Buterin recasts Ethereum as a cryptographic computer aiming for 8-to-32-second finality

Start with the finality gap: 8 to 32 seconds against roughly 200 for the 12 confirmations users wait through today ^[4][1]. That is a latency claim, and it points to where the protocol's engineering effort is going before any of the machinery ships.

Buterin's argument is that Ethereum is already a hybrid of Satoshi Nakamoto's blockchain and cryptographic tools that "did not even exist" in 2009, and is still called a blockchain "to a large extent for historical reasons" ^[2][3]. The relabelling matters less than what he says gets rebuilt to earn it.

Three properties are being replaced outright, in his telling ^[5]. Verification stops meaning every node downloads and re-executes every block, and starts meaning a node samples data with PeerDAS and checks a SNARK ^[6]. Consensus moves to what he calls a "much better optimized" proof-of-stake model ^[7]. Block construction stops being one party deciding what goes in and becomes several parties building a block together ^[8]. Slots would run 4 to 8 seconds, and a node would need far less hardware ^[9].

Cheaper node hardware will decide whether the decentralisation Buterin is selling actually holds. His case is that decentralisation stops being purely a cost: it lets data sit in parallel, computation run in parallel in the mempool, and metadata hide in ways a centralised system cannot manage ^[10]. He says this was the mid-2010s dream and it failed then for one reason, that there was no way to verify split-up work had been done correctly, and that committees sampled at random were "complicated to set up and expensive" ^[11][12]. Modern cryptography, he argues, solves it, with overhead "decreasing month by month" ^[13].

The timeline is a draft. The post is a companion to the Strawmap, the Ethereum Foundation's draft schedule of seven forks through 2029 ^[17], and Hegota, next year's fork, is only "likely to be Ethereum's last 'normal' fork" ^[1]. And Buterin is candid that the hard near-term work is unglamorous: making zero-knowledge proofs fast and safe enough is "encapsulated complexity" already being worked on with AI tools, while managing and parallelising access to very large amounts of state is "more difficult, and systemically complex" ^[15][16].

Privacy arrives in pieces on his account, with general-purpose computation staying expensive and only partly private while special-purpose applications get "very strong privacy" from zero-knowledge proofs ^[14]. He is explicit that the furthest-out piece, indistinguishability obfuscation, is not something the conclusions depend on: "all of the conclusions in this post will apply long before any of that becomes available," he wrote ^[18][19].