User experience is subsidized. The low fees for a swap on zkSync or Starknet are an illusion, paid for by the protocol's prover infrastructure costs. The real expense is in generating the zero-knowledge proof that batches thousands of transactions.
the-ethereum-roadmap-merge-surge-verge
Blog
ZK Rollups and Prover Infrastructure Costs
A cynical breakdown of the hidden, unsustainable costs of generating zero-knowledge proofs. The promise of cheap L2s is being undermined by a prover market that's becoming a centralized, capital-intensive arms race.
introduction
THE COST PARADOX
The ZK Mirage: Cheap Txs, Expensive Proofs
ZK rollups promise cheap user transactions but shift massive computational and capital costs to specialized proving infrastructure.
Proof generation is capital-intensive. Running a prover for zkEVMs like Scroll or Polygon zkEVM requires high-end GPUs or ASICs and significant ETH staking. This creates a centralizing force contrary to decentralization narratives, concentrating power with a few specialized operators.
Prover markets are nascent. Projects like Risc Zero and Succinct are building generalized proof markets, but today's prover economics are opaque. The cost to prove a block is not a simple linear function of transactions, creating unpredictable operational overhead.
Evidence: A single zkEVM proof generation can cost $0.50-$2.00 in compute, meaning a rollup must batch thousands of sub-$0.01 transactions to achieve net profitability, a fragile economic model.
key-trends
SCALABILITY, COST, DECENTRALIZATION
The Prover Cost Trilemma
ZK-Rollups face a fundamental trade-off: you can only optimize for two of three critical attributes in prover infrastructure.
01
The Centralization Trap: High-Performance Provers
To achieve low latency and low cost, projects like zkSync and Starknet initially rely on centralized, high-spec provers. This creates a single point of failure and control, undermining the core blockchain value proposition.
- Risk: Censorship and liveness failure.
- Reality: ~10-30 second proving times, but controlled by a single entity.
1 Entity
Centralized Risk
<30s
Prove Time
02
The Cost Wall: Decentralized & Fast
Aiming for decentralization and speed requires a competitive network of provers, like RiscZero's Bonsai or Espresso's decentralized sequencer. This inflates costs as you pay for redundant compute and coordination overhead, pricing out small users.
- Result: Prover costs dominate transaction fees.
- Example: Ethereum L1 data availability becomes the cheaper part of the stack.
$0.50+
Est. TX Cost
High
Overhead
03
The Latency Penalty: Decentralized & Cheap
Optimizing for low cost and decentralization means using slower, commodity hardware in a distributed network. This leads to high finality latency, breaking UX for DeFi and gaming. Polygon zkEVM's planned decentralized prover network faces this exact challenge.
- Trade-off: Finality takes minutes, not seconds.
- Use Case: Suitable only for non-interactive batches.
2min+
Finality Time
Low
Hardware Bar
04
The Escape Hatch: Proof Aggregation
Projects like Polygon AggLayer, Avail's Nexus, and zkBridge protocols attack the trilemma by amortizing cost across many chains. A single aggregated proof secures multiple states, diluting cost per chain while maintaining security and decentralization.
- Mechanism: Recursive proofs or proof of proofs.
- Outcome: Enables viable sovereign rollups.
>10x
Cost Amortization
Multi-Chain
Security Scope
05
The Hardware Endgame: ASICs & GPUs
Specialized hardware from Cysic, Ingonyama, and Accseal provides a 10-1000x efficiency gain for specific proof systems (e.g., STARKs, Halo2). This pushes the cost curve down, making decentralized, fast proving economically feasible.
- Catch: Creates hardware centralization and vendor lock-in risks.
- Target: Reduce prover cost to <$0.01 per transaction.
1000x
Speedup
<$0.01
Cost Target
06
The Market Solution: Proof Auctions
Decentralized prover markets, pioneered by Espresso Systems and Georli, treat proof generation as a commodity. Rollups post proof jobs; a permissionless network of provers bids to compute them, creating a cost-discovery mechanism that balances all three trilemma points dynamically.
- Efficiency: Market forces optimize for price and speed.
- Future: Likely model for Ethereum's native zk-rollups.
Dynamic
Pricing
Permissionless
Network
deep-dive
THE COST BREAKDOWN
Anatomy of a Prover Bill: Where the Money Goes
A prover's operational costs are dominated by hardware and electricity, not algorithm complexity.
Hardware is the primary cost driver. Proving a ZK-SNARK batch requires specialized, high-memory servers. The cost of a single AWS r6i.metal instance for a prover like zkSync Era or Polygon zkEVM exceeds $4 per hour, dwarfing software licensing fees.
Electricity consumption scales linearly with compute. Unlike general compute, ZK proving is a fixed, intensive workload. A prover for Starknet or a custom RISC Zero circuit draws constant high wattage, making energy cost the second-largest line item after hardware rental.
Prover marketplaces like RiscZero and Ulvetanna create price competition. They allow rollups to auction proof generation, commoditizing hardware costs. This model pressures in-house prover teams to justify their capital expenditure versus spot market rates.
Evidence: A zkSync Era prover cluster requires 512 GB of RAM and 32 vCPUs. At AWS on-demand rates, this hardware alone costs over $100,000 annually before a single proof is generated.
ZK-ROLLUP PROVER ECONOMICS
Prover Infrastructure Cost Matrix: zkSync vs. Starknet vs. Scroll
A direct comparison of capital and operational costs for running a prover node across leading ZK-Rollups, critical for protocol architects evaluating infrastructure overhead.
| Cost & Performance Metric | zkSync Era | Starknet | Scroll |
|---|---|---|---|
Prover Hardware Requirement | High-End Consumer GPU (e.g., RTX 4090) | Custom Prover Service (SHARP) or High-CPU Server | Consumer GPU (e.g., RTX 3080) or Cloud Instance |
Prover Execution Time (L2 Block) | 3-5 minutes | ~15-30 minutes (SHARP batch) | 5-10 minutes |
Estimated Prover Cost per Tx (Current) | $0.10 - $0.30 | $0.02 - $0.05 (shared via SHARP) | $0.15 - $0.40 |
Prover Decentralization Model | Permissioned Prover Network (zkSync Boojum) | Centralized Prover (SHARP) → Permissioned Future | Centralized Prover → Decentralized Roadmap |
Proof System | SNARK (Boojum / PLONK) | STARK | SNARK (Scroll zkEVM / Halo2) |
Prover Software Open Source | |||
Requires Trusted Setup | Powers of Tau Ceremony (Universal) | Powers of Tau Ceremony (Application-Specific) | |
L1 Verification Gas Cost (Avg.) | ~400k gas | ~600k gas | ~450k gas |
counter-argument
THE HARDWARE TRAP
The Optimistic Rebuttal: It Gets Cheaper, Right?
The long-term cost trajectory for ZK provers is constrained by hardware, not just software.
Prover costs are hardware-bound. The asymptotic efficiency gains from recursive proofs and folding schemes like Nova and Plonky2 face a physical limit. The prover's computational work must be executed on real silicon, where performance-per-dollar improvements follow Moore's Law, not exponential ZK theory.
Specialization creates centralization pressure. High-performance proving for chains like zkSync and Starknet requires custom hardware (ASICs/FPGAs). This creates an infrastructure moat for firms like Ulvetanna and Ingonyama, moving cost efficiency from decentralized software to centralized capital expenditure.
The cost floor is data availability. Even with a zero-cost proof, posting the state diffs and calldata to Ethereum L1 remains the dominant expense. This anchors ZK rollup transaction costs to Ethereum's base layer data pricing, a bottleneck shared with Optimistic Rollups.
risk-analysis
ZK PROVER ECONOMICS
The Centralization Endgame: Four Scenarios
The cost to prove a block is the ultimate bottleneck for ZK rollup decentralization. We model the economic forces shaping the prover market.
01
The Commodity Hardware Trap
The dream of permissionless proving on consumer GPUs is economically non-viable for high-throughput chains. Specialized hardware (ASICs, FPGAs) delivers 100-1000x better $/proof efficiency. This creates a natural oligopoly of capital-intensive proving farms, mirroring Bitcoin mining.
- Cost to Prove a Block: Ranges from $0.50 to $50+ depending on TPS and circuit complexity.
- Market Structure: Leads to 3-5 dominant proving pools controlling >80% of security.
- Risk: Recreates the L1 validator centralization problem one layer up.
100-1000x
Efficiency Gain
$0.50-$50+
Proof Cost
02
The Prover-as-a-Service Cartel
Rollup teams outsource proving to centralized services like =nil; Foundation, Ulvetanna, or Geohot's zkSharding to avoid capex. This trades technical decentralization for operational centralization, creating systemic risk.
- Dominant Model: Adopted by Starknet (via Madara), zkSync, and most new ZK L2s.
- Key Risk: Single prover failure = chain halt. Creates a trusted setup for liveness.
- Mitigation: Requires a robust multi-prover network with slashing for downtime.
>70%
Of New L2s
1->Many
Failure Point
03
Decentralized Prover Networks (Espresso, Brevis, Lagrange)
A coordination layer that auctions proof generation to a permissionless set of nodes. This is the only path to credible decentralization but faces massive coordination and economic challenges.
- Economic Flywheel: Requires sustainable $PROVER token rewards funded by sequencer fees.
- Latency Trade-off: Adds ~2-10 seconds of proving latency vs. a centralized service.
- Verdict: Likely only viable for high-fee, high-value chains (e.g., future EigenLayer AVS).
+2-10s
Latency Penalty
Token-Driven
Incentive Model
04
The Shared Sequencer/Prover Hybrid (Fuel, Eclipse)
Bundles sequencing and proving into a single decentralized network. Uses a unified token to secure both data ordering and proof generation, maximizing capital efficiency.
- Synergy: The sequencer selects the winning prover, internalizing the market.
- Example: Fuel's beta-3 testnet uses a PoS network for both roles.
- Endgame: Could enable truly sovereign rollups that rent security from a shared physical infrastructure layer.
Unified
Security Model
Max
Capital Eff.
future-outlook
THE COST CURVE
The Verge is Coming, But Not for Provers
Zero-knowledge rollup scaling will shift computational burden to users, but the prover market will consolidate into a capital-intensive oligopoly.
User-proven computation is inevitable. The Verge, a core component of the Ethereum roadmap, moves ZK proof generation to the client side. This reduces L1 data costs but transfers the proving overhead to end-users or specialized services.
Prover hardware is the moat. Efficient proving requires specialized hardware like GPUs, FPGAs, and eventually ASICs. This creates a capital-intensive barrier that startups cannot cross, leading to market consolidation around firms like Ulvetanna and Ingonyama.
The prover market centralizes. The economics favor a few large, optimized proving farms, similar to Bitcoin mining pools. Decentralized prover networks like RiscZero and Succinct must compete on cost efficiency with these industrial-scale operators.
Evidence: A single ZK-SNARK proof for a complex transaction can cost $0.50-$5.00 in compute. At scale, this creates a multi-billion dollar market where only the most efficient hardware operators survive.
takeaways
ZK-ROLLUP PROVER ECONOMICS
TL;DR for Protocol Architects
The cost of proving is the primary bottleneck for ZK-rollup scalability and decentralization. Here's the breakdown.
01
The Prover Cost Wall
Generating a ZK proof is computationally intensive, creating a centralizing force and a direct cost passed to users. This is the core scaling tax.
- Dominates L2 transaction fees (often >50% of total cost).
- Creates high hardware barriers for decentralized prover networks.
- Limits proving throughput, capping TPS.
>50%
Of Tx Fee
~$0.10+
Prove Cost
02
Parallel Proving & Hardware Acceleration
The solution is massive parallelism and specialized hardware (GPUs, ASICs) to amortize cost. Think zkVM architectures like zkSync, Starknet, and Polygon zkEVM.
- GPU clusters slash proving times from minutes to seconds.
- Custom ASICs (e.g., Cysic, Ingonyama) target ~100x efficiency gains.
- Enables sub-second proof times for viable decentralized sequencing.
~100x
Efficiency Gain
<1s
Target Proof Time
03
Proof Aggregation & Shared Sequencing
Amortize fixed proving overhead across many transactions and even across different rollups. This is the network effect play.
- Aggregators (e.g., Espresso, Astria) batch proofs from multiple rollups.
- Shared sequencers enable cross-rollup atomic bundles.
- EigenLayer restakers could back decentralized proving networks, reducing capital cost.
10-100x
Amortization
Multi-Rollup
Scope
04
The Endgame: Prover Markets
The logical conclusion is a commoditized proving market, separating proof generation from sequencing. Rollups become proof consumers.
- Proof-as-a-Service providers compete on cost and latency.
- zkRollups (like Taiko) already architect for this.
- Drives cost toward marginal electricity + hardware depreciation.
~$0.01
Long-Term Cost/Tx
Commodity
Market
ENQUIRY
Get In Touch
today.
Our experts will offer a free quote and a 30min call to discuss your project.
NDA Protected
Confidentiality24h Response
Time to ValueDirectly to Engineering Team
No Sales Fluff10+
Protocols Shipped
$20M+
TVL Overall