The replay test data landed at 2:47 AM Sydney time. I was cross-referencing the Ethereum Foundation's public dashboard against my own Dune queries when the number hit me: 2,687,652 transactions. That's how many historical transactions break under EIP-8037's new state-gas pricing. Add EIP-8038's impact and the total climbs past 5.7 million. Out of 929,731,274 transactions replayed, that's a 0.3% failure rate. In any other engineering context, 99.7% success is a triumph. In smart contract land, it's a fault line running directly beneath the DeFi ecosystem's most critical infrastructure.
The code doesn't lie, but it does break silently. And this particular breakage is scheduled for Q4 2026.
Context: The Glamsterdam Upgrade and Its Two Proposals
Ethereum's next scheduled network upgrade, Glamsterdam, isn't a single EIP. It's a package deal built around two complementary proposals that together aim to roughly triple the network's base throughput. The first, EIP-8037, introduces a separate state-gas dimension that runs parallel to execution gas. The second, EIP-8038, adjusts the costs of accessing existing state. Both are currently in the formal Review stage, meaning the Ethereum Foundation has already done substantial homework but hasn't locked anything in.
The core problem these EIPs address is state bloat. Every new account, every new storage slot, every deployed contract adds to the Ethereum state trie. That state must be stored by every full node, forever. The current pricing model charges a flat fee for state creation, which doesn't reflect the long-term storage cost. EIP-8037 fixes this by charging 1,530 gas per byte of new state, targeting annual state growth of 120 GiB under a 150 million gas limit scenario. Even in the worst case—a 200 million gas cap—growth stays bounded at 160 GiB per year.
The fee adjustments are dramatic. Creating a new account jumps from 25,000 gas to 183,600 gas—a 7.3x increase. New storage slots go from 20,000 to 97,920 gas, a 4.9x jump. Deploying a 24 KiB contract with a new account costs 37,784,880 gas versus 4,947,200 today. That's a 7.6x multiplier. These aren't marginal tweaks; they're structural repricing of state creation.
Based on my audit experience during the 2017 ICO sprint, I can tell you that repricing state access is one of the most dangerous changes you can make to an EVM-based chain. It doesn't just affect gas costs—it changes the economic calculus of every contract that assumes certain operations will remain cheap.
Core: The On-Chain Evidence Chain
The Ethereum Foundation didn't just theorize about the impact. They ran a massive replay test covering 929,731,274 transactions from December 2024 through June 2026. The methodology is sound: take historical transactions, re-execute them under the proposed gas rules, and see what breaks. This is the kind of empirical rigor I respect—data is the only witness that never sleeps.
The results reveal a clear pattern. Under EIP-8037 alone, 174 million transactions are repairable—meaning they'd fail but could be fixed with code changes. But 2,687,652 transactions are potentially broken in ways that might not be fixable. EIP-8038 adds another 84.7 million repairable and 3,036,537 potentially broken. The overlap between the two EIPs' impact zones is where the real danger lives.
The Ethereum Foundation's public outreach report identifies specific protocols with repeated failures: Across, Socket/Bungee, CoW Protocol, and 0x. These aren't small players. They're core DeFi infrastructure—bridges, DEX aggregators, and trading protocols that handle billions in volume. The report prioritizes the eth-infinitism ERC-4337 EntryPoint and the broader smart account stack, which makes sense. Smart accounts are the highest-risk category because they involve multiple interacting contracts: entry points, account implementations, factories, and validators. Each layer must be updated in coordination, and immutable contracts can't be updated at all.
Here's what the raw numbers don't tell you: repeated activity likely dominates the count. The same contract called millions of times by different users inflates the transaction count. The actual number of unique affected contracts is probably much smaller. But that's cold comfort when the affected contracts include the ERC-4337 EntryPoint that thousands of smart accounts depend on.
The gas repricing also breaks tooling assumptions. Wallets, RPC providers, indexers, node tools, and gas estimators all need to understand the new state-gas dimension. The Platåberget testnet announcement explicitly warns that old tools may not work correctly. This is a systemic adaptation burden that hits infrastructure providers first and hardest.
Contrarian: The 3x Narrative Is Engineering, Not Marketing
Here's where I push back on the prevailing narrative. The "triple network speed" headline is technically accurate but practically misleading. The 3x figure is an engineering support target, not a guaranteed mainnet capacity increase. It means the network could handle 3x the current throughput if demand materializes and the gas limit is raised accordingly. It doesn't mean your transactions will suddenly be 3x faster or cheaper.
More importantly, the replay test's "counterfactual failure" framing deserves scrutiny. These failures only occur under the new rules—they don't exist today. That's precisely why they're dangerous. The market has priced in the current behavior. When the rules change, contracts that were operating at the edge of their gas assumptions will fail in production, not in testing. The 0.3% failure rate sounds small until one of those failures is a bridge holding $500 million in user funds.
Liquidity is just trust with a price tag. When that trust breaks due to a silent compatibility failure, the price tag gets renegotiated under panic conditions.
There's also a hidden assumption in the state growth projections. The 387 GiB extrapolation from a 30 million to 60 million gas limit assumes linear scaling. But the response to gas limit increases is demonstrably non-linear. The 160 GiB worst-case figure under a 200 million gas cap suggests the Foundation has already modeled the upper bound internally. They know something about the non-linearity that the public extrapolation doesn't capture.
The Ecosystem Ripple Effect
The transmission chain is clear. Upstream, node operators and validators see no direct impact—the consensus mechanism doesn't change. But the infrastructure layer—RPC providers, indexers, gas estimators—must adapt to the new pricing rules. That's a short-to-medium-term negative as they scramble to update. The DeFi layer faces compatibility risks that could manifest as user fund losses if contracts fail in unexpected ways. The L2 ecosystem, however, may benefit indirectly. If L1 throughput increases, L2 settlement costs drop, which could accelerate rollup adoption.
The smart account stack is the highest-risk area. ERC-4337 EntryPoint, ZeroDev, Alchemy, and similar infrastructure all need new implementations. The Foundation's outreach report explicitly prioritizes these, which tells me they've already identified the critical path. But immutable contracts present a special problem. If a contract can't be upgraded, the only fix is deploying a new version and migrating users. That's expensive, risky, and some projects may simply choose not to do it—leaving zombie contracts that still hold user funds but can't function under the new rules.
Speed is an illusion when the ledger is honest. The honest ledger here says millions of transactions will break, and the ecosystem has roughly six months to prepare.
Takeaway: Watch the Signals, Not the Headlines
The Glamsterdam upgrade is a necessary pain. State growth is an existential threat to Ethereum's decentralization—if node operation becomes too expensive, the network consolidates and loses the security that makes it valuable. The Foundation's approach is methodical: replay tests, public dashboards, outreach reports, dedicated testnets. This is how you ship a risky upgrade responsibly.
But the market hasn't priced in the compatibility risk. The narrative is still "Ethereum gets faster," not "Ethereum breaks 5.7 million transactions." That gap is where the opportunity and the danger both live.
We don't need to speculate about the outcome. The data will tell us. Watch the EIP status move from Review to Last Call. Watch the Platåberget testnet for stability. Watch for adaptation announcements from Across, CoW Protocol, and 0x. And most importantly, watch the replay test dashboard—if the potentially broken transaction count starts dropping as protocols fix their code, the risk is being managed. If it stays flat, we have a problem.
The next 90 days will determine whether Glamsterdam is a smooth evolution or a compatibility earthquake. The code doesn't care about narratives. It only executes. And in Q4 2026, it will execute under new rules.