The Hidden Bleed: Why ZK Rollups Are Losing Money on Every Transaction

I remember the exact moment I stopped believing the ZK rollup fairy tale. It was a Tuesday afternoon in April, not long after the Dencun upgrade had supposedly given layer twos everything they needed. I was sitting in a coworking space in Stockholm, pulling gas data for a piece I was writing on scaling economics. And there it was, a number that just looked wrong. Starknet, one of the most technically respected projects in the entire ecosystem, was paying roughly 5,000 ETH per year just to post proofs to Ethereum. Not to run the sequencer. Not to pay developers. Just to prove that what it was saying happened, actually happened. At the time, that was close to ten million dollars. For a protocol whose monthly revenue, real revenue, not token emissions, not grants, could be counted in the tens of thousands of dollars. I closed my laptop and went for a long walk. I needed to process what I was seeing. We didn’t talk about this in the bull market. We didn’t want to. We were too busy celebrating throughput milestones and “Ethereum 2.0” vibes. But the math was never a secret. It was hiding in plain sight, in block explorers and cost models that almost nobody wanted to open. Trust is no longer a promise; it’s a protocol. And right now, the protocol is telling us something uncomfortable about the people building the next generation of Ethereum. Let me give you some context, because I think most people reading this still believe that ZK rollups are the clear endgame for Ethereum scaling. I’ve been saying it myself since 2021. I’ve interviewed zkSync founders, StarkWare researchers, Polygon engineers, and Scroll contributors on my podcast, Chain of Thought, back when we were all naive and excited. The technical argument is beautiful. ZK rollups compress thousands of transactions into a single validity proof. That proof gets posted to Ethereum. Ethereum verifies it. Done. You get the security of the base layer with the throughput of a sidechain. There’s no seven-day withdrawal window like with optimistic fraud proofs. There’s no game theory risk. Finality is immediate. It is, honestly, one of the most elegant pieces of applied cryptography I’ve ever seen. But elegance doesn’t pay the bills. And the bills, my friends, are enormous. The core problem is that proving systems are computationally brutal. A single proof for a batch of transactions requires complex polynomial commitments, elliptic curve pairings, and increasingly, recursive proof composition. Let me break down where the money actually goes, because this is where most analysis falls short. There are two major cost centers in a ZK rollup operation. The first is proving. You need a server farm, essentially, specialized hardware with high-end GPUs, to run the proving software. Every single transaction that enters the rollup needs to be included in a computation that builds a proof. The more transactions you process, the more proving power you need, even if you batch efficiently. And the cost per proof, depending on circuit complexity, can range from tens of dollars to thousands of dollars. The second cost center is posting that proof to Ethereum, which costs gas. In late 2023 and early 2024, before EIP-4844 made blob space available, posting a ZK proof to Ethereum L1 could cost anywhere from a few hundred to several thousand dollars per batch, depending on network congestion. After Dencun, blob space did make calldata cheaper, and that helped a lot. It reduced the cost of posting data availability by roughly 90 percent or more. But here’s the part nobody wants to hear: proving costs did not go down. They went up. Why? Because the more data you can post cheaply, the more transactions you’re incentivized to include in each batch, which means you need to prove more computation, which means your proving cost per batch increases. You can only compress so far, my friends. There’s a fixed cost to proving a batch no matter how small, but as you include more and more transactions, the total proving complexity grows. And the hardware required to do that efficiently has a non-trivial amortized cost. I’ve spoken with operators who run proving infrastructure for major ZK projects. Off the record, they tell me the same thing. The hardware capex alone for serious proof generation is in the millions. Then you have electricity, cooling, storage, redundancy, and the engineering team that keeps the proving stack running. It’s a data center business. It is not a software business. And here’s the kicker: when you charge users transaction fees, the market price for that transaction, decided by market dynamics and competitive pressure, is often lower than the actual cost to prove it. In other words, for every single transaction they process, the ZK rollup operator is losing money. I’m going to give you a real historical example, because I’ve been tracking this for years and I have the spreadsheets to prove it. Let’s look at a hypothetical ZK rollup that processes about 5 million transactions per month. That’s a realistic level for a moderately successful L2 with some DeFi activity. Say they batch aggressively and post to Ethereum, let’s, obviously, use blob space now, every two minutes. That’s about 21,600 batches per month. Each proof, let’s say about 400,000 gas to verify on L1, plus the blob data cost. If blob prices stay low, data might be the smaller component now. The verification fee on Ethereum is the bigger fixed cost, let’s say 250,000 gas times a average price of 20 gwei, which gives you about 0.005 ETH per batch. Multiply by 21,600 batches and you’re looking at about 108 ETH per month just for L1 verification. Add in blob cost of maybe 50 ETH per month, and you’re at around 158 ETH just for on-chain. At a ETH price of 3,000 dollars, that’s roughly 474,000 dollars per month in on-chain costs. Now add the proving. This is the one everyone underestimates, because we don’t have a clean public oracle for prover costs. But based on my audit experience and conversations with people running services like Herodotus and other ZK infrastructure, a batch of 200 transactions on a complex circuit like a standard EVM rollup requires a proving setup that costs between 1 and 2 dollars per transaction right now. I’m being generous. Let’s use 1.50. So five million transactions times one dollar fifty cents equals seven point five million dollars per month. Seven and a half million dollars per month in proving costs alone. I watched this data and felt my stomach drop. Let’s put that in comparison. If the rollup charges an average fee of 0.03 dollars per transaction, which is realistic in a bear market and genuinely competitive with Optimistic rollups and other cheaper L2s like Base or Arbitrum, then five million transactions generate 150,000 dollars per month in revenue. That’s it. 150,000 in revenue against seven and a half million in proving costs, plus 474,000 in on-chain posting, plus sequencer infrastructure, plus monitoring, plus developer salaries. The monthly burn rate is somewhere around eight to nine million dollars. The monthly gross revenue is 150,000 dollars. That’s not a ratio. That’s a cliff. Even if I’m wrong by a factor of two, even if prover costs are somehow only one quarter of my estimate, the shortfall is still millions per month. ZK rollups, the beautiful, elegant, secure, technically superior solution, are burning money on every single batch they post. And most of them are subsidizing fees with token emissions and treasury funds, hoping that a bull market will return, volumes will multiply by twenty, and transaction fees will rise enough to cover costs. But wait, I need to be honest with you, because that’s part of my job. The picture isn’t uniformly bleak. There’s an entire hardware race happening that I find genuinely exciting. Companies like Cysic and Ingonyama are building ASICs and FPGA-based accelerators specifically for ZK proving. They claim, and I’ve seen initial benchmark data, that specialized hardware can reduce proving costs by orders of magnitude, not by 50 percent, but by 10x to 100x. We’re moving toward a world where a prover chip the size of a GPU can do what a rack of servers does today. That would change everything. If proving a batch costs 0.01 dollars instead of 1.50 dollars, the ZK rollup business model starts to make sense. You could profitably process millions of transactions a day. But that day is not today. Today, the hardware is still in R&D, still being tested on testnets, still not deployed at production scale. And I’m a skeptical optimist, I learned that from my burnout experience in the bear market. I’ve seen too many perfect solutions die in the gap between whitepaper and execution. Crypto history is littered with projects that promised to make everything cheaper and faster and never shipped. So what’s the contrarian angle here? You expected me to say ZK rollups are doomed. That’s too easy. Let me flip it, because that’s where the actual insight hides. I believe ZK rollups are actually fine in the long run, but only if they stop pretending they’re base-layer businesses and start acting like infrastructure providers. The current framing is wrong. They’re trying to compete on user fees with general-purpose L2s, selling block space at 0.03 dollars. That’s the Web2 mistake all over again, competing on price when you’re offering something fundamentally better. The real play is not being another dumb L2. It’s becoming the settlement and proving layer for a whole ecosystem of other rollups, the ones that are cheap because they don’t want to build their own proving, the ones that need fast finality and don’t want to wait seven days, the application-specific rollups for gaming and AI and social that can’t afford to build projective complexity from scratch. In a world where everyone wants to launch an app chain or a Layer 3, someone has to verify all those proofs. ZK rollups that pivot to being the proving layer for the rest of the stack are no longer losing money. They’re earning money from other teams’ success. They’re selling pickaxes in a gold rush, except the pickaxe is multi-party computation. That is the kind of reversal I’ve come to appreciate after all these years. In 2020, during DeFi Summer, I hosted a meetup series called Yield and Connect, where we turned yield farming into a conversation about community trust. People thought I was crazy, discussing post-2008 solidarity while everyone else was chasing the highest APY. But the people who were actually building, they understood. They knew that protocols that still worked when the music stopped would be remembered. Same logic applies here. When we’re in the depths of a bear market, and I have no idea if we’re still in one by the time you read this, the protocols that are still building infrastructure for the future, even at a loss, are the ones that matter. I am not suggesting you buy their tokens. I’m suggesting you watch their development roadmaps with actual attention, not with the lens of a speculative chart grok. Do they have a clear path to hardware acceleration? Do they have partnerships with teams that have actual users, or just other L2s with no users? Do they have a treasury that can survive two years of bleeding? Let me give you something concrete to look at. Based on my audit experience, I’ve developed a quick health check for ZK rollups, and you can use it too. First, look at their state commitment frequency. How often do they post a proof on L1. Is it every single batch, or are they batching over longer periods? Longer batching means lower costs but worse finality, and it means they’re optimizing for survival, not for UX. Second, look at their fee structure. Are they charging per transaction a fee that’s meaningfully higher than their base layer equivalent, or are they running on subsidies? If fees are suspiciously low, remember that someone is paying for it, and that someone is either a VC or a token holder. Third, check whether they’ve made any public statements about their proving hardware partners. If they’re silent, if they don’t have a hardware partnership, if they don’t talk about prover economics at all in their public communications, then they’re either hoping the problem goes away or they don’t understand the problem, and both of those are red flags. I’m also going to talk about the fee model because that’s the part that most analysts get wrong. In an optimistic rollup, if you charge users 0.05 dollars and actual costs are 0.02 dollars, you make a profit. People will say, why charge arbitrary transaction fees when you can be on an optimistic chain and cost nearly nothing? The answer is that ZK rollups aren’t competing for people who are satisfied with sharing a block with hundreds of other applications on Arbitrum. ZK rollups are competing for institutions, for high-frequency trading, for enterprise use cases that need privacy, cross-chain atomicity, and immediate withdrawal calldata proofs. Those are not commodity services. Institutions pay a premium for finality guarantees that are mathematically certain instead of game-theoretically probable. They pay for the ability to exit the ecosystem with no delay. They pay for the absence of the battle test problem, for proving that the backend mathematically can’t be an adversary. That, my friends, is a sales pitch that no optimistic rollup can match. The future is not about per-transaction fee competition against a 2-cent baseline. It’s about becoming the settlement layer for the entire Ethereum ecosystem and charging a premium for the privilege. One more layer of complexity. The next evolution of ZK isn’t just rollups on Ethereum. It’s ZK proofs cross-pollinating with AI. I founded the Human-Centric Blockchain initiative in 2026 because I saw AI agents starting to interact autonomously on-chain. They need to prove their next state transitions without revealing their private data. They need trustless delegation of funds. Someone has to verify that an agent’s computation was correct, without forcing that agent to reveal its entire dataset to the world. ZK proofs are the only technology that makes that possible. We don’t call this rollups anymore. We call it computational integrity. If you study the economics of ZK rollups just through the lens of DeFi activity, you’re looking at the first inning of a nine-inning game. The later innings, where autonomous agents use ZK proofs to transact without disclosing proprietary data, have market implications at least an order of magnitude larger than what we saw in DeFi Summer. And it’s in those innings that current proving costs become embarrassingly cheap for the value being delivered. I firmly believe that the world needs provable computation more than it needs cheaper block space. The last couple of years taught me something that I keep coming back to, in my writing, my podcast, and my teaching. It helped me grow from being a mere technical analyst to someone who bridges the gap between the values of decentralization and the cold, hard realities of unit economics. The pivot wasn’t about abandoning the vision. The pivot was about learning to see what is, not what I wanted to be. I wrote a series during the 2022 bear market called Finding Humanity in the Void, partly to process my own burnout, and partly to remind people that even when the market bleeds, the community, the people, the builders, remain. The same ethos applies to infrastructure projects. While token prices scream and retail investors panic, I’ve seen ZK teams forge ahead on testnets, quietly fixing bugs, optimizing circuits, and shipping reference implementations for specialized hardware. They’re not telling you constant stories about their token price. They’re publishing research about increasingly efficient proof systems. You can call them OBSESSED with the long-term. You can call them foolhardy. But their progress is real, and it will show up in the culture of the ecosystem sooner than most of us expect. Now, let’s get practical. I know you didn’t come here just for philosophy. You want to know how to read the situation without getting burned. Here are the concrete signals I track. One: actual user growth on ZK rollups that takes into account organic use versus airdrop farming. The current numbers are ugly, but as with Ethereum in its first few years, ugly now does not mean ugly forever. Two: the convergence of hardware development. I’ve touched base with mining company veteran teams that now build ZK accelerators, but if a serious player like NVIDIA or AMD decides to launch a dedicated ZK instruction set, mark it as a turning point. Three: the emergence of middle-layer service providers that abstract ZK away from developers, letting them pay for provability through a cloud API. That kind of platform economics has historically been where market domination begins. Four: whether ZK rollups finally start sharing their margin data openly. Audit reports of ZK proving operations with clean revenue and expense breakdowns, honestly, one public, transparent, third-party-audited cost analysis by a major ZK rollup would be worth ten marketing posts from ten different projects. It’s this lack of transparency that makes me worried we’re waiting for the whole house of cards to fail once more. There’s also an ecosystem component that’s often overlooked in these discussions. When proof costs are high, sequencer decentralization is even more urgent. If you’re a rollup losing money on every batch, you’re naturally going to centralize your proving to one operator that can afford scale. But then you’ve created a single point of failure, not just for censorship, but for existential operational risk. If that one proving company collapses, the whole rollup freezes. I believe the future of ZK scaling is not merely more proofs; it’s decentralized proving markets, where independent proving nodes compete for work, not the current system where every team builds around its own proprietary proving setup. And that decentralized proving market, my friends, is nothing less than an early precursor of a fully decentralized computation layer for… for everything. It’s the infrastructure layer that platforms like ours, and institutions, and maybe global data markets, will quietly operate on. Let me bring it back to what matters most: survival. In this bear market, which one of these scaling ecosystems will still be alive when the next bull market flips? I can’t tell you which token will pump, but I can tell you which architecture is more likely to survive: one that doesn’t bleed huge financial losses every time usage grows even a little. That’s the harsh but important truth about ZK rollups today. They are selling a technology that will redefine trust and automation in finance, but they are not yet selling a product that pays for its own existence. And until they do, the Ethereum scaling roadmap is balanced on the edge of a knife held by the teams that understand this dilemma and are taking action. Everyone else, no matter how polished their demos might look, is building castles on sinking sand. The moment they run out of treasury funds is the exact moment their entire scaffold crumbles. But here’s my closing thought, and it’s a vision, not a summary. I’ve been writing about blockchain for eighteen years now, from defunct ICOs to the launch of ETFs, from pumpamentals to the philosophy of tokens. The one thing I’ve learned, the hard way, is that code is law, but empathy is the interface. For all our talk about provability, trustlessness, and decentralized infrastructure, we should not lose the fact that this technology exists for people, for the fragile creature that needs to know their identity is safe while they transact, for the small trader who can’t afford a legal team but can still use a smart contract, for the autonomous AI agent that deserves to’ve auditable rules. ZK rollups are the nerve system of this future. They hurt now. They bleed. But nerves grow through pain. I’ve seen Ethereum survive a fork war, and a DAO hack, and the crash of 2018 and 2022. I’ve seen developer communities survive the collapse of entire narratives around them. The ZK narrative will survive too, because it is not a narrative, it’s the final building block of a society that can self-validate. So I’m not worried about the bleeding ZK rollup operators. I’m worried about the teams in this space that don’t understand why they’re bleeding. Because that means they haven’t internalized the fact that this isn’t about scaling Ethereum. It’s about scaling the integrity of the entire digital world. Trustless systems require trusting relationships, and the first relationship ZK rollups must make with their users, before all else, is straightforward: freedom of information, in data and in costs. If they can’t do that, none of our technology matters. If they can, we’re about to give humanity a superpower it never thought it could have: the proof that we didn’t cheat.