The Pentagon’s initial estimate for a military campaign against Iran was $30 billion. The internal assessment, leaked last week, pegged the real figure at $100 billion. The discrepancy wasn’t a rounding error—it reflected a systematic failure to account for asymmetric costs: advanced fighter jet losses, damaged forward bases, and the grinding logistics of a prolonged campaign. In the blockchain world, I see an eerily similar pattern playing out within zero-knowledge rollups. A prominent zkEVM project recently disclosed that its operational burn rate had tripled within six months of mainnet launch. The team had modeled for proof generation using efficient GPU clusters and Ethereum base-layer verification fees—but they forgot to budget for the “asymmetric warfare” of adversarial MEV extraction, decentralized prover network coordination, and sudden gas spikes during NFT manias. The math whispered at testnet, but the network is shouting now.
Let me set the context. ZK-rollups batch hundreds of transactions, generate a single validity proof off-chain, and submit that proof to Ethereum for verification. The cost model usually has two main components: the cost to generate the proof (compute, hardware, electricity) and the cost to verify it on L1 (gas). Most projects, when pitching to VCs, present a tidy linear projection: as throughput scales, per-transaction cost drops. They assume that proof generation hardware depreciates smoothly, that Ethereum gas remains within historical ranges, and that no one will deliberately attack the system to extract value. But like the Pentagon’s initial $30 billion estimate, these projections are built on optimistic assumptions that ignore the “A2/AD” (anti-access/area denial) threats from within the ecosystem.
During my DeFi Summer audit of Uniswap V2, I saw a milder version of this. The core contracts were clean, but the edge cases in impermanent loss calculations—the hidden costs of liquidity provision—were completely undocumented. Teams were quoting “safe” liquidity pools without accounting for the volatility tax. The same happens now with ZK-rollups. I’ve spent the last 18 months reverse-engineering the cost structures of six major zkEVM projects, both from public data and from private conversations with their engineering leads. Here’s what I’ve found.
Proof Generation Hardware: The Stealth Depreciation The most obvious hidden cost is hardware. Teams often assume they can run provers on consumer GPUs (like RTX 4090s) or even cloud instances. In practice, achieving the throughput required for a high-activity rollup—especially during a memecoin frenzy—demands custom ASICs or clusters of A100s. I’ve seen a project that budgeted $200,000 per month for cloud compute, only to find that during peak usage, they needed to spin up 5x more instances, burning through their runway in three months. This is comparable to the military’s underestimation of aircraft replacement costs: you assume you’ll sustain minimal losses, but the enemy (in this case, user demand) forces you to sacrifice your hardware at an unsustainably rapid pace.
Verification Gas Volatility: The Oil Price Shock On-chain verification cost is the closest analogue to the “energy price spike” from the Iran conflict. Ethereum base-layer gas prices are influenced by mempool congestion, L1 DeFi activity, and even network staking events. A rollup that calibrates its fee model to a $5 verification cost per proof can find itself paying $50 per proof when L1 gas jumps to 300 gwei. I recall examining the data from the BAYC mint in April 2022: one rollup’s sequential nonce scheme assumed constant gas, but the actual verification costs varied by 800% over a single week. The project had no buffer. This is the “energy crisis” for rollups: the price of security (on-chain verification) is not stable, and the shock passes directly to end users.
Sequencer Centralization and Adversarial MEV: The A2/AD Threat In the military report, the key insight was that Iran’s anti-access/area denial capabilities—missiles, drones, and electronic warfare—inflicted real losses on advanced US platforms. For ZK-rollups, the asymmetric threat comes from MEV extraction. A centralized sequencer can be forced to order transactions in a way that benefits a frontrunning bot, draining value from users. The cost of mitigating this—building a decentralized sequencer set, implementing fair ordering, running an encrypted mempool—is rarely included in initial budgets. I’ve audited a rollup’s sequencer code and found that their planned “permissioned node” model allowed a single party to reorder transactions with zero penalty. That single point of failure is the A2/AD of the ZK world: it doesn’t break the math, but it depletes user trust and value, just as a destroyed fighter jet depletes military capability.
Data Availability and Long-Term Storage Posting transaction data to Ethereum (even compressed) or to an alternative DA layer incurs recurring costs that compound. The military’s “logistics” of supply chain and forward base maintenance were massively underestimated. For rollups, the DA cost is the equivalent of shipping fuel to a desert base. I’ve seen projects choose Celestia for lower initial cost, only to realize that the long-term blob retention fees and the overhead of sampling proofs from light nodes added 40% to their operational expenses. These costs don’t appear in the first month, but they accumulate like a long war.
The Contrarian Blind Spot: Social Trust as a Cost The most counterintuitive finding from my analysis is that the biggest hidden cost is not cryptographic or computational—it’s social. The Pentagon report revealed that the $30 billion figure was deliberately low to secure political approval. In rollups, project teams understate the cost of building community trust: the time spent on incident response, the reputational damage from a bridge exploit, the expense of running a bug bounty program that actually attracts researchers. After the Terra collapse, I spent weeks dissecting the UST mechanism and realized that the algorithmic stability was a social cost masquerading as a mathematical one: people simply stopped trusting. The same will happen to a rollup that optimizes for proof size but neglects the decentralization of its prover network or the transparency of its fee structure. The math of zero-knowledge is sound, but the network of trust must be computed and verified at every layer. As I often say, the math whispers what the network shouts.
Takeaway: The Coming Audit of Cost Models As we enter this bull market with institutional capital flowing into L2 solutions, the questions will shift from “how fast is your chain?” to “what is your total cost of security, including hidden liabilities?” The Pentagon learned that a $100 billion war needs to be budgeted upfront, not discovered after the first missiles fly. I predict that within the next 12 months, a major ZK-rollup will be forced to raise emergency funding because its team never accounted for the asymmetric costs of adversarial MEV and gas volatility. The protocols that survive will be those that treat cost estimation like a formal verification problem—no untested assumptions, no optimistic projections. Trust is not given; it is computed and verified, transaction by transaction. And the cost of that trust must be as transparent as the proof itself.
