Every computation that matters on-chain is on a path toward being proven rather than re-executed. Rollups prove state transitions. Bridges prove consensus. Applications increasingly prove off-chain work, from machine learning inference to solvency attestations. Each proof requires substantial computation to generate, and someone has to perform that computation, at a price. A market is forming around that work, and it is worth analyzing with the tools we would apply to any young commodity market, because that is what it is becoming.
The anatomy of a proving trade
A proving job has three economic properties worth noting. It is verifiable by construction: the buyer can check the output cheaply and objectively, which eliminates most of the trust problems that plague other outsourced-compute markets. It is latency-sensitive in tiers: a rollup posting hourly can shop for price, while an exchange settling in seconds cannot. And it is hardware-elastic: the same proof can be generated on a laptop slowly, or on specialized hardware quickly, at very different unit costs.
Verifiability plus elasticity is the classic recipe for commoditization. When buyers can switch suppliers without trust and suppliers can enter with capital alone, margins compress toward the cost of hardware and electricity. We are watching that compression happen in real time: proving costs for standard workloads have fallen by more than an order of magnitude since we first modeled the space, driven by better provers, better proof systems, and simple competition.
Proving is on its way to becoming the electricity of the verifiable internet: indispensable, enormous in aggregate, and priced at the margin like a utility.
Where margins survive commoditization
Commodity markets still produce excellent businesses; they are just found at specific points in the chain. We see four such points here. First, the proof systems themselves, where a performance edge translates into a durable cost advantage for whoever integrates it. Second, specialized hardware and acceleration, the picks and shovels of any compute commodity. Third, aggregation: batching many small proofs into one on-chain verification amortizes the fixed cost that dominates small workloads, and aggregators sit at a natural toll point. Fourth, the marketplace layer that matches jobs to provers and nets exposure, which in mature commodity markets is where the best risk-adjusted economics tend to live.
The demand question
The bear case is that proving supply is scaling faster than proof demand. Today that is true. Our view is that demand curves for cheap verifiable compute behave like demand curves for cheap bandwidth: each order-of-magnitude cost decline unlocks applications that were previously absurd. Proving every bridge message was absurd at 2023 prices. Proving ML inference is borderline today. Proving ordinary web services' claims about their own behavior is absurd at current prices, and will not remain so. We are underwriting the category on that trajectory, positioned at the toll points rather than the treadmill.