Disclaimer: This is a thought experiment, not a recommendation, a forecast, or a plan. The numbers rest on a pile of assumptions influenced by our experience. There are too many variables for this to be accurate in general, or for anyone else without redoing the math. The point is to show one way to evaluate public chains as a ledger option — an example framework, not an answer you should copy.
In "The API We Didn't Build" we explained how putting NDD on public blockchain rails does three things at once. It saves clients the time and money of building another bank integration, it provides more utility and fees to Ethereum, and it offloads compute, network traffic, and API load onto infrastructure we don't own.
Which got us wondering: what would this bank have looked like if we'd put the whole ledger on the public Ethereum network from day one? Not just NDD, everything in our private Ethereum network as well. And if we're pricing that, we may as well price a traditional banking core alongside it.
Per the disclaimer above: set aside the regulatory, security, and operational reasons you probably wouldn’t do this. We're doing the math anyway, with assumptions stated as we go and guesses called out.
TL;DR
A traditional core is the most expensive way to run our ledger at every size we modeled, and not by a little. That gap never closes anywhere on the curve, which should tell you nobody buys a core for the price.
Between public and private blockchains, it's close at our scale and a landslide above it. (We model Ethereum L1 directly, not Layer 2s, because rollups do not yet match the decentralization, censorship resistance, or sovereign consensus of mainnet). At Regional bank level, private wins by 13% and the entire margin is less than a quarter of an engineer’s salary. Move our staffing assumption from 0.75 FTE to 1.0 and public wins instead. But at Money Center bank level private wins by 83%, because gas and signing are metered and a private fleet isn't. Every transaction you add is free on one and billed on the other.
The tiebreaker is time. Gas is the only input in this model that gets cheaper, falling somewhere between 20% and 64% a year depending on whether you trust the published Ethereum roadmap or the last five years of history. Everything else moves the other way. Salaries rise, vendor contracts renew upward, and nobody's core gets cheaper at renewal.
Ethereum set out to be a "world computer." As throughput climbs and execution costs continue to collapse, you can start to see a future where public rails displace both legacy core vendors and private cloud fleets alike. That was the original promise of cloud computing: paying strictly and exclusively for what you actually use. Cloud platforms and core vendors answer to shareholders who demand expanding margins, annual price escalators, and locked-in enterprise minimums. Ethereum’s incentive is structurally aligned in reverse: driving unit execution costs down to maximize volume and global network adoption. It becomes the true realization of pay-as-you-go utility computing.
Of course, cost is only one hurdle: public rails still require lower latency and privacy before an institutional bank can migrate live customer ledgers. But on cost alone, we'd have been fine starting on public rails, and we'd probably be better off there within five years. We didn't rent the ledger, but we may yet.
The Results
Modeled at $10 billion in assets and 250,000 transactions a month, a mid-size Regional payment bank:
A private network runs 13% below public Ethereum and 80% below a traditional core.
What you're paying for
The Public Ethereum column is mostly variable volume costs—gas (58%) and MPC/HSM signing (7%), plus RPC (6%) and labor (29%). Gas and signing are the only lines in the model that grow when volume does.
The Private Ethereum column splits evenly between people and infrastructure (~50/50), and neither half responds to volume. Internal HSMs and nodes are fixed overhead. You pay it whether you settle a thousand transfers or a million.
The Traditional Core column is mostly invoice. People are only 25% of it, the other 75% is a contract you don't get to tune. Buying a core doesn't remove staff either, it changes what they do: vendor management, release coordination, DR tests, security reviews, reconciling the invoice against what you used.
How It Scales
Three architectures, three different shapes:
Every figure includes labor. Headcount runs 0.5 to 1.0 engineers on a private chain, 0.25 to 0.5 on public, and 1 to 6 operations staff on a core — the appendix breaks out all three by tier. Gas is priced at 0.569 gwei throughout, the 45-day median of the published network average. Public chain includes tiered MPC/HSM signing fees ($0.05 down to $0.002/tx).
A core is the most expensive option at every size. A De Novo bank pays a $12,500 vendor minimum and staffs 1.0 FTE against it before anyone opens an account. Grow to Money Center and the bill is $779,500 a month — 31x what you paid at the start, mostly because the vendor prices on assets and account counts. Some vendors charge per transaction too, but it's fractions of a penny and disappears next to the platform fee. Per-transaction cost drops as volume rises ($0.60 at regional, $0.31 at money center). Private at that tier is $0.016. Not close.
Private Ethereum charges the same no matter what you do. The bank grows 2,000x. The bill grows 1.6x. QBFT has a minimum viable size, so the node count stays flat. Most of the increase is a second half-engineer and bigger production machines, not more nodes. Volume is free. The millionth transfer of the month costs nothing.
Public Ethereum has the lowest floor and the only flat headcount. At De Novo scale the entire bill is $105 of gas, $200 of RPC, $63 of MPC signing, and a quarter of an engineer. No minimum, no contract, nothing to provision. Half an engineer still covers a Money Center, because there's no private chain capacity to grow into.
The public chain wins the bottom two tiers. Then variable costs catch up: 2.5 million transactions a month costs $215,500 in gas and signing fees ($210,500 gas + $5,000 signing) against a private chain configuration that runs the whole ledger for $39,850.
One caveat on all of it: this is priced at today's gas. Fees have fallen roughly 64% a year for five years — 100 gwei in 2021, 0.57 now — and the roadmap points the same way, with Glamsterdam clearing a path to about 3.3x current L1 capacity. Take the conservative read, capacity alone and no further migration to L2s, and it's 20% a year. Anywhere in that 20–64% band, public chain takes the Regional payment bank within five years and Money Center bank within ten, and at the historical rate it does both inside three. Worth noting because every other column here moves the other way: salaries rise, vendor contracts renew upward, and nobody's core gets cheaper. Gas is the only input in this model with a roadmap to reduce it, funded by somebody else.
Appendix: Cost Inputs
Reference figures behind the model. Estimates meant to be representative, not precise.
Labor
Roles loaded at base × 1.4 for benefits, payroll taxes, equity, overhead.
Which rate applies depends on the architecture. A private network runs on engineers; a traditional core runs on operations staff. That's why a headcount advantage doesn't convert one-for-one into a cost advantage.
Neither blockchain figure scales much. A private chain configuration needs the same patching whether it's settling a thousand transactions a month or a quarter million, so private holds at 0.5 FTE through Community, then doubles to 1.0 across the top two tiers as specs, volume, and examiner attention climb. Public starts at 0.25 and stops at 0.5 — public never needs more than half an engineer, because the work doesn't grow with the balance sheet.
The core column runs one person to six. That's deliberately light for a bank of that size, because N3XT doesn't lend — no origination, no servicing, no collateral, no credit risk, and none of the staff those carry. A full-service bank at $100 billion would need multiples of this. What's left is vendor management, release coordination, DR scheduling, security reviews, batch monitoring, and invoice reconciliation. Real work that lands in your payroll rather than their invoice, which is why a core looks cheap in a vendor benchmark and expensive in a budget. All three figures are additive to the hard costs below.
Staffing sensitivity
The engineer count is the softest assumption. We model 0.75 FTE at Regional bank level because we built the systems and automations ourselves. Need a full engineer instead? Private goes from $31,250 to $36,500. Public is $36,050. The entire private advantage over public is a quarter of a salary.
Private Network Costs
Eleven nodes is our floor for a network — seven block producers for QBFT plus at least four RPC nodes. Under QBFT's 3f+1 quorum formula, seven validators tolerate up to two simultaneous validator failures (f=2) without stalling consensus, and four RPC nodes ensure high availability and failover across regions. Six networks put the baseline floor at 66 nodes across the org. Running five test environments (two of which are client-facing integration sandboxes) might seem excessive from the outside, but when you're settling real money, you should be obsessed with testing, staging parity, and correctness.
The node count doesn't increase at Money Center scale — 66 nodes remain sufficient across all six networks. Instead, scaling throughput means vertically scaling compute (more cores and memory) and storage on the production nodes, raising the fleet average from $205 to $225 per node. Observability and networking also scale with traffic(500 at De Novo, $1,000 at Community, $2,000 at Regional, up to $4,000 at Money Center). A De Novo bank deploys the same 66 nodes at about $14,000 in hard costs, because you can't run a fractional QBFT network just because you're small. That's $14,000 to $18,850 across a 2,000x range of bank sizes.
Per-node cost rises with specs and storage on the production machines, and storage never comes back down since chain history doesn't compact. Key management runs on internal cloud HSMs / KMS bundled into standard infrastructure and node operations, because validator and transaction access is constrained behind private VPC perimeters and permissioned consortium rules.
Public Network Costs
This column beats a fixed private configuration below 199,000 transactions a month and loses above it. Price gas instead at the 0.046 gwei and the threshold moves to roughly 1.1 million (or 2.4 million at Money Center signing rates) — public would win nearly every tier.
Gas basis. 0.569 gwei, the median of the published daily average over 45 days — $0.078 a transfer, or $0.084 with 8% added for mints, burns, and allow-list updates.
RPC assumes two providers, but the second is failover rather than load sharing — commit spend at the primary, keep the secondary minimal.
Signing and Key Management basis. On a public chain, key compromise could mean immediate, loss on an open network. A bank cannot sign public mainnet transactions with basic server keys; it requires robust institutional-grade HSM/MPC infrastructure (e.g., Fireblocks, Turnkey, Utila) with multi-party quorum policies, hardware isolation, and strict compliance automation. We model standard volume-tiered pricing: $0.05/tx at De Novo ($63/mo), $0.04/tx at Community ($1,000/mo), $0.01/tx at Regional ($2,500/mo), and $0.002/tx at Money Center ($5,000/mo).
Traditional Core Costs
Built bottom-up rather than from a percentage of assets, because percentage-of-assets benchmarks collapse to absurd figures at the small end. Cornerstone Advisors' 2025 numbers put bank technology spend at 0.3–0.6% of assets with core systems at 15–25% of that, which implies a $50 million bank runs its core for a few thousand dollars a month. Vendor minimums generally start between $100,000–250,000 a year regardless of size, and Cornerstone's definition folds internal staff into the percentage, so using it directly both understates the vendor floor and hides the headcount. Per-item posting fees are real but tiny — we model $0.001 a transaction, fractions of a penny, which is where blended core pricing usually lands and barely moves the total next to the platform fee.
This is still the softest column in the model. We haven't priced a core contract at this level of detail ourselves; these are estimates built from vendor minimums, Cornerstone benchmarks, and what we'd expect per-item posting to cost. The asset-percentage method would put regional around half this number. Both can't be right. We ran the bottom-up version because it's closer to what a bank actually signs, but treat it as the weakest input here.
Transaction volumes
This is the assumption that decides the whole model. N3XT is a narrow bank, does not loan, and is focused on institutional and enterprise payment flows. The premise is that wires, ACH, and FedNow are overpriced and or slow, and that if we can get enough of a client's payment flow onto our ledger, most of it settles as a book transfer. Deposits aren't stored value here, they're working capital. More deposits means more clients, which means more transactions.
Flat at 25 transactions per $1M of assets per month, scaling linearly. That's a finger in the air, and it's the number to argue with first. A conventional institutional deposit book runs closer to 10–15, because balances sit still and tickets are large — a $10 million wire is one transaction, and so is a $10,000 one. We assume roughly double that, on the theory that capturing payment flow means more transactions per dollar of deposits, but nothing like the velocity of an actual payment processor.
The threshold that matters: public Ethereum beats a private network below about 20 transactions per $1M of assets, and loses above it. We model 25. That's a 20% margin on a finger-in-the-air number, so a reasonable person could put us on either side of it.
Why Ethereum
There are dozens of chains faster and cheaper than Ethereum on paper, and Layer 2 rollups that cut gas fees to pennies today. We price Ethereum L1 directly because enterprise settlement requires sovereign, battle-tested decentralization without sequencer centralization risk or administrative multisigs. The tiebreaker isn't raw performance, it's the first question a bank asks a core vendor: will you still exist at renewal?
That's the real reason banks pick FIS and Fiserv. Nobody chooses a forty-year-old batch ledger for its elegance. You choose it because a core conversion could cost millions and take years, and you can't survive your ledger vendor getting acquired and sunset.
Run pubic chain options through that filter and most of the field falls out. A chain whose validators are paid by a foundation treasury is a vendor with a runway. A chain whose operators are paid by organic fee revenue is a utility. Fee revenue is the credit rating of a blockchain, and gas isn't a toll — it's the price of a globally distributed operator set you don't hire, patch, or wake up at 3 a.m.
