The seven-day withdrawal window was the price users paid for cheaper L2 execution.

Optimistic rollups made Ethereum transactions cheaper by replacing upfront validity proof generation with fraud challenges. Users gained lower fees and accepted delayed exits, weaker finality, and bridge liquidity as normal parts of the L2 experience.

ZK rollups offered a cleaner route because once a validity proof is generated, posted, and verified, withdrawals can move back to Ethereum without the same challenge period.

One of the important trade-offs here was cost, since proving large batches of transactions required expensive compute, so the market picked lower fees over faster exits and stronger trust assumptions, meaning better cryptography lost on price.

The Rollup Market Came With a Compromise

Optimistic rollups became the early default because they kept operating costs low and gave developers familiar EVM paths. On July 8, 2026, L2BEAT listed roughly $34 billion in rollup value secured, with Arbitrum One and Base among the largest Ethereum L2s. Both grew from the optimistic model, while the largest ZK systems remained smaller in value secured and user activity.

The trade-off was built into the model. Optimistic rollups assume batches are valid unless someone challenges them. Ethereum’s own docs describe an exit period of roughly seven days, and Optimism’s docs use the same one-week challenge period for state proposals. Users receive lower transaction costs while accepting delayed withdrawals and softer finality during the challenge window.

ZK rollups prove state transitions with validity proofs, and once a proof is verified on Ethereum, exits can complete without waiting through a fraud challenge period. This gives ZK rollups a stronger settlement model: verification replaces the fraud-challenge trust assumption rather than lowering trust across the board. Other trust vectors remain; sequencer control, upgrade keys, data availability, and bridge design can each still introduce trust regardless of the proof system. Proof generation has also historically carried a heavy cost.

Ethereum’s docs name proof generation and verification as part of the ZK fee model and describe validity proof generation as resource-intensive. They also warn specialized hardware requirements can concentrate prover operations among fewer actors.

The lesson from the past cycle is plain, since users and builders chose cheaper execution first while stronger guarantees came second.

Why Proving Became the Expensive Part

Zero-knowledge proofs compress trust into computation, and computation has a bill.

Proof generation for large transaction batches needs heavy parallel processing. GPUs became the default because they were available, programmable, and already deployed across compute markets. They also consume significant power, carry high capital costs, and compete with AI, gaming, and cloud demand.

FPGA chips offer a different path because field-programmable gate arrays can be reconfigured after manufacture to perform specific tasks extremely efficiently. In ZK proving, purpose-built FPGA hardware can improve energy efficiency rather than raw speed. Our Venus zkVM prover crossed a 1.3x proofs-per-joule milestone against an RTX 5090 GPU baseline, bringing energy per proof down to roughly 5.96 kJ from the GPU's roughly 7.76 kJ. The measured GPU baseline consumed about 7,760 J per proof at an efficiency of 0.000129 proofs per joule, and the FPGA path moved past roughly 0.000168 proofs per joule, an efficiency result, not a speed or total-cost result. The gain reflects the practical costs of a real prover system, including active power, data movement, and scheduling overhead, and shouldn't be read as universal speed or lower total cost of ownership.

Proving costs set the boundary for many ZK applications. Private stablecoin payments, identity proofs, on-chain games, verifiable AI, and micropayments all become harder when each proof costs too much to justify frequent use.

The missing piece has been open access. FPGA proving work has often been closed, attached to one prover network, or focused on narrow primitives. Rollup teams could use the service, but they had limited ability to audit the implementation, port it, benchmark it independently, or improve it in public.

A public blockchain cannot depend forever on private proving engines.

Open Source as an Infrastructure Requirement

ZK rollups are meant to reduce trust in operators. Closed proving systems create a new dependency in the place where trust should be smallest.

If a rollup depends on one prover, one hardware vendor, or one closed implementation, users inherit a hidden operational risk. The chain may verify proofs on Ethereum, but the machines generating those proofs can still become a private chokepoint.

With open-source proving code, rollup teams can run more of their own proving, researchers can inspect the implementation, independent operators can build new proving services, and hardware teams can improve acceleration designs. The prover market becomes easier to test, compare, and decentralize.

This is what trustless execution requires at the machine level. Open smart contracts are only one part of the system. Open proof generation becomes just as important once ZK rollups carry payments, identity, games, AI verification, and consumer transactions. ZK only reaches its potential when proving is fast, cheap, and verifiable by anyone.

Open-sourcing the first FPGA zkVM shows the ecosystem can develop together rather than behind closed doors.

The Cost Equation Is Being Rewritten

The rollup race is as much of a fee race as it is a security race.

Optimistic rollups won early because they delivered cheaper execution quickly. ZK rollups carried stronger technical guarantees, but proof generation made them harder to operate at the same cost. The user experience also suffered because markets grew around bridges, liquidity workarounds, and waiting periods.

Cheaper proving would narrow one of the cost gaps between ZK and optimistic rollups, but it wouldn't settle the fee comparison on its own. Proving is only one line in rollup economics, data availability, Ethereum settlement, sequencing, batch size and utilization, hardware capex, and engineering overhead all feed into per-transaction cost. If those other variables are comparable, falling proving costs could let ZK rollups compete on fees while retaining their faster finality and different trust assumptions.

The effect could reach far beyond rollup withdrawals:

• Private stablecoin payments need low-cost proofs so users and businesses can prove compliance conditions without exposing complete payment histories;
• Identity verification needs affordable proofs so people can prove attributes without handing over full documents;
• On-chain games need cheap state proofs so gameplay can involve frequent verifiable actions;
• Micropayments need proof costs small enough to avoid eating the payment itself.

These categories have been discussed for years because the demand exists. They remained constrained because proof generation made frequent use uneconomical.

Lower-cost FPGA proving loosens the constraint. Open-source code accelerates the competitive response. More operators, more hardware routes, and more public benchmarking should put pressure on proof prices over time.

The Seven-Day Window Was Always a Compromise

The fraud-proof window gave Ethereum cheaper execution during the first major L2 phase, helped rollups attract liquidity, gave developers usable blockspace, and made L2 applications cheaper for users, yet it remained a compromise.

ZK rollups have always offered a compelling endgame for verifiable computation, though optimistic and hybrid designs still carry their own trade-offs in compatibility, cost, governance, and operational complexity. The open question was whether proving could become cheap enough, fast enough, and open enough for the market to treat ZK as the default path rather than the premium path.

The future of this market will be defined by open proving markets, cheaper validity proofs, and hardware built to make stronger guarantees cost less than weaker assumptions.