09.06.2026
CEX vs DEX: Centralized and Decentralized Crypto Exchanges Explained
Read article
Your merchants have been asking about USDT for two months, and "we don't do that yet" has stopped sounding like a roadmap answer. So the choice is on your desk. Put your own engineers on a crypto payment gateway, or take someone else's infrastructure under your own brand — white label — and sell it as part of your payment product. The board will want a figure.
Feature comparison will not decide this. Three numbers will, and they come in a fixed order. First, your monthly volume — it decides whether "buy on partner terms" is even available to you. Second, the licence — bought infrastructure does not come with one, and somebody in the chain has to hold it. Third, payroll against a percentage of turnover. Building never finishes; it is a permanent cost line. What stands against it is a rate somewhere between 0.4% and 2%.
Work them in that order and the answer usually arrives before you open a vendor deck. Below the volume threshold and without your own authorisation, you are not choosing between building and buying at all — you are choosing which partner carries the licence and the infrastructure while you carry the merchants.
This question gets asked last and should be asked first, because the answer is binary.
Infrastructure providers that sell to payment companies set an entry threshold. At some of them it starts at $500,000 in monthly payments and six months of operating history, going by published onboarding terms as at 22 August 2026. No market standard exists here: some sit higher, some lower, and most publish nothing at all. Assume that order of magnitude until a provider tells you its own.
Now the part that decides your year. Below the threshold, outright refusal is rare. You are quietly moved to a different product: an ordinary merchant account, with your logo on the payment page and retail terms behind it. Fine, if a branded checkout was the goal. A problem, if you promised the board a payment product you resell to fifty merchants at your own margin.
If you are under the bar today, the move is to grow into it. Ask for one clause in writing while you do: a rate review at a stated monthly volume, with the partner terms named in advance. That turns the second conversation into a calculation instead of a negotiation.
Questions to ask before the demo rather than after it:
That last one only looks commercial: it is the licence question, which is the next number.
Bought infrastructure does not issue a licence. This is the most expensive misunderstanding in the whole build-versus-buy conversation, because a contract with an authorised provider makes the question look settled without settling it.
In the European Union, the MiCA transitional period — the grandfathering window that let firms keep operating under their national regimes — ran out on 30 June 2026. What expired is the window, not the rule: from 1 July 2026, firms that had been trading on those national permissions must hold CASP authorisation or stop serving EU clients. ESMA set out the wind-down in its public statement on the end of the MiCA transitional period, dated 23 June 2026, reference ESMA75-113276571-1710. There is no grace period in that text, and the date is already behind you.
Most PSPs ask whether the provider has a licence. Most serious providers do, and they will show you the register entry. The question that decides your own position is a different one: who, in this chain, is the crypto-asset service provider to my merchant? There are two answers, and they put the authorisation in different hands.
White label makes the second shape look like the first on screen. Contracts do not care what the payment page looks like, so read yours for this one thing before you read it for rates.
If the answer is that you need your own permission, the UK register gives you the honest odds. Since 10 January 2020, and as at 1 August 2026, its cryptoasset registration regime has produced the following. The figures sit on the FCA's own page for firms that need to register:
That accounts for 391 of the 412; the rest had no outcome recorded as at that date. The last line says the most. Those 263 firms were not turned down. They looked at what was being asked and stopped.
On timing the regulator is precise, and slightly cruel about it. On a complete application it allows itself three months to decide, counted from the point at which it holds all the information it needs — not from the day you filed. That wording sits in the FCA's own guidance for registration applicants. The gap between those two dates is where most of the calendar goes.
Two caveats for the board room. This is UK AML registration: a different permission from a payment institution licence, with a different process. And the figures cover the UK only — MiCA authorisation in the EU runs on its own track. Neither is a quarter-long project. If your plan for this year involves your own permission, it is next year's plan. None of this is legal advice: jurisdictions and dates are stated so your counsel can start from the right place. If Europe is your market, the wider picture is in the guide to stablecoin payments in Europe after MiCA.
Now the arithmetic the board actually wants.
Start with the cost of the team, because that is the number that never goes away. The Stack Overflow Developer Survey 2025 puts the median annual salary for a backend developer at $175,000 in the US, $108,913 in the UK and $87,011 in Germany. Read those for what they are: country medians for backend developers in general, and gross salary at that. Blockchain specialists cost more, employer costs sit on top, and a median is one person.
That team also outlives the launch. Networks change their fees, new token standards appear, and a customer sends USDT on a network you do not support — someone has to answer that merchant by Tuesday. Those people are on the payroll in month thirty as surely as in month one.
Against that stands a percentage of your turnover. Published infrastructure rates run from 0.4% to 2% — the basis is the public price lists of infrastructure providers, checked on 22 August 2026. Inside that range the rate steps down as volume goes up, and one of those price lists prints the steps:
Your own quote lands in there depending on volume, assets and settlement currency.
Both lines, same page. The fee is tied to your turnover; payroll is not. That is why a crossing point exists, and here is how to put a figure on yours.
The fee side is arithmetic. At $1 million a month you are on the top published step: 1% plus $0.25 per transaction. Call it $10,000 a month, roughly $120,000 a year before the per-transaction cents.
The build side is your headcount times a median. Pick the N you would actually hire, because nobody can pick it for you: engineering, plus whoever carries security and on-call. Say you model three people. At the UK median that is about $327,000 a year in gross salary, and at a 1% rate the fee line only reaches it at roughly $2.7 million a month of processed volume.
And that comparison still flatters the build, because the two sides are not on the same basis. The $327,000 is gross salary and nothing else, while the fee at least buys a running service. To compare like with like, add employer costs, security review and key management, on-call rotation, incident work, AML tooling, reconciliation and reporting. Add the support hours that arrive with every network edge case. Compute the point against the full cost of your own solution, not the budget line labelled "development". The guide to why crypto payments fail is a fair inventory of what that support queue inherits.
So the crossing point is real, and for most PSPs it sits in the millions per month. Which gives three answers rather than two.
Below the thresholds, without your own authorisation. Stop framing it as build versus buy. You are choosing which partner carries the licence and the infrastructure while you carry the merchants, and this year's goal is the volume that unlocks partner terms. Get there, then renegotiate. CryptumPay has a White Label mode.
Volume yes, authorisation no. Do not start the build this year. Your constraint is a calendar you do not control: three months of assessment that begins only once the file is complete, and one applicant in six reaching the UK register. Buy the rails, run the application in parallel, revisit the arithmetic when the permission is in hand.
Above the thresholds and already authorised. The calculation is worth doing properly. Build if crypto processing is what you intend to sell as your own technology, and if your volume puts the fee line clearly above a fully costed team. Otherwise buy, and spend the engineering you saved on merchant acquisition, where your margin actually lives.
Whichever way the numbers fall, one decision is not optional. It is also the cheapest insurance in this whole exercise: buy the infrastructure, keep the business logic.
From day one your own platform holds all of this, whatever the provider's dashboard shows as well:
It costs a few weeks of engineering at the start. It is exactly what makes the second calculation possible a year later. Without your own operational history you cannot compute the switch, cannot migrate and cannot add a second provider.
Which raises a data question for any provider before signing: what am I actually receiving, and when? The workable baseline: API, widget and payment links, a webhook on every status change, and a lifecycle you can map onto your own order states. CryptumPay, for instance, exposes the stages as created, pending, crediting and finished. Whatever a provider calls them, every transition has to arrive in your system by webhook. Run that technical pass before the contract: the crypto payment API checklist is the list to hand your engineers.
Risk and support stay yours too. When funds are held, your merchant calls you. So learn the mechanics you resell: in CryptumPay, suspicious funds are held before they reach the balance, and two-factor authentication applies at login and at withdrawal. Ask any shortlisted provider to describe its equivalents as plainly. The background is in the guide to AML and KYC in crypto payments. The support queue is the same story — wrong network, missing gas, underpayment by a few cents, an expired invoice. Most of it is USDT network mechanics. Read USDT without gas before you write your merchant instructions.
Infrastructure — wallets, blockchain monitoring, status logic, settlement — supplied by a specialist and sold onward under your brand. The merchant sees your payment page and your reporting. Commercially, what matters is the paperwork underneath: a partner contract with resale terms, or an ordinary merchant account with a logo swap.
Bought infrastructure does not transfer anyone's permission to you. What matters is who is the crypto-asset service provider to the merchant under the contract. In the EU, the MiCA grandfathering window closed on 30 June 2026, so since 1 July a firm that was trading on a national permission needs CASP authorisation to carry on. Settle the contract shape with counsel before launch, in your specific jurisdiction; treat this as background for that conversation.
Whatever you charge, minus your step on the rate card, minus support. See it per quarter-point. At $1 million a month, each extra 0.25 percentage point on your merchant price is $2,500 a month of gross margin, and your own cost there is 1% plus a fixed few cents per transaction. So your pricing power lives in the spread, and crossing the $1 million step cuts your own cost from 1.5% to 1%.
The rails are liquid: USDC alone had $72.7 billion in circulation as of 20 August 2026, according to issuer Circle. But that is coins outstanding, and it says nothing about how much moves through checkouts or what share is B2B. Your decision runs on your own number: how many merchants activate it, and what they process in the first quarter.
Store your own merchant, customer and order identifiers, your own status mapping and your own event history from the first transaction. Check the export path before you sign. A provider whose data lives only in its dashboard is one you cannot leave. It is also one you cannot benchmark when you sit down to compute whether building has started to pay.
Create an account and connect the checkout yourself, or talk to sales and we will plan the integration with you.
