Telcos Are Trying to Become Banks Again

In December 2025 the French banking regulator withdrew Orange Bank’s licence. What is left is a dormant shell, after losses that ran to roughly a billion euros.

A couple of months later e& money, the financial arm of e&, the UAE’s largest telecom operator and formerly Etisalat, announced that the Central Bank of the UAE had approved a Finance Company licence allowing it to move into lending. More than two million existing customers; credit cards, buy-now-pay-later, early wage access and other credit products to follow.

Telco operators have been trying to become banks for fifteen years on an argument that has never changed: we already have the customers.

The first wave lost, and lost badly. The second wave is under way now, in the Gulf and in South Asia, with better licences and much better products.

Why did companies with the best distribution in the world keep failing at this, and what is different about the ones now succeeding?

Distribution gets you the customer only where the customer has nowhere better to go. In a mature market, where most people change bank account once or twice in a lifetime, the largest retail footprint in the country buys you almost nothing, which is the lesson Orange paid a billion euros to learn.

It works where banking is thin, or when the product is worth moving for (so far, that means yield). And even where it works, it does not tell you who captures the economics.

Two things decide that instead. First, whose money is on the line when a loan goes bad, the telco’s or a partner bank’s. Second, what the telco has left to sell once the central bank starts moving money for free. Brazil did this with Pix in 2020: instant account-to-account transfers became free for individuals, compressing transfer fees banks had previously earned. Large banks could absorb that because payments are only one part of a business that also earns from deposits, lending and other services. A wallet that only moves money has no such cushion.

Kenya is where the modern version of this began.

Safaricom is the country’s dominant mobile network operator, part-owned by the Kenyan government and by Vodacom. In 2007 it launched M-Pesa, which at the time meant something very simple: you handed cash to a shopkeeper who was signed up as an agent, electronic value appeared in your wallet, and you could send that value to anybody else with a phone. No bank account, no branch, no smartphone. It was e-money built for basic phones, with cash-in and cash-out handled by the agent network.

This worked because there was nothing else. Kenyan bank branches were concentrated in cities, account opening required paperwork most people did not have, and the alternative for sending money home to a village was handing an envelope to a bus driver and hoping.

M-Pesa was not competing with a bank account. It was competing with a bus driver.

It went on to become one of the most successful mobile-money systems ever built, and it is now embedded in Kenyan life to a degree that is hard to convey. Rent, salaries, school fees, matatu minibus fares, market stalls. Last time I was there Kenyans told me they hadn’t touched physical money in years.

Then came the credit products, which is where my understanding fell apart.

M-Shwari arrived in 2012, a savings and small-loan product inside the M-Pesa menu. KCB M-Pesa followed. Then Fuliza in 2019, an overdraft that quietly tops up your wallet when you are a hundred shillings short at the till. To a Kenyan customer these all look like Safaricom products. They sit inside the Safaricom app, they carry M-Pesa branding, and the money moves through M-Pesa rails.

Safaricom does not lend any of it.

M-Shwari is underwritten by NCBA, a Kenyan commercial bank formed from a 2019 merger of NIC Bank and Commercial Bank of Africa. KCB M-Pesa is underwritten by KCB, the largest bank in Kenya by assets. Fuliza is underwritten by both of them together: Safaricom’s own terms say the service is offered by KCB and NCBA under licence from the Central Bank of Kenya.

Reporting at launch put the revenue split at forty percent Safaricom, forty percent NCBA and twenty percent KCB. Safaricom’s forty percent is a share of the fees, not of the loans, which took me longer to notice than I would like to admit. NCBA and KCB are the ones putting up the money.

So two companies sit behind one line on one screen, and they are in completely different businesses.

When you borrow through Fuliza, a bank funds the advance out of deposits it holds, and books it as a loan, which is an asset on that bank’s balance sheet. If you never repay, the bank writes the asset down, and the write-down comes out of the bank’s capital. Safaricom’s accounts never feel it. Safaricom collects its share of the fee whether the loan performs or not.

There is a second version of the same arrangement, hiding in plain sight, which is the float. The cash backing everybody’s M-Pesa balances sits in pooled trust accounts at commercial banks. The interest earned on that pile is paid into a not-for-profit trust, and Safaricom cannot take it. That constraint is the specific device that allows Safaricom to run the largest deposit-like system in East Africa without being regulated as a bank, which the World Bank documented years ago and which almost nobody quotes.

Safaricom does not keep the interest on the M-Pesa trust float, and it does not book the Fuliza loan principal or the resulting credit loss. It is a distribution company with a very good fee line.

If you have ever bought insurance you already know the cleanest analogy for this.

An insurance broker sells you the policy and takes a commission. An underwriter pays the claim. Same customer, same product, same piece of paper, and only one of the two is awake at three in the morning when a hurricane forms.

Most embedded-finance arrangements are some version of that split, and the marketing is designed to blur it (that is what marketing is for). When a phone company, a retailer, an airline or a ride-hailing app announces that it has “launched credit”, what has usually happened is that somebody else has agreed to hold the loans and the announcing company has agreed to introduce the customers.

Three things fall out of knowing which side a company sits on, and all three matter to anyone valuing it.

You learn what a downturn does to it. A wave of defaults does not hit Safaricom’s capital through Fuliza loan write-downs, because Safaricom does not book those loans. A licensed bank has no such insulation.

The party eating the losses writes the underwriting rules, sets the limits and can withdraw the product. However prominent the logo on the app, a broker is a tenant on somebody else’s balance sheet.

And you learn what kind of growth you are looking at. A distributor with no capital requirement can grow until it runs out of human beings. A lender stops when it runs out of equity. Those two businesses get valued on different multiples for good reason, and a company that is quietly one while sounding like the other is mispriced in one direction or the other.

When a company tells you it has launched credit, ask whose balance sheet absorbs the loss. If the answer is somebody else’s, you are usually looking at a marketing channel with unusually good margins.

Orange is the largest mobile operator in France. In 2016 it bought control of Groupama Banque, an existing French retail bank, and in November 2017 it launched Orange Bank, with the reasoning that should by now sound familiar: tens of millions of customers, a bill-paying relationship with each of them, shops on every high street, and banking is only software.

It passed a million customers by July 2020, which sounds like success and was not. In June 2023 Orange opened exclusive talks with BNP Paribas to wind the thing down, agreements were signed in February 2024, the Spanish operation stopped in June 2024, and in December 2025 the French banking regulator, the ACPR, formally withdrew the licence. The shell that remains is called Orange OBK and does nothing.

The losses run to somewhere around a billion euros, which makes this one of the most expensive strategic mistakes in European telecoms, and it was made by people who could all recite the M-Pesa case study.

Orange did the harder thing. It bought an underwriter, took the credit risk onto its own books, held capital against it, and hired the people. Safaricom did the easier thing and rented the risk out.

Orange also faced the opposite distribution problem. Safaricom was selling access to people who often had no good alternative; Orange was selling a slightly nicer current account to people who already had one. A French customer never lacked access to a bank. The branch was around the corner, the account worked, the money was insured, and switching current accounts is one of the most reluctantly performed acts in consumer finance.

Distribution is only worth something when it is scarce. In Kenya in 2007 it was the scarcest thing in the economy. In France in 2017 it was free.

(I would love to know what the internal memo said. Somebody at Orange must have written the sentence “we have more customers than any bank in France” in a deck, and somebody more senior must have nodded.)

The Kenyan model has a hidden assumption in it: that the state stays out of the way. India shows what happens when it does not.

Around 2016 India had the same patchwork everybody else had: bank apps that only talked to themselves, and a pile of closed mobile wallets, several of them run by telecom operators. Airtel Money, Vodafone m-pesa, Idea Money, Jio Money. You could not send money from one wallet to another. Each one was a walled garden and the walls were the business model.

Then the National Payments Corporation of India, a not-for-profit set up by the central bank and the banks’ own association, launched the Unified Payments Interface. UPI is what the industry calls a rail: the underlying system that moves the money, on top of which everybody else builds their apps. It let any bank account talk to any other bank account through one open protocol, addressable by a simple handle (something like an email address for money) rather than an account number, instant, and free to the user. For years India also held the merchant discount rate at zero across UPI, turning the transfer layer into something close to a public utility.

A columnist in Business Standard wrote in September 2016 that UPI would probably supersede mobile wallets. He was right, and the scale of how right he was is difficult to hold in your head.

In January 2026 alone, UPI processed 21.7 billion transactions. It accounted for around eighty-one percent of India’s retail digital payments by volume in FY2024-25, and around 228 billion transactions ran across it in 2025. The IMF has described UPI as the world’s largest retail fast-payment system by transaction volume.

The telco wallets did not survive it. Vodafone’s Indian m-pesa venture was wound down around 2019 with an impairment of roughly two billion rupees booked alongside the collapse of the associated payments bank. The walled gardens had been built to solve a problem the state then solved for everybody.

The companies that won the resulting market are not making much money either. The top two apps, PhonePe and Google Pay, carry more than eighty percent of UPI volume between them (one owned by Walmart, one by Alphabet, which is its own story about who ends up running a nation’s payments), and for years they carried it on rails they did not build and could barely monetise. Mastercard’s chief financial officer said out loud in 2023 what everyone in the industry was saying privately, that UPI had been an “incredibly painful experience” for the firms participating in it.

Winning the distribution of a free rail turns out to be an expensive hobby.

Then, while I was finishing this article, the model changed. On 15 September 2026 India announced a limited return of merchant discount rates from 15 October for specified merchant payments above ₹2,000. Person-to-person payments remain free, small merchants remain protected, and the government says approximately ninety-six percent of merchant transactions will remain unaffected. Fifteen months earlier the Finance Ministry had called reports that MDR would ever be charged on UPI “completely false, baseless and misleading”. None of this rebuilds the old walled gardens. It does make an important concession: even a public rail has to fund itself.

None of this proves that state rails destroy value. UPI created an enormous amount of value, much of it captured by consumers and merchants through cheaper, easier payments. What it shows is that value can stop accruing to whoever owns the interface.

Yes, in Pakistan, and the timing is interesting because the wallets got big first and the rail arrived second.

Pakistan’s two large mobile wallets are JazzCash, run by Jazz, the country’s largest mobile operator, and Easypaisa, its longer-established rival. JazzCash reports over sixty million customers. Together the wallet sector runs to more than eighty million active accounts on the State Bank of Pakistan’s numbers, in a country where formal bank account ownership was until recently among the lowest in Asia. Classic Kenyan setup: no incumbent worth the name, distribution scarce, wallet wins.

Then the State Bank of Pakistan, the central bank, built Raast.

Raast is the same idea as UPI. Instant, twenty-four hours a day, addressable by phone number, connecting banks and wallets alike, and free at the point of use with no merchant discount rate. It has scaled fast: something close to fifty trillion rupees across nearly two billion transactions in 2025 by the central bank’s own reporting, roughly a hundred and eighty billion dollars of gross payment value. By the first quarter of 2026, ninety-two percent of Pakistani retail payment transactions were digital by volume, though only around forty percent by value (bank branches still handled nearly a hundred trillion rupees in the same quarter, so the big money still walks in through a door).

The instinct for someone in London or Chicago is to file this under financial inclusion in a developing market, and that instinct is backwards.

Pakistan has instant, free, phone-number-addressed payments. Kenya solved retail payments on feature phones while the United States was still posting cheques (some of it still is). The executive reading this is paying somewhere around two percent to card networks for the privilege of being fifteen years behind on instant account-to-account payments. These are among the most advanced consumer payment systems in the world, and the firms operating inside them are living in our future.

What Raast did to JazzCash is visible in the product itself. Both big wallets now route free Raast transfers out to bank accounts, because they have no choice: a rival that charges for something the central bank provides for nothing does not stay a rival for long. The transfer fee, which was the original business, is being pushed toward zero by a state that offers the same thing for nothing.

Jazz is owned by VEON, a group that deserves a sentence of introduction because it should be a household name by now and it isn’t. VEON is listed on Nasdaq, headquartered in Dubai, and operates mobile networks in Pakistan, Ukraine, Bangladesh, Kazakhstan and Uzbekistan, serving around a hundred and fifty million connectivity customers. No state owns it. It is run by Kaan Terzioglu, who before joining in 2019 spent four years as chief executive of Turkcell, Turkey’s largest operator, which means the man exporting this playbook across five countries learned it in the market that went furthest with it first.

The first thing VEON did differently was refuse to be a broker.

Rather than renting a bank the way Safaricom does, VEON owns one outright. Mobilink Microfinance Bank has been operating in Pakistan since April 2012, reports over forty-two million registered users and sixteen million active digital wallets, and took an Islamic banking licence from the State Bank of Pakistan during 2025. A microfinance banking licence is a lower tier than a full commercial banking licence, with tighter limits, which is precisely why it was attainable.

You can see the consequence in the group accounts, and this is my favourite detail in the whole piece. In VEON’s second quarter 2026 results there is a line, in a footnote naturally, explaining that around five hundred and ten million dollars of group cash relates to the Pakistani banking operations and has to be excluded when calculating net debt, because a bank’s deposits are not the group’s money to spend. That is a telecom company explaining ring-fenced bank funding to equity analysts. The accounts stop looking like a telco’s the moment you stop being the broker (and based on current share price, I think analysts are not valuing VEON correctly but that’s not my lane…)

Then the products. With transfers going free, what the wallets now advertise is what the rail cannot provide on its own: a return on the balance, and credit against it.

And in June 2026, JazzCash started selling government debt.

Any verified customer with an active Mobilink Bank deposit account can now buy three-month Pakistani Treasury Bills inside the app, from five thousand rupees, which is about eighteen dollars. It was built with the Ministry of Finance and the central bank, Mobilink Bank acts as the regulated custodian and trustee, and the stated target is a million active investors.

Strip the branding off and a mobile network operator has turned its wallet into a retail distribution and custody channel for short-term government debt, available to tens of millions of people who have never owned a government security. The demand being served is yield, and delivering it needs regulated custody and a licensed institution behind it, which is why the broker model could never have delivered it.

There is a version of this move that is a hundred and fifty years old. Western Union built a telegraph network, and in 1871 started moving money across it for the simple reason that it already owned the wires. The telegram business died slowly and then completely, the last one sent in 2006, while the money transfer business outlived it by generations. The communications half was the half that did not make it. Western Union today owns fifteen percent of stc bank in Saudi Arabia, the licensed digital bank that the Saudi telecom operator’s wallet turned into, which means the company that made this exact transition before the invention of the telephone is now on the cap table of a telco making it in 2026.

e& is the other route to the same place. VEON got there by owning a bank. e& money is the state-linked route: the operator sits inside a majority state-owned group, and a Finance Company licence lets it put credit on its own balance sheet without buying a bank. Two routes, and the thing they share is that both companies stopped being brokers. The announcement quoted Khalifa Al Shamsi, who runs the division, saying the gap in UAE credit is “not a risk issue, it’s an access issue”. I would frame that as the single most testable sentence any operator has said this year. If he is right, data-rich underwriting has changed what a telco can safely lend against. If he is wrong, e& has just taken on the balance sheet Orange took on, with a better app and a friendlier regulator.

I wrote about this in Follow the Float, which argued that the significant change in the new stablecoin consortium model is the migration of float income from the issuer to the distributors. That piece asked who keeps the money while it sits still. This one asks who eats the money when it disappears. Both are questions about whether the brand on the front of the product is the party at risk behind it, and in payments the answer is usually no.

Stablecoins are the cross-border version of the same clock. They do not make FX, compliance or credit free. They do make the transport layer more open and more contestable. If a telco can move dollar value between two regulated wallets without paying the old correspondent-banking tolls, the value migrates again: toward the licence, the FX spread, custody, underwriting and the idle float.

There is an earlier version of this in Indonesia. In 2022 Fasset partnered with Indosat Ooredoo Hutchison to put its digital-asset platform inside the telco’s myIM3 and bima+ apps rather than build distribution from scratch. When Fasset launched in Indonesia the following year, it said more than a million people joined the waitlist in its first week. Fasset brought the financial product; Indosat brought the customers.

Three years after that Indonesian launch, Fasset was valued at $1 billion. It now describes its infrastructure as a network connecting banks, telcos, payment providers and liquidity providers across more than 100 banking corridors.

On 15 September a startup called Iris came out of stealth with a forty-three million dollar commitment from Balesia Group, a family office with telecom businesses across the Americas, and announced that VIVA, a Bolivian mobile operator with around two million customers, was the first carrier on its network. VIVA settles in USDi, a dollar stablecoin issued and redeemed by Agora, and can hold its operating reserves in dollars rather than bolivianos. Agora’s co-founder summarised the pitch on X: operators bring “customers, verified identities and distribution”.

Agora issues and redeems the token, so Agora’s reserves are the balance sheet, and VIVA is the distributor with its name on the wallet (Balesia also owns most of VIVA, so the network’s first customer is its funder’s own telco, which says something about conviction and nothing yet about demand). That is the Safaricom arrangement with a stablecoin where the trust account used to be. Bolivia ended its fifteen-year dollar peg in June, and the dollar rose roughly forty percent against the boliviano within a few weeks. The country had already been dealing with a prolonged shortage of dollars, and virtual-asset transactions through electronic payment channels have been permitted since June 2024.

The argument so far says a free state rail takes the wallet’s transfer business away and leaves it selling only the things that need a licence, a balance sheet or both. Two pieces of evidence point the other way, and one of them is uncomfortable.

Turkey is the first. Turkcell’s payments arm, Paycell, operates under payment and electronic money licences from the Turkish central bank, and its consumer lending sits in a separate finance company, Financell, because Turkish law will not let one entity hold both (sensibly). The Turkish central bank launched its instant payment system, FAST, in 2021, with a directory, run by the banks’ shared card-processing body, that lets any bank or e-money firm register your phone number against your account. On my theory Paycell should have been hollowed out. It has instead kept expanding into Europe and Northern Cyprus, and it is open to customers of rival operators. Perhaps the rail takes the transfer without touching the relationship, and the wallet survives by moving into merchant services and credit.

Pakistan’s own numbers half-support this idea. In the first quarter of 2026 Raast handled 742 million transactions, and 664 million of them were person to person. Payments to merchants came to 56 million. The state rail has eaten the transfer and barely touched the shop counter, which is where the wallets and the card schemes still live. Whether that is a lag or a limit is the whole question.

The second is a pilot study published this year in the Journal of Pakistan Administration, which looked at why informal-sector workers keep using JazzCash and Easypaisa when Raast is free, and found a strong negative correlation between fear of state supervision and use of the state rail. The sample was sixty people, small enough that nobody should build a thesis on it. The direction is plausible, though. A central bank rail is free and legible. A wallet is habitual, agent-based, and feels less visible to the tax authority.

If that holds, part of the telco wallet’s residual moat may be illegibility, and the business model becomes, at least in part, a bet against the formalisation of the economy. That is a far less comfortable article than the one I have written, and possibly the truer one.

My prediction, moderate confidence: within three years at least two more large emerging-market mobile operators will acquire or apply for a balance-sheet lending or deposit-taking licence rather than remain pure distributors, and will say publicly that it is about financial inclusion. (It will be about the yield.) Watch the Airtel Money listing. Airtel Africa, which operates across fourteen African countries, confirmed London as its preferred venue in July 2026 and intends to list before the end of the year, subject to market conditions. If that business reaches the market still looking primarily like a payments platform, with lending sitting on partner balance sheets, investors will get a clean chance to price how long the fee annuity can last.

Lower confidence: the first operator to move money between its own wallets across a border using a dollar stablecoin will be one whose home currency is failing rather than one whose regulator is friendliest.

The next time a company tells you it has launched credit, or embedded finance, or banking, ask who writes the loan down when it goes bad.

Then ask: how long until the state builds the thing this company is charging for? India did it with UPI. Brazil with Pix. Thailand with PromptPay. Turkey with FAST. Pakistan with Raast. The architectures differ: FedNow in the United States is an interbank instant-payment service, not UPI with an American flag on it, and Europe’s digital euro is still in preparation.

Every distribution business in payments is running down the same clock. The firms that keep their position are the ones that used the good years to buy something the rail cannot give away. In every case I have looked at, that something has a licence attached and a balance sheet behind it.

Disclosure: I work on settlement infrastructure in Abu Dhabi, including regulated stablecoin projects.

See you next week.

James Smith

Usually not. In Kenya, M-Shwari, KCB M-Pesa and Fuliza all appear inside the Safaricom app under M-Pesa branding, but they are underwritten by the banks NCBA and KCB. Reporting at launch put the Fuliza revenue split at 40% Safaricom, 40% NCBA and 20% KCB, and Safaricom's share is of the fees, not the loans. The bank books the advance as an asset and absorbs the write-down if it is never repaid. Safaricom collects its fee either way.

Orange bought an underwriter rather than renting one: it took control of Groupama Banque in 2016, launched Orange Bank in November 2017, and held the credit risk and the capital itself. It also had the opposite distribution problem to Safaricom, selling a slightly better current account to French customers who already had a perfectly good one. The French regulator withdrew the licence in December 2025, with losses of roughly a billion euros. Distribution is only worth something where it is scarce.

It removed their reason to exist. Before 2016 India had a set of closed telco wallets that could not send money to each other, and the walls were the business model. UPI made transfers interoperable, instant and free, and for years held the merchant discount rate at zero. Vodafone's Indian m-pesa venture was wound down around 2019. In January 2026 alone UPI processed 21.7 billion transactions.

By selling the things a public rail cannot give away, which in practice means yield and credit, and both need a licence or a balance sheet behind them. In June 2026 JazzCash began selling three-month Pakistani Treasury Bills inside the app from around eighteen dollars, with Mobilink Bank acting as the regulated custodian. The wallet became a retail distribution channel for sovereign debt.

The broker introduces the customer and takes a commission. The underwriter holds the loan and absorbs the loss. It decides three things: what a downturn does to the company, who writes the underwriting rules and can withdraw the product, and whether growth is capped by equity or only by the number of human beings available. The two get valued on different multiples, which is why a company that is quietly one while sounding like the other is mispriced.

Follow the Float: Who Earns the Yield on OUSD asks the companion question: not who eats the loss, but who keeps the money while it sits still.

Blockchain Makes FX Cheaper. It Doesn't Make Liquidity Free. traces the same point through a cross-border payment: the transport layer gets cheaper, the funding does not.

Swift Never Moved Money is the institutional version: a coordinator that never becomes the settlement asset.