Kaspi.kz

Imagine that Visa, Amazon and your bank were one company, that the company had one app, and that three out of every four people in your country opened that app every day — to pay for coffee, buy a fridge on ten interest-free instalments, send money to their mother, renew a driving licence, book a flight and keep their savings. That is Kaspi.kz in Kazakhstan. Roughly 15 million monthly users and 10 million daily users in a country of 20 million (approx. — company data); 77 transactions a month per active customer; about seventy percent of the nation's non-cash retail payments and close to half of its e-commerce (approx. — third-party estimates). A country that went from mostly cash to mostly cashless in a decade did so largely inside one company's software.

The financial result is what you would expect of a monopoly-shaped network: in fiscal 2025, revenue of KZT 3.9 trillion (about $8.5 billion), net income of KZT 1.04 trillion (about $2.3 billion), a return on equity of 41% on the trailing twelve months, and — the number that separates it from every bank Buffett ever disliked — a cost of credit risk of 0.7% a quarter, because the lender can see every purchase its borrowers make. Revenue is up seven and a half times since 2019; earnings per share five and a half times.

And the shares trade at eight times earnings, with a dividend that at the newly raised rate yields about nine percent. The honest description of a share of KSPI today, then, is this: a national utility with software margins, priced like a Central Asian bank, because it is domiciled in one. Mohnish Pabrai, who put it in his fund this spring, calls it "heads I win; tails I do not lose much." This report is an attempt to find out what "tails" actually looks like — and it is not, we will argue, Turkey.

Two lessons for an owner. First, the moat was built in a vacuum: Kaspi digitised a country that had no incumbents worth the name, which is why it could take seventy percent of payments and half of e-commerce in a decade — and why the same feat in Turkey, against Alibaba's Trendyol, is a different sport. Second, the founders have always used the public markets as an exit, not a source of capital: the London IPO, the Nasdaq listing and the 2024 block trades were all secondary; Baring has been selling for twenty years; Kim sells on a schedule. That is not a governance failure — it is the normal life-cycle of a private-equity-backed founder company — but it means the share register tells you who is leaving, not who is arriving. The exception, this April, was Tencent, and the CEO buying with his own money.

Our segment feed returns nothing for Kaspi — the platform disclosures do not map to the standard schema — so the mix below is from the Q2 2026 release. Revenue of KZT 1,112bn in the quarter, up 15%:

How to read a company that is one-third bank. This is the metric-teaching part every new business type earns on our board — banks got NII and ROTCE, insurers got float — and for Kaspi the lesson is what not to read. Our feed prints a return on invested capital of 11%, an Altman-Z of 1.9 (the "distress zone"), interest cover of 1.6× and receivables of 634 days. All four are the model treating a bank as if it were a factory: loans are "receivables," deposits are "debt," interest paid to savers is "interest expense." Every one of those numbers is meaningless here, exactly as they were at Bank of America. The four that matter are the ones a payments-and-lending operator watches: the take rate (payments 1.00%, marketplace 12.1%, both moving in the direction you would expect — down where the state competes, up where advertising grows), the cost of funding (14.5%, the price of a central bank that held rates at 18% into 2026), the cost of risk (0.7% a quarter — extraordinary for unsecured consumer credit, and the direct product of seeing every purchase), and leverage: equity of KZT 2.7tn against assets of ~KZT 12tn, about 4.4×. Buffett's 1990 warning about banking was assets "twenty times equity." Kaspi is a bank levered like a retailer.

Why the profit stopped growing while the business did not. Revenue is up 15%; net income is flat; the margin fell from 26.7% to 23.3% in a year. Three reasons, all named by management: the bank corporate tax rose from 20% to 25% in January; the National Bank raised reserve requirements twice; and Hepsiburada's consolidation brings first-party retail revenue at retail margins plus a Turkish loss. None of the three is the flywheel slowing. All three are the price of the jurisdiction and the acquisition — which is the theme of this report.

Buffett's test — would a competitor with money, talent and will want to attack this? — has a plain answer in Kazakhstan: they have tried, and the daily-user gap (10 million against Halyk's 2.8 million) says how it went. The competitor that matters is not a bank. It is the state, which in July built the one thing Kaspi's payments moat depended on nobody else building — a QR code that works across every bank — and did it in the name of "technological independence." That is what a government does to a national champion that has become the plumbing: it does not break it, it makes it share. The payments take rate, already drifting from 1.07% to 1.00%, is the line it will show up in. Score: 8 — as wide as Visa's at home, and, unlike Visa's, subject to a sovereign who can redraw it by decree. Not narrower than Alphabet's or Nvidia's, but contested by a different kind of rival.

Mohnish Pabrai's thesis, in his own words from his fund's March call: "It is more of a 'heads I win; tails I do not lose much' type of bet… if they fail [in Turkey], I do not believe we lose much because they still have a great business and a strong position in Kazakhstan." And in June: "We were buying into the Kazakh business at a low multiple… We have a moonshot with Turkey." That is a good thesis, and we think it has the coin the wrong way up. Turkey is not the tails. Turkey is a small heads. The tails is Kazakhstan.

Why 8×, honestly. The Kazakh flywheel alone earned KZT ~1.04tn last year — about $2.3 billion — and a network of this quality, growing volumes at 15–20%, would command 15–20× in Ohio. It commands 8× in Almaty for reasons that are all real: a currency that has devalued in steps five times in twelve years (−15% in 2024 alone); a central bank that held rates at 18% and taxed the banks to 25%; a regulator that just launched competing rails; a border that makes every deposit a compliance question (the Culper report's allegations were answered, the class action was dropped, and the discount remained); a founder-and-PE register of ~62% with the public holding tenge-law rights through a depositary; and disclosed material weaknesses in internal controls, now including a Turkish subsidiary. None of these is a flaw in the business. All of them are the price of where it lives, and the market has priced them at roughly half of what the same company would fetch elsewhere.

Why Turkey is the smaller question. Kaspi has now spent about $1.5 billion — the purchase, the top-ups, the capital calls, a $300 million bank — on a business the market values at about $0.9 billion in total, of which Kaspi's 86.7% is ~$0.8 billion (approx. — Hepsiburada's Nasdaq price, 10 Sep 2026). Hepsiburada lost TRY 1.9bn in Q2, twice the prior year; real GMV grew 2.8%; Trendyol holds ~45% of the market; and the lira fell 21% against the tenge in a year. That is, so far, a capital-allocation error of perhaps $700 million against a company that earns $2.3 billion a year — painful, survivable, and reversible: management runs Hepsiburada "around breakeven" by choice and could stop investing tomorrow. It cost shareholders one year of dividends, which is the real grievance. It did not cost the Kazakh flywheel anything.

Our read: the bet is asymmetric, but the asymmetry runs through Almaty, not Istanbul. If Kazakhstan stays roughly as it is — a volatile currency, a tax-hungry but champion-friendly state, rates drifting down — the Kazakh business alone justifies the price several times over and pays you nine percent to wait, and Turkey is a free option that is currently worth less than nothing. If Kazakhstan does not — a devaluation, a sanctions event, a regulator that decides 70% is too much — the multiple does not re-rate from 8 to 15; it goes from 8 to 5, as the January 2022 internet shutdown and the September 2024 short report both showed in miniature. That is the tails, and "not losing much" depends entirely on whether the reader can hold a Central Asian jurisdiction through it. Pabrai can. Most of our readers, honestly, should ask themselves whether they can.

The pattern is the mirror of what we found at Broadcom this week, where the customers were the competitors. Here the regulator is the competitor: no private rival has dented the daily-user gap in a decade, so the state built the rails itself. In Turkey the situation inverts — Kaspi is the challenger, with a tenth of the incumbent's lead and none of the data advantage that made the Kazakh lending book so clean. A model built in a vacuum is being exported into a crowded room. It may work; Pabrai thinks the "jockey" can do it. But the base rate for super-apps crossing borders — Grab, Gojek, Paytm, Mercado Libre's early years — is a long, expensive learning curve, and the market's $0.8 billion valuation of the stake is a reasonable prior.

Integrity: no veto, with two asterisks. The Culper allegations of 2024 were answered by the regulator and abandoned by the plaintiffs; the related-party history (classifieds bought from the CEO, a bill-payment aggregator owned by the chairman) is disclosed and old; the 20-F discloses material weaknesses in internal controls that a Nasdaq company of this size should not still be remediating. Capital allocation is where the dial loses its points. For a decade the company paid out 50–60% of earnings quarterly and bought little back. Then it spent ~$1.5 billion on Turkey — funded by skipping every dividend in 2025 — for a stake now worth ~$0.8 billion, bought no minority out, and has put more capital in twice since. Buffett's one-dollar test asks whether each retained dollar has created a dollar of value; on the market's arithmetic this one has created about fifty cents so far. The dividend is back (KZT 850, then 1,000 — a 73% payout of Q2 earnings) and Lomtadze buys with his own money, both of which count. Dial: 6. An operator of the first rank; an allocator on probation.

The feed and the tenge. This is the first company on our board that reports in a currency other than dollars or pounds, and the feed shows it. The automated DCF prints an intrinsic value of $935.84, +859% — because the model computed value in tenge and compared it to a dollar price; divide by the exchange rate and it says roughly $2, which is equally useless. The estimates feed has one or two contributors per year, in tenge (FY2026 EPS ~KZT 6,050; FY2027 ~7,000; FY2028 ~8,270), which at ~460 per dollar imply forward multiples of about 7.4×, 6.4× and 5.4× (approx., own conversion) — thin, but directionally what the street says. And a caution for dollar holders that no ratio captures: FY2025 net income was flat in tenge and, at year-end rates, down in dollars; the H1 2026 figure is flat in tenge and up in dollars. Nothing changed in Almaty; the exchange rate changed. That is the whole investment in miniature.

The dividend is affordable and it is the reason most holders own the stock — at KZT 4,000 a year it yields about nine percent in dollars (approx.). But the six questions land in an unusual place: the capacity to pay is beyond doubt and the willingness was suspended for twelve months two years ago, with no warning, to fund an acquisition. A dividend that has been zero within living memory is a policy, not a promise, and we price it as such — which, at nine percent, the market already does. Note the contrast with the American tradition: McDonald's is six weeks from its fiftieth consecutive raise; Kaspi's streak is two quarters old.

Value it two ways, because the two answers disagree and the disagreement is the point. As a business: Kazakh earnings of ~$2.3 billion from a network with 41% returns on equity, growing volumes at 15–20%, deserve 14–18× in any jurisdiction we can underwrite — $170–220 a share, before Turkey, which we carry at the market's $0.8 billion (~$4 a share) rather than the $1.5 billion spent. As a jurisdiction: Buffett's tiers say that when intrinsic value cannot be reliably estimated, the answer is not a bigger discount but "do not invest," and a currency that can move 15% in a year against a state that can redraw the rails by decree is, for most readers, exactly that case. The market has split the difference at about half of the business value, which is where emerging-market national champions with concentrated registers have traded for a generation.

So our number is conditional, and we will not pretend otherwise. For a reader who can hold Kazakhstan — who will not sell on the next devaluation or the next short report — the stock is cheap today and pays nine percent to wait, and below ~$80 (about 6× earnings, a yield above 10%, the March low) the Turkish option comes free with a cushion even against a bad tenge year. For a reader who cannot, no price fixes the problem, because the risk is not in the price; it is in the passport. We are, honestly, in the second group.