How Arweave prices storage for 200 years

Paying Once for Permanent Data and the Bet on Falling Storage Costs

This page is general education about how a storage network funds itself. It is not financial or investment advice, and it does not recommend buying any token.

Most digital storage is rented. A cloud account, a web host, or a domain name keeps working only while someone keeps paying, and when the payments stop, the data tends to disappear. Arweave tries a different model: the uploader pays once, up front, and that payment goes into a shared storage endowment meant to fund copies of the data far into the future. The idea behind it is a futurism question with an abundance answer. If the cost of storing a byte keeps falling, then a finite payment today can cover an indefinite future, and the scarcity of reliable long-term memory starts to turn into abundance. This page explains how that pricing works, where the idea came from, and what would make it fail, so readers can judge any "store it forever" promise on its numbers instead of its slogans.

The core idea in one sentence

Permanent storage at a fixed price is not a technology claim so much as an economic bet that storage keeps becoming less expensive, and the size of that bet can be measured.

How the price is set

According to the Arweave lightpaper, users "pay upfront for the storage of 20 replicas for 200 years, at present costs." The protocol estimates the present cost of storing one gibibyte for one minute, averaged over a six-week window, doubles it to get the cost for one block of about two minutes, and multiplies by 20 copies, 200 years, and the size of the data.

That payment is not handed straight to miners. Most of it enters the endowment, which the lightpaper describes as "a faucet through which miners are paid out over time" as they prove they still hold copies of the dataset. The earlier yellow paper explains that the network only draws on the endowment when block rewards and fees are not enough to cover the cost of keeping the whole dataset stored for that period.

The key design choice is the margin of safety. The lightpaper states that because users pay for 200 years at present prices, "only a 0.5% Kryder+ rate is sufficient to sustain the endowment for an indefinite term, in the absence of token price changes." The Kryder+ rate is the yearly decline in the full cost of storing a unit of data, including hardware, electricity, and operations.

Why falling costs make a finite payment last forever

Imagine paying for one year of storage today. If next year the same storage costs a little less, the money left over covers slightly more than a year. If the decline continues every year, the total cost of all future years adds up to a finite number, the same way a shrinking series of payments can have a fixed sum. The yellow paper writes the perpetual price as the infinite sum of each future period's storage cost, which only stays finite because each term keeps getting smaller.

The historical record behind this is strong. The yellow paper reports an average yearly decline in the cost of storing a gigabyte for an hour of 30.57 percent over 50 years, while the later lightpaper cites a figure of about 38.5 percent per year. The two documents use different numbers, which is a reminder to read such figures as estimates rather than laws of nature.

Libraries tried this before blockchains

The endowment model did not start with crypto. Digital preservation researcher David Rosenthal noted that in 2010 Princeton described an endowed data service based on an analysis that charging about double the initial cost could store data forever. Rosenthal built an economic model with two unknowable inputs, the interest rate earned by the unspent endowment and the Kryder rate. At the historic 30 to 40 percent yearly Kryder rate, the endowment needed was modest and not very sensitive to the exact rate. Below about 20 percent per year, the needed endowment rose quickly, roughly doubling by 10 percent, and became very sensitive to the rate.

Rosenthal also pointed to a real shock. In late 2011, floods in Thailand destroyed about 40 percent of the world's disk manufacturing capacity, and the price per byte almost doubled, staying high for more than two years. Any pay-once system has to survive events like that, not just the long average.

What recent hard drive prices show

Backblaze, a cloud storage company that buys drives in bulk, published its purchase records from 2009 to November 2022. Its average cost fell from about $0.114 per gigabyte in 2009 to about $0.014, an 87.4 percent decrease, or about 0.52 percent per month. From 2017 to 2022 the decline was slower, a little over 9 percent per year.

Read together, these sources tell a clear story. The decline has slowed a great deal from the 30 to 40 percent era, which is exactly the zone Rosenthal warned about for traditional endowments. Yet even 9 percent per year is far above the 0.5 percent that Arweave says its design requires. The margin is the whole point of charging for 200 years at present prices.

What could still break the model

• Token price. The endowment is held in the network's own token, so its purchasing power rises and falls with the token's market price. The lightpaper itself notes that endowment health depends on token price changes as well as on the actual Kryder+ rate.
• Storage cost shocks. A disaster like the 2011 floods can reverse the decline for years.
• Hidden costs. Rosenthal's minimal model notes a rule of thumb that ingest is about half of total lifetime preservation cost, storage about one third, and access about one sixth, so storage price alone understates the job.
• Network survival. The yellow paper separates data permanence from network permanence. Its authors say they do not expect the network to keep producing blocks in true perpetuity, and they expect the data to be carried into a later storage system, with social incentives taking over from financial ones.

Why this matters for freedom tech and abundance

Rented storage has a single point of failure: whoever stops paying, or whoever is pressured into deleting. A pay-once, many-copy model shifts that risk. It cannot remove every risk, but it turns the long-run fall in storage costs, one of the steadiest abundance trends in technology, into a way to fund memory that does not depend on one company's goodwill or one person's monthly budget. That matters for people who cannot afford ongoing fees, for archives of censored material, and for wikis and research notes meant to outlast their authors. It connects directly to the problem of link rot and to efforts at backing up civilization.

Honest limits

The figures on this page come from the project's own documents, one company's drive purchases, and one research program's models. None of them can predict storage costs or token prices 200 years ahead. The 0.5 percent requirement is the project's own claim, made under the stated assumption of no token price changes. Readers should treat any promise of permanent storage as a well-reasoned bet with known weak points, not as a guarantee.

Discussion questions, essay ideas, and learning related AI prompt ideas

• Explain why a payment that covers 200 years of storage at present prices could last indefinitely if storage costs fall by only 0.5 percent per year.
• Compare the Princeton endowed data service with the Arweave storage endowment and describe what changed between the library approach and the blockchain approach.
• Describe how the 2011 Thailand floods would affect a pay-once storage system, and what design choices could absorb a shock like that.
• Argue whether holding a storage endowment in a volatile token makes permanent storage stronger or weaker than holding it in conventional assets.
• Calculate how long a fixed storage budget would last at Kryder rates of 0, 5, 9, and 30 percent per year, and explain what the results show.
• Discuss whether falling storage costs are a form of abundance that ordinary people can capture, or one that mainly benefits large companies.
• Explore which kinds of knowledge most deserve permanent storage, and who should decide.

See also

• Permaweb basics
• Permanent decentralized wiki design
• IPFS and Arweave for agent memory
• Link rot and reliable knowledge
• Local-first software
• Time prices and abundance
• Censorship-resistant money