Pay-Per-Proof vs. Subscription: The Prover Revenue War

Fixed-fee models like those from RISC Zero or Succinct Labs create misaligned incentives. Provers are paid the same for a trivial proof as for a complex one, leading to:

• Inefficient capital allocation and unpredictable costs for high-volume protocols.
• Centralization pressure as only large, capitalized provers can absorb the risk of variable compute costs.
• A hidden tax on innovation, discouraging novel, compute-heavy ZK applications.

Auction-based models, pioneered by Espresso Systems with Tiramisu and seen in Polygon zkEVM, create a true spot market for compute. This aligns cost with actual resource consumption.

• Dynamic pricing ensures users pay for what they use, no more, no less.
• Prover competition drives down costs and fosters a decentralized network of specialized hardware.
• Enables long-tail applications by making sporadic, complex proofs economically viable to produce.

The end-state isn't a single winner. The market will stratify, mirroring cloud computing. Celestia-inspired data availability layers will need different prover economics than an EigenLayer AVS.

• Subscription for baseline throughput (e.g., L2 sequencers).
• Pay-per-proof for burst compute (e.g., on-chain gaming, privacy bridges).
• Specialized ASIC/GPU networks will emerge, creating verticalized prover markets akin to Render Network for ZK.

Pay-per-proof models like EigenLayer AVS and AltLayer create a cutthroat spot market. Provers undercut each other to win work, driving fees to marginal cost. This creates a liquidity trap where only the largest, most efficient provers survive, centralizing the network.

• Winner-Take-All Dynamics: Lowest bidder wins the entire proof batch.
• Unsustainable for Small Players: High fixed costs (hardware, R&D) cannot be covered at rock-bottom prices.
• Security Risk: Ultra-thin margins disincentivize honest participation, making collusion or exit more likely.

Subscription models, as seen in RISC Zero B2B deals, offer predictable revenue but create vendor lock-in and stagnation. Long-term contracts protect incumbents and reduce the competitive pressure to improve performance or lower costs.

• Barrier to New Entrants: New prover tech cannot displace an entrenched subscription base.
• Hidden Costs: Protocols pay for idle capacity and lack flexibility during low-usage periods.
• Innovation Tax: Revenue certainty removes the incentive for provers to aggressively optimize hardware or algorithms.

Both models fail when proof demand is volatile. In pay-per-proof, provers face idle capital during troughs. In subscriptions, protocols overpay during peaks. This mismatch leads to inefficient resource allocation and systemic fragility.

• Boom-Bust Cycles: Prover hardware is either overwhelmed or sitting unused.
• Slashing Ineffectiveness: Penalties cannot compensate for revenue lost during sustained low demand.
• Protocol Risk: Reliance on a small set of provers who may exit the market during a downturn.

The fix is a two-tiered market combining base-layer subscriptions with a spot auction for excess capacity. Projects like Espresso Systems with time-based sequencing hint at this future. Proof futures could let protocols hedge cost and provers hedge revenue.

• Guaranteed Base + Competitive Edge: Protocols reserve core capacity, provers auction surplus.
• Derivative Markets: Financial instruments to smooth out revenue/cost volatility.
• Dynamic Allocation: Automated systems like EigenLayer restaking could route work based on real-time price and performance.

Subscription models lock capital into a single prover network, creating inefficiency. Builders overpay for unused capacity while provers fight for a finite pool of staked capital.

• Capital Inefficiency: Staked capital sits idle, earning zero yield when not actively proving.
• Fragmented Security: Each network (e.g., EigenDA, Avail) must bootstrap its own staking pool, diluting security.
• Builder Lock-in: Switching provver networks requires unbonding periods and re-staking, creating friction.

Decouples payment from staking. Anyone with a GPU can become a prover and get paid per job, creating a liquid spot market for compute.

• Capital Efficiency: Stakers can provide security to multiple networks simultaneously (e.g., via EigenLayer).
• True Competition: Provers compete on cost and latency, driving prices toward marginal cost.
• Builder Flexibility: Developers can route proofs to the cheapest/fastest prover instantly, no lock-in.
• Representative Projects: Risc Zero, Succinct, Lumoz.

Subscription models argue that committed capital creates superior security and service guarantees, appealing to high-value applications.

• Sybil Resistance: A $1B+ staked pool is harder to attack than a pay-per-proof network with minimal skin-in-the-game.
• Predictable Costs & SLAs: Enterprises and L2s (e.g., Arbitrum, zkSync) prefer fixed monthly costs and guaranteed uptime.
• Long-Term Alignment: Stakers are incentivized to maintain network health over years, not just for the next proof job.
• Representative Projects: Polygon zkEVM, Scroll, Starknet.

The market will bifurcate. High-security, high-TVL applications will pay premiums for subscriptions. Cost-sensitive, bursty workloads will dominate pay-per-proof.

• Market Split: ~70% of TVL in subscription models, ~80% of proof volume in pay-per-proof by 2026.
• Hybrid Emergence: Look for subscription networks to offer spot markets for excess capacity, and pay-per-proof networks to offer staking slashing for high-value clients.
• Investor Takeaway: Bet on protocols that enable the capital layer (EigenLayer) and the middleware that routes demand (Hyperliquid, Espresso).