Ethereum co-founder Vitalik Buterin has declared the zkEVM “battle for speed” to be “definitively won” and says the top priority should now “shift fully to security.” In an October 15 post on
Ethereum co-founder Vitalik Buterin has declared the zkEVM “battle for speed” to be “definitively won” and says the top priority should now “shift fully to security.” In an October 15 post on X, he congratulated ZKsync’s Airbender team on a milestone that redefines the zero-knowledge race: real-time Ethereum proving on a single graphics card.
The statement reads as more than praise. It moves the finish line for every zkEVM — away from faster benchmarks and toward proofs that are themselves provably correct.
ZKsync unveiled Airbender Prime on October 13, calling it the first Ethereum prover, to its knowledge, to demonstrate real-time proving on a single GPU. Running on one NVIDIA RTX PRO 6000 Blackwell card, the system generated proofs for a consecutive batch of 7,200 Ethereum mainnet blocks; 99% of those proofs finished in under 10 seconds, and every block completed within 12 seconds including witness preparation. Competing provers have required large GPU clusters to reach comparable timings, so a single card changes who can realistically run a prover.
The prover is built on the open RISC-V instruction set and uses hash-based STARK technology, which ZKsync positions as resistant to quantum attack. It also ships with a standalone verifier, so results can be checked without trusting the team’s numbers.
Buterin spelled out what “formally verifying” now means: proving the arithmetization is equivalent to the Ethereum specification, setting a hard maximum bound on prover cost relative to the gas in a block, and verifying the security of the proof system itself.
The distinction matters because a benchmark can be gamed or measured loosely, while the Ethereum spec leaves no slack. A prover that is fast on a friendly workload still has to be shown to behave exactly like Ethereum’s rules on every input. That is the same reason zero-knowledge proofs are treated as a trust fix only when the proof system itself is sound.
Optimizing proof throughput is an engineering problem with clear, comparable numbers. Matching a mathematical proof to the Ethereum specification is slower, harder work that produces no headline benchmark. Quantum-resistance claims carry the same burden: saying a hash-based proof system is post-quantum ready is not the same as independently verifying its cryptographic assumptions hold.
ZKsync’s own recent trajectory makes the point. Before Airbender Prime, the team spent months hardening the security of its older EraVM, work that matters precisely because raw proving speed is no longer the bottleneck.
The next observable milestone is not a faster benchmark but formal-verification progress and independent scrutiny of the prover’s security. If the field treats Buterin’s checklist as the new bar, the center of gravity moves from proving speed to provable correctness — a slower race with a much higher bar.
