and far less on Shape. The full fee schedule, priced live from the chain, is in our guide The Weight of Forever.