Ethereum's proving infrastructure crossed a threshold in 2026 that would have been dismissed as speculative two years ago. Multiple independent teams — Succinct (SP1 Hypercube), RISC Zero (R0VM 2.0), and ZisK — now generate zero-knowledge proofs of full Ethereum L1 blocks in under 12 seconds on c...
"The trilemma has been solved – not on paper, but with live running code, of which one half (data availability sampling) is on mainnet today, and the other half (ZK-EVMs) is production-quality on performance today – safety is what remains." — Vitalik Buterin, Ethereum Co-Founder
Ethereum's proving infrastructure crossed a threshold in 2026 that would have been dismissed as speculative two years ago. Multiple independent teams — Succinct (SP1 Hypercube), RISC Zero (R0VM 2.0), and ZisK — now generate zero-knowledge proofs of full Ethereum L1 blocks in under 12 seconds on consumer-grade GPU hardware. The Ethereum Foundation's ethproofs tracker logged sub-half-cent proving costs for eligible configurations on September 2, 2026, down from $1.69 per block proof in January 2025.
The practical consequence is EIP-8025 (Optional Execution Proofs), a backwards-compatible specification that introduces "zkAttesters" — validators that verify cryptographic proofs instead of re-executing every transaction. The proposal requires no hard fork and is targeted for inclusion in the Hegotá upgrade (2027). If activated, it would eliminate the requirement for validators to store execution-layer state, reducing hardware costs and potentially expanding the validator set. The Ethereum Foundation mandated 128-bit provable security as the minimum standard for any zkEVM proof system used on mainnet, setting a May 2026 interim target of 100-bit and a year-end deadline of full 128-bit with sub-300 KiB proofs.
The economic implications are significant. Proving competes for the same GPU silicon as AI training, creating a structural cost floor tied to AI compute demand. Three proving teams have converged on NVIDIA RTX 5090 clusters as the hardware baseline, and the Foundation set a $100,000 ceiling for proving rigs to preserve decentralization.
The trajectory of Ethereum block proving times follows a power-law cost curve. In early 2025, proving a single Ethereum block took roughly 16 minutes. By December 2025, RISC Zero's R0VM 2.0 had compressed that to 44 seconds. Succinct's SP1 Hypercube, released in 2026, achieved a 10.3-second average per Ethereum block on a 16x RTX 5090 rig, with 99.7% of blocks completing under 12 seconds and 95.4% under 10 seconds.
ZisK, the open-source zkVM spun off from Polygon by Circom creator Jordi Baylina, posted a 9.62-second p99 on just four RTX 5090 GPUs on August 18, 2026, claiming 99.7% of tested blocks finished under 10 seconds. The claim carries caveats: ZisK has not disclosed the block range, sample size, proof size, or whole-system power consumption for the run. As of September 2026, it remains an unverified benchmark.
The cost decline is equally steep. According to the ethproofs tracker, average proving cost per block fell from $1.69 in January 2025 to under $0.04 by December 2025. On September 2, 2026, the live cohort showed roughly half-cent cost fields for its two eligible configurations. However, only 39.7% of all evaluated block slots achieved sub-10-second success, because offline and ineligible proving runs are counted in the denominator.
EIP-8025 (Optional Execution Proofs) is the consensus-layer specification that translates proving capability into protocol architecture. Published by the Ethereum Foundation's zkEVM team, it introduces two new opt-in node roles:
Provers (proof-generating mode) altruistically produce zkEVM proofs for each block's execution and gossip them over a dedicated P2P network.
zkAttesters (proof-verifying mode, also called stateless attesters) verify received zkEVM proofs and attest on that basis, rather than running a full execution client.
The specification requires a 3-of-5 threshold: zkAttesters must verify three out of five independent proofs — produced by different proving systems — before accepting a block's execution as valid. This multi-prover requirement enforces implementation diversity.
The practical consequence for zkAttesters is significant: they do not need to hold execution-layer state, they do not need to sync the full execution-layer chain, and they do not need to process every transaction. According to the EIP specification, "A zkAttester does not need to hold EL state. It does not need to sync the full execution layer chain." Verification becomes faster and lighter on hardware, potentially lowering the barrier for solo stakers and home validators.
EIP-8025 is backwards compatible and does not require a hard fork. It is targeted for the Hegotá upgrade, tentatively scheduled for 2027.
The Ethereum Foundation is coordinating six sub-projects across the proving stack, ensuring compatibility between multiple independent proving systems and Ethereum's various execution clients. Three proving teams dominate the current landscape:
Succinct (SP1 Hypercube): The first zkVM built entirely on a multilinear polynomial-based proof system, eliminating reliance on proximity gap conjectures. SP1 Hypercube claims formal verification of all 62 core RISC-V opcodes, completed with Nethermind Security and the Ethereum Foundation. A proving cluster capable of real-time proving for over 90% of mainnet blocks requires approximately 160 RTX 4090 GPUs and can be built for $300,000-$400,000, or approximately $100,000 using 16 RTX 5090 GPUs. SP1 Hypercube is integrated into the Succinct Prover Network and is live on mainnet.
RISC Zero (R0VM 2.0): Reduced Ethereum block proving from 35 minutes to 44 seconds. RISC Zero launched the Boundless decentralized proving network on mainnet in September 2025, then terminated its hosted proof service that December, forcing all generation through the decentralized network. RISC Zero targets a proving cluster buildable for approximately $120,000. Its code and GPU orchestration software are 100% open-source.
ZisK: Founded by Jordi Baylina after spinning off from Polygon in June 2025. ZisK's v1.1.0-alpha prover claims 9.62-second p99 on four RTX 5090 GPUs with 128-bit security and post-quantum resistance. The project has yet to publish detailed methodology for its benchmark claims. Baylina called the result "a milestone combining four-GPU proving with claimed 128-bit security and post-quantum resistance."
All three teams are building toward compatibility with the ethproofs benchmark standard set by the Ethereum Foundation.
On December 18, 2025, the Ethereum Foundation announced a pivot from performance optimization to security hardening. Having declared the real-time proving goal effectively met, the Foundation published a 2026 roadmap centered on provable security guarantees.
The mandate requires all zkEVM teams to:
The Foundation warned that "security remains the elephant in the room," noting that many STARK-based zkEVMs rely on unproven mathematical conjectures — specifically proximity gap conjectures — that recent research has begun to disprove. The 128-bit standard aligns with recommendations from leading cryptographic standardization bodies and is positioned as the minimum for institutional adoption.
SP1 Hypercube claims to be the first general-purpose hash-based zkVM to completely eliminate dependence on proximity gap conjectures. ZisK claims 128-bit security but has not published supporting documentation. RISC Zero's security posture relative to the new standard has not been publicly detailed as of September 2026.
A structural tension underlies the entire proving ecosystem: zero-knowledge proving competes with AI model training for the same GPU hardware. According to Leo Fan, CEO of hardware proving company Cysic, "We're bidding against trillion-dollar data centre budgets for the same silicon."
By mid-2025, nearly every proving team converged on NVIDIA RTX 5090 GPUs as the globally available, consumer-grade, high-parallelism baseline. The Ethereum Foundation set explicit decentralization constraints for proving infrastructure:
These constraints are designed to prevent proving from concentrating in the hands of entities with access to AI-scale compute budgets. But the tension is real: as AI demand for GPUs intensifies, the cost floor for ZK proving hardware rises correspondingly. Cysic open-sourced its Venus multi-GPU ZK proving stack in April 2026, arguing that the hardware layer must be opened up rather than left to proprietary provers.
The economic question for the Ethereum ecosystem is whether altruistic proving — the model assumed by EIP-8025 — can sustain itself when GPU hours have a high opportunity cost in AI markets. At roughly half a cent per proof, the revenue opportunity for provers is marginal. The Ethereum Foundation has not yet published a sustainability model for proving incentives.
The ethproofs platform, maintained by the Ethereum Foundation, serves as the neutral benchmark for the proving ecosystem. It provides a proof explorer, dashboards, and open schemas that allow independent measurement of proving performance across teams and configurations.
The September 2, 2026, snapshot revealed the current state of production readiness:
The ethproofs data underscores a point often obscured in proving-team press releases: real-time proving under controlled conditions is not the same as real-time proving under production conditions across all block types and network states.
The path from current proving capability to protocol integration follows a multi-year sequence:
2026 (Current):
2027 (Hegotá Upgrade):
2027-2030 (Vitalik's Timeline):
Dependencies are sequential: 128-bit security must be demonstrated before EIP-8025 can be safely activated, Glamsterdam's ePBS must ship to provide the timing infrastructure, and Hegotá must finalize its scope to incorporate EIP-8025. Any delay in the Glamsterdam pipeline — and Devnet-9 failed to finalize on September 1, 2026 — pushes the entire chain forward.
The technical achievement is real: proving an Ethereum block in under 12 seconds for less than a cent represents a 99.7% cost reduction in under two years. Three independent teams have demonstrated the capability, and the Ethereum Foundation has published both the security standards and the protocol specification (EIP-8025) to translate proving into validator architecture.
The unresolved questions are economic and operational. Production reliability — measured by the ethproofs cohort — lags headline benchmarks by a wide margin. The 128-bit security mandate may force proving-time regressions as teams strengthen their cryptographic foundations. GPU competition with AI creates uncertainty around hardware costs. And the absence of a proving incentive model leaves the economic sustainability of the system undefined.
The comparison to the Merge is apt in one respect: the technical work is further along than the coordination work. Ethereum proved it could switch consensus mechanisms in 2022. Whether it can build a decentralized, economically sustainable proving market by 2027-2030 is the open question that EIP-8025 poses but does not answer.