Summary

Sam Williams’s core re-rating case is that Arweave already behaves like a benchmark L1, not a speculative storage app. The protocol has operated for six years, stores roughly 6 billion pieces of information and now approaches 1 billion transactions a month—versus slightly under 1 billion across its first five years. Some 30 chains use it for chain data, roughly 85% of leading Solana NFT collections rely on it, and Sam’s shorthand is “just like Bitcoin, except for permanent storage at scale.”
AR’s value-accrual mechanism ties real usage to long-duration scarcity without charging users a shareholder-style rent. Customers prepay 200 years of storage across 20 replicas; assuming a stable token price, the endowment remains sustainable if storage costs decline just 0.5% annually, versus the roughly 38.5% historical Kryder rate. Compared with Filecoin’s maximum two-year deal, Sam estimates the circulation effect is amplified about 100-fold: the tokens “may as well be burned.”
Arweave’s answer to centuries of unknowable technical change is a permissionless market for protocol forks. An innovator can ship a complete upgrade at a specified block height, mint compensation and let users decide whether the improvement justifies its dilution; competing forks can offer the same fix more cheaply while preserving the data set. The roughly 274-word “constitution” has never been triggered, but Sam believes it creates an “efficient market for improving the protocol.”
AO turns Arweave’s permanent logs into an arbitrarily scalable computer by abandoning global shared state for message passing. Deterministic processes infer state from data on Arweave, while separate networks handle computation, sequencing and messages—an architecture Sam contrasts with Ethereum’s roughly 70,000 nodes performing “the amount of computation that you would get in a literal desktop calculator.” His pitch is the familiar “world computer, but this time it actually works.”
A showcase for the AI-crypto thesis was Llama 3’s 7 billion-parameter model running inside an AO smart contract and making financial decisions. Llama Coin’s “Llama Fed” evaluates requests to print tokens, with users burning Arweave tokens for compute; Sam explicitly warned it is “not a long-term store of value.” The broader thesis is “agent finance”: autonomous hedge funds, portfolio balancers and other AI strategies whose rules and custody remain trustless.
Composable data is Sam’s consumer-scale bet: creators upload once, retain rights and let every application reuse the same open data set. The Universal Data License can set different terms for personal use, commercial use or AI training, while content funds could buy future content or royalties—down to selling 20% of a tweet immediately instead of waiting months for a platform payout. His promised combination is “maximum distribution and maximum ownership rights.”
The scalability claim is already testable rather than purely architectural. Arweave typically handles 350–450 transactions per second and has suffered one cited outage—14 minutes on mainnet’s sixth day; AO already generates 25% of transactions and Sam thinks it might reach 95% within a year as agents interact up to 10 times a second. Spam is priced through storage and compute fees, while AO lets users buy access to a chosen amount of collateralized security for a chosen duration.
AO’s zero-premine distribution is designed to address Ethereum’s key moat—liquidity—without first forcing investors to sell ETH. Supply converges on 21 million with four-year halvings, and slightly less than two-thirds is allocated over time to bridged assets; applications earn ownership by attracting that liquidity and may share rewards with users. Sam’s formulation is “put your economic residency there,” creating a potential liquidity flywheel while bridgers retain exposure to assets such as staked ETH.

Deep dive

1. Arweave turned permanent storage into a protocol, not a product

Sam defines Arweave as a “true protocol, not a product”: it has no core team in the conventional sense, only multiple organizations—including Forward Research—that advocate for and build around a decentralized permanent-information network.
Development began roughly seven years before the recording, with mainnet launching just over six years earlier. Its mandate is deliberately narrow: on-chain storage comparable to Bitcoin’s, scaled to arbitrary data sizes and priced for “the maximum possible time period.”
The first product-market fit was less solemn than humanity’s permanent archive. Sam calls NFT storage the “seats in the casino”: roughly 85% of top Solana NFT collections use Arweave, while some 30 chains store data there so future node operators can reconstruct their histories.
The hosts’ emerging reframe was that genuinely decentralized applications increasingly require this persistence layer. Sam accepts the “backbone of Web3” description, though he says careful marketing left actual utility ahead of market understanding.

2. A 200-year endowment makes “pay once” economically legible

A storage payment is sized to fund 200 years across 20 replicas at current prices. As hardware gets cheaper, the endowment’s purchasing power rises; with a stable token price, average annual cost declines above 0.5% leave it stronger after each year.
The historical comparison supplies the safety margin: Kryder’s law has delivered approximately 38.5% annual storage-cost declines over 50 years. Sam reaches further back to cave paintings—humans have reduced the labor required to encode information for tens of thousands of years.
Jonah’s challenge was whether permanence eventually meets a physical wall. Sam placed theoretical density near (10^{68}) bits per cubic centimeter versus roughly (10^{13}) today, adding that storage could still be stacked afterward: “You actually have to care about this stuff” when designing for 1,000 or 10,000 years.

3. Evolutionary forks trade ossification for a market in upgrades

The hosts’ governance question—Bitcoin-style immutability or Ethereum-style responsiveness—occupied Sam for two years. He thinks today’s Arweave can last decades, but “realistically something will need to change” on century and millennium horizons.
Conventional DAOs failed his test because they are effectively “publicly traded on-chain companies,” eventually incentivized to maximize extraction from users. Bitcoin forks preserve rights better, but Sam’s calculations suggested that below $500 million worth of Bitcoin, the resulting price increase would rarely recover the cost of development, persuasion and publicity.
An evolutionary fork instead ships one complete code change activating at a stated block height—no “trust me, bro, I’m going to implement it later.” Its author can mint tokens; the market then decides whether the improvement deserves that dilution, while rivals can offer the same change for 3% rather than 100%.
The shared data survives across branches, allowing a successful fork to route around a future failure. The roughly 274-word constitutional framework has never been invoked and might sit unused for years; meanwhile, the team is already considering DNA storage at (10^{27}) bits per cubic centimeter, perhaps commercially viable in half a decade to a decade.

4. Temporary decentralized storage lacks the demand permanence creates

Sam’s distinction is functional: encrypted temporary data on AWS or Google Cloud is rarely censored, so decentralization adds little. Permanent storage, by contrast, cannot credibly be promised by a company whose business model and existence may change.
His own experience sharpened that view. After mining Bitcoin around 2011, he bought roughly $15 of ETH for compute credits and also invested in Storj; Storj did not appreciate much because, despite attempts to build demand, the market for decentralized temporary storage remained limited.
Filecoin is Sam’s cautionary specimen: roughly 70% of supply was allocated to mining rewards, yet storage allegedly found few buyers at one-thousandth of AWS pricing, then free, then with users paid to store it. Despite market-cap recovery, he said the token traded only around 1.5 times its 2017 price; any viable temporary-storage service should “probably be tokenless.”

5. AR’s accrual comes from usage, not subsidized supply

Sam’s first token-design rule is blunt: “Don’t print crazy numbers of tokens to subsidize supply when you don’t have any demand.” Arweave instead compounded utility through bear markets, with six years of stable operation making permanence progressively more credible.
Filecoin locks payment for a maximum two-year deal; Arweave collects 200 years upfront. Sam estimates this magnifies the reduction in token velocity roughly 100-fold, making the distinction between locking and burning economically “kind of immaterial” over centuries.
The design does not require raising prices to extract value. Users pay what the protocol believes supply truly costs—“literally, you pay zero cents above”—while greater usage removes more tokens from circulation; the safety reserve remains available for unknowable storage conditions 50 years out.
Avi’s investor question captured the market gap: unlike most middle-risk-curve altcoins, AR purchases pay for an already-used product. Sam thinks recognition has improved recently but spread slowly because “storage” was reflexively bucketed with Filecoin; listings also emerged organically, with Arweave reaching most major exchanges after signing very few agreements.

6. Ethereum’s scaling trap motivates AO’s message-passing design

Sam’s provocation is that Ethereum’s roughly 70,000 nodes collectively produce calculator-scale computation: “the world’s most enormous calculator.” Jonah noted that the calculator is at least reliable; Sam then softened the trash-talk, calling arbitrary decentralized computation an extraordinary innovation while maintaining that Ethereum’s scaling constraints trapped the community.
His deeper criticism is sequencing. Ethereum built a large community before solving scale, and “protocols are about communities”; once applications depend on a particular trade-off, securing consensus for a fundamentally different architecture becomes much harder than pivoting a Web2 product.
Global shared state—every program potentially accessing every other program’s memory—is the constraint Sam believes Solana has optimized nearly to its practical limit. AO chooses the internet’s model instead: “Screw global shared state,” because the global financial system and the internet itself are message-passing machines.

7. AO turns stored computation into an agent-capable supercomputer

Arweave has stored deterministic computation logs since SmartWeave around 2020. The hosts’ key pushback was latency: infinite storage does not help if reconstructing state takes forever. AO answers with a peer-to-peer computation subnet rather than requiring every participant to perform the same work.
The architecture separates blockchain functions into computation, messaging, sequencing and data availability. Compute nodes stake against correct state transitions; messaging units route outputs to scheduling units, which order and upload them to Arweave—nicknamed “MUs, CUs and SUs.”
Its showcase was Llama 3’s 7 billion-parameter model running on-chain. Llama Coin lets users petition randomly assigned AI central-bank personalities to print tokens, burning Arweave tokens for computation; Sam calls it a “fiat simulator” but warned listeners not to treat the coin as durable value.
The serious extension is “agent finance.” Sam imagines autonomous hedge funds and portfolio balancers whose models execute inside smart contracts; Jonah added the access angle—a bedroom developer without institutional credentials could deploy a strategy, and strangers could trust its custody and withdrawals as they trust Uniswap.

8. Composable data could make creators—and applications—portable

Arweave held roughly 6 billion pieces of information at recording. Transactions had risen from slightly under 1 billion in the first five years to nearly 1 billion monthly; Forward Research was pairing that infrastructure with Odysee’s 7–8 million monthly users and the smaller Solarplex network.
Sam’s TikTok example makes composability concrete: rather than populating an empty database, a developer queries the shared Arweave data set for vertically oriented videos under 30 seconds and builds a new interface. The application starts with content instead of undertaking a decade-long cold start.
The Universal Data License leaves creators as rights holders. They can make individual use free, charge commercial or AI-training royalties, tokenize those rights, or sell 20% of a tweet immediately; autonomous content funds could bid for work they expect to monetize.
Identity travels with the content through cryptographic keys, letting audiences verify whether a politician’s clip is authentic across applications. Sam ties that to speech rights and portability: the underlying identity and data remain in the open system even if a platform deplatforms a user, while developers inherit an open “data lake” rather than platform lock-in.

9. Arweave says the scale is already observable

Normal throughput was 350–450 transactions per second, which Sam compared with Solana’s occasional 500–600. His complete outage history was “14 minutes on day six of mainnet”; otherwise, nodes had remained online doing work consistently for six years.
AO already produced about 25% of Arweave transactions, and Sam thought it might reach 95% within a year. Agents could interact “10 times a second” without human decision costs, making machine activity a likely source of the next volume surge.
A storage-spam attacker still pays the quoted fee; AO’s testnet charged specifically for intensive LLM compute, while mainnet was expected to price all computation. Sam’s answer to becoming a “victim of your own success” was that insufficient adoption remained “the least of my worries.”
AO also prices security directly. A user might request four hours for fraud detection and $1 million of claimable collateral around a $50,000–$100,000 transfer, temporarily renting stake rather than purchasing scarce block space to subsidize security indirectly.

10. AO’s fair launch is designed as a liquidity magnet

Asked why an institution should still settle on Ethereum and merely archive on Arweave, Sam identified two moats: “mimetic prominence” and liquidity. Developers choose Ethereum because capital already resides there, even when another execution environment fits better.
AO’s response is a 100% community distribution with zero premine, a 21 million terminal supply and four-year halvings. Slightly less than two-thirds goes to users bridging assets such as ETH or staked ETH from other networks.
When bridged capital uses an application, the right to future AO rewards can flow to that process; developers may retain it or return 50% or 100% to users. No grants council has to select winners—applications earn protocol ownership by attracting genuine liquidity.
Sam calls the mechanism “put your economic residency there.” A fund can preserve required ETH exposure while earning AO, then borrow through tools such as the overcollateralized Astro stablecoin and explore the ecosystem; at recording, the distribution mechanism had been live only three or four days.

11. Bitcoin, Ethereum and Solana supply the benchmark—and the contrasts

Bitcoin remains Sam’s first mechanism-design inspiration: virtually nothing else achieved decentralized commodity status, yet Satoshi did so without precedent. Some ingredients may have been a “happy accident,” which only makes the result more striking.
Ethereum created arbitrary decentralized computation; Solana then pushed global shared state “to its nth degree” and executed aggressively on adoption. Sam saw that firsthand when Metaplex stored its data on Arweave and both teams worked 16–18-hour days through Arweave 2.0’s scaling transition.
His market lesson is that “the best technology doesn’t always win. But, when it’s well packaged, it tends to.” By the close, Jonah explicitly moved Arweave from the generic altcoin bucket into benchmark infrastructure, while Sam kept the caveat that markets are only “eventually efficient.”

Surviving Crypto's Summer Chop | 1000x The New Administration Means Major Change for Crypto, AI & Tech w/ Salim Ismail | EP #143