368 Neobanks, 127 Licences: What the Data Shows
An open dataset tracking 368 active neobanks found only 127 hold a full banking licence. What that means for the 1.46 billion people banking through an app, and why the failures stay invisible.
Last updated: 28 July 2026
Most writing about neobanks counts funding rounds. An open dataset published in July 2026 by neobankbeat did something more useful: it counted the companies, and the ones that quietly disappeared. The headline finding is uncomfortable.
Of 368 verified-active neobanks, only 127 hold a full banking licence. Two thirds of the "banks" in the app stores are not banks. They rent the right to exist from a sponsor bank, an e-money licence or a card issuer, and their customers almost never know which side of that line they are standing on.
The dataset is MIT-licensed and reproducible, which is rare in an industry where the equivalent analysis usually sits behind a four-figure paywall. Below are the findings we think matter most, and what they imply about where you keep money.
The rise is real
It is worth stating plainly, because the licensing point that follows is not an argument that neobanks failed. They did not. Summing company-reported figures across the dataset gives roughly 1.46 billion users. That is reported customers, not a market-size projection.
The geography also corrects a bias in most Western coverage.
| Region or company | Reported users |
|---|---|
| Asia (total) | ~817 million |
| WeBank alone | 400 million+, more than every US and European neobank combined |
| Nubank | 131 million, more than all US neobanks together |
| Revolut | 50 million+ |
For a very large number of these people, an app was the first bank account they ever had. That is a genuine achievement and it should not be lost in the risk discussion.
The licence gap
The structural problem is what sits underneath the app. A neobank without its own charter depends on a partner: a sponsor bank holding the deposits, a BaaS platform running the ledger, a processor issuing the cards. That dependency is invisible from the customer side and decisive when it breaks.
The dataset's framing of this is the sharpest line in the whole analysis: when a real bank fails, deposit insurance pays out; when a neobank's infrastructure fails, customers get a queue number in a bankruptcy proceeding.
There is precedent, repeatedly:
- WaveCrest, 2018. Visa pulled one issuer and dozens of crypto card programmes died overnight.
- Wirecard, 2020. A 1.9 billion euro hole in a single processor froze products across Europe whose only mistake was building on it.
- Synapse, 2024. A banking-as-a-service middleman collapsed and American customers discovered that "FDIC insured" did not mean what they assumed, because the failure was in the ledger recording whose dollars were whose.
- Ready, 2026. The same pattern again.
Note the shape of each: the consumer brand was not the thing that failed. The layer beneath it was. The dataset counts 106 infrastructure providers carrying 368 consumer brands, with a handful of sponsor banks and processors each supporting dozens. That concentration is exactly what produces the next incident.
Why you never hear about the failures
Five entities came off the list in a single month, liquidated, absorbed or quietly pivoted. No press releases, no post-mortems.
Neobanks do not fail loudly. The app stops updating, support stops replying, and eventually the domain redirects to a partner's landing page. Fintech media covers launches and funding rounds because that is where access and advertising are, so the graveyard stays invisible and each new founder walks into the same traps believing they are the first to see them.
This is a data-availability problem with real consequences: customers cannot price a risk that is never reported.
The part that surprised us
Of neobanks founded in the 2020s that are still operating, 30% are web3-native self-custodial apps where no company holds the balance at all. In the 2010s cohort that figure was 4%.
The dataset's own composition reflects it: 254 traditional custodial challengers, 58 hybrid fiat-and-crypto apps, and 56 web3-native. The marginal energy in the category has moved toward architectures that do not take custody, and it has done so precisely as the custodial layer's failure modes became visible.
That is not a crypto-maximalist reading. It is what the founders of the last five years actually built.
The AI claims do not survive an audit
One more finding worth repeating. Every neobank pitch deck now mentions AI. The dataset audited all 368 against filings, regulator disclosures and production evidence rather than marketing pages. 67 passed. Eighteen percent. The rest were piloting, exploring, or presenting a partner's model as their own.
The leaders are also not the famous names. They are emerging-market lenders in Nigeria, the Philippines, Mexico and Bangladesh, where credit bureaus are thin and a model that can underwrite a first-time borrower is not a feature but the reason the business exists.
What this means for where you keep money
The practical takeaway is not "avoid neobanks". They are good at what they do, and for daily spending, salary and cards they are usually better than the incumbents. The takeaway is narrower and harder to argue with: know which entity actually holds your money, and what happens if the layer beneath it fails.
- Check the licence. Does the app hold a banking charter, or does it name a partner bank in the small print? Both can be fine; only one of them is a bank.
- Find out who the partner is. If a BaaS provider or sponsor bank sits underneath, that company is your real counterparty.
- Understand the insurance. Pass-through deposit insurance depends on accurate records at the intermediary. Synapse is the case study in what happens when those records fail.
- Do not concentrate. The single control you fully own is how much sits with any one provider.
This is the same analysis we applied to the 2026 crypto exchange closures, and the answer rhymes. A balance held by a company is a claim against that company. In an orderly wind-down the claim is honoured; in a disorderly one it becomes an unsecured claim among many.
The self-custodial alternative
The 30% figure above describes the structural response: applications where no company holds the balance, so there is no intermediary whose failure can strand it. JewelSwap sits in that category. It is non-custodial across MultiversX, Sui and Radix, with users interacting with smart contracts from wallets they control, and it does not take custody of user assets.
It is also not a bank, and does not attempt to be one. There is no card, no salary deposit, no insurance and no support desk that can reverse a mistake. Smart contract risk and user error replace counterparty risk rather than eliminating risk altogether. For most people the sensible arrangement is exactly the split the data implies: a neobank for the fiat side of life, self-custody for the part of the balance sheet you want no company to be able to freeze. Our guide to self-custody covers doing that properly.
Frequently asked questions
How many neobanks are there?
The neobankbeat dataset tracked 368 verified-active neobanks as of July 2026, split into 254 traditional custodial challengers, 58 hybrid fiat-and-crypto apps and 56 web3-native self-custodial applications, supported by 106 infrastructure providers.
Are neobanks actually banks?
Often not. The dataset found only 127 of 368 hold a full banking licence. The remainder operate through a sponsor bank, an e-money licence or a card issuer. The customer experience is identical, but the entity holding the money and the protection that applies are not.
Is my money safe in a neobank?
It depends on who holds it. With a licensed bank, deposit insurance applies directly. With a partner-bank model, protection depends on accurate records at the intermediary, which is precisely what failed in the Synapse collapse in 2024. Check which arrangement your provider uses.
How many neobank users are there worldwide?
Roughly 1.46 billion, summing company-reported figures across the dataset. About 817 million are in Asia. WeBank alone reports more than 400 million users, exceeding every US and European neobank combined.
Why do neobank failures go unreported?
Because they are gradual rather than dramatic. Apps stop updating, support stops responding, and domains eventually redirect to a partner. Media coverage concentrates on launches and funding, so exits are rarely recorded, which leaves customers unable to price the risk.
Do most neobanks really use AI?
No. An audit of all 368 against filings, regulator disclosures and production evidence found only 67, about 18%, had AI genuinely in production. The strongest implementations were emerging-market lenders using models to underwrite borrowers with no conventional credit file.
Keep reading
