The Leapfrog: How Emerging Markets Built Their Attention Economies Backwards

The order the West took, and the order emerging markets skipped

The rich world built its attention economy in a specific sequence, and the sequence mattered as much as any single technology in it. A landline telephone network, laid down over the better part of a century, created the wiring and the household billing relationship that dial-up internet later rode into homes. A personal computer, usually shared by a household and often bought on credit, gave that connection a keyboard, a larger screen and, over time, an operating system full of separate installed programs. Broadband replaced dial-up through the 2000s without changing the basic shape: one machine, one browser, a growing directory of separately built websites that a search box became the natural way to find your way around, because no other tool could keep up with a decade of independently launched destinations. Plastic payment cards followed their own, parallel sequence, issued against a bank account that itself presumed a branch network and a credit history. Smartphones, when they arrived after 2007, mostly layered onto habits already formed on a keyboard and a mouse, a second screen rather than a first one.

In India, Kenya, Nigeria, Indonesia, Brazil and much of the rest of sub-Saharan Africa and Southeast Asia, that sequence mostly did not repeat, and the gap is measurable rather than impressionistic. Kenya had roughly 200,000 landlines in use across the entire country as late as 1991; two decades later the fixed-line base had shrunk further rather than grown, and by June 2024 the Communications Authority of Kenya recorded 68.9 million mobile SIM subscriptions, a penetration rate of 133.7 percent, alongside a fixed-internet base that had only just crossed 1.5 million lines. India shows the same shape at a larger scale: mobile cellular subscriptions stood at 80.56 per 100 people in 2023, World Bank data show, while fixed telephone subscriptions have sat in the low single digits per 100 people for two decades. Neither market built a mass desktop-and-landline internet on the way to a mobile one. They largely went straight to mobile.

This piece traces that record across four layers, connectivity, payments, bundled commerce, and messaging, with sourced figures for each, then asks the harder question directly: where does the leapfrog thesis hold as a real description of what happened, and where is it an overstatement of what connectivity alone can do. It closes with a forward question, framed explicitly as a projection rather than a measured result, about what a mobile-first, chat-first starting point might mean now that discovery itself is shifting toward AI answers.

Mobile-first, mobile-only: the connectivity leapfrog

Widespread network coverage has not translated into universal use, and the gap between the two is where the connectivity story actually lives. GSMA's State of Mobile Internet Connectivity 2024 report, published 23 October 2024, found that 96 percent of the world's population lives within reach of a mobile broadband signal, yet 3.45 billion people, 43 percent of the global population, still do not use mobile internet. Of that 3.45 billion, only around 350 million live where no network exists at all; the remaining 3.1 billion, a 39 percent "usage gap" nine times the size of the pure coverage gap, live within range of a signal and simply do not connect, for reasons GSMA ties to device cost, digital skills and the relevance of available content. Ninety-three percent of the unconnected live in low- and middle-income countries, and in sub-Saharan Africa specifically, only 27 percent of the population uses mobile internet despite near-total coverage; an entry-level smartphone there can cost up to 99 percent of a month's income for the poorest earners, GSMA's report found.

For the majority who are connected, mobile is not the primary way onto the internet so much as the only way. GSMA's Mobile Economy 2025 report describes mobile as the primary, and in many cases only, route to the internet across low- and middle-income countries, where it accounts for 84 percent of all internet connections. India's own numbers illustrate the scale: the Internet in India Report 2024, jointly produced by the Internet and Mobile Association of India and Kantar, put active internet users at 886 million in 2024, up 8 percent year on year, with rural India, at 488 million users, accounting for 55 percent of that base and outgrowing urban India's 397 million for a fourth consecutive year, on a base that is overwhelmingly mobile. Indonesia shows a related pattern from a different angle: DataReportal's Digital 2025 report on Indonesia counted 212 million internet users, a 74.6 percent penetration rate, against 356 million active cellular mobile connections, a figure equivalent to 125 percent of the population that reflects widespread multi-SIM ownership rather than a literal majority of Indonesians owning two phones each.

Bank cards were optional: the payments leapfrog

The clearest, best-measured leapfrog in this study is in payments. Rich-world digital payments were built on top of a card network that itself sat on top of a banking relationship, a sequence that took decades and a dense branch footprint to reach mass adoption. Several emerging markets instead built a payments rail directly on the mobile phone, without most users ever holding a payment card at all.

The bundle instead of the browser: super-apps

The West's web stayed unbundled by default: a retailer's website, a separate payments processor, a separate bank app and a separate messaging app, loosely connected by a browser, an app store and, eventually, card networks that could move money between all of them. Several emerging markets instead consolidated chat, payments, commerce and, in some cases, government services into a small number of dominant apps. WeChat is the reference case other markets' apps were explicitly built against, even though China sits somewhat outside the emerging-market set this study otherwise covers: Tencent's own third-quarter 2024 results reported combined Weixin and WeChat monthly active users of 1.382 billion, with WeChat Mini Program commerce alone generating more than 2 trillion yuan of gross merchandise value in the quarter.

Southeast Asia built its own version of the same idea. Grab, spanning ride-hailing, delivery and digital financial services across eight countries, reported 43.9 million group monthly transacting users in the fourth quarter of 2024, up 17 percent from 37.7 million a year earlier, according to the company's own results. GoTo, formed from the 2021 merger of Indonesia's Gojek and Tokopedia, reported 71 million annual transacting users for full-year 2024, up 19 percent, with monthly fintech transacting users up 29 percent to 28.8 million. Both sit inside a regional digital economy that Google, Temasek and Bain's e-Conomy SEA 2024 report put at $263 billion in gross merchandise value for the year, up 15 percent, with sector profit growing two and a half times, from $4 billion in 2022 to $11 billion in 2024, as the region's platforms shifted from growth funded by subsidy toward actual monetization.

India's Paytm supplies a useful correction to any assumption that bundling only ever concentrates strength. Paytm's own investor disclosures reported average monthly transacting users reaching 90 million in the quarter ended March 2023. Company and press reporting on subsequent quarters shows that figure falling to roughly 72 million in the quarter ended March 2025, down from about 96 million a year earlier, after the Reserve Bank of India restricted Paytm's payments-bank affiliate from most new business in early 2024. A single regulatory action against one licensed piece of a bundled app moved tens of millions of users in a matter of months, a concentration of risk that has no close analogue in a more unbundled web of many separate, smaller services.

The inbox becomes the storefront: messaging as commerce

WhatsApp passed 3 billion monthly users in 2025, Meta chief executive Mark Zuckerberg told investors on the company's first-quarter 2025 earnings call; India and Brazil, each with well over 100 million WhatsApp users, are among its largest single markets. The scale alone understates what the app has become in those markets: increasingly, it is the storefront itself, not merely a channel that points to one.

Meta has begun testing AI-driven "Business Agent" tools on WhatsApp with small merchants, deliberately starting in India, Mexico and Brazil, with more than a million businesses signed up for the early rollout and, separately, more than a million businesses reported using some form of Meta AI agent on WhatsApp and Messenger every week. In Brazil specifically, Meta has enabled paying a local small business directly inside a WhatsApp chat, collapsing catalog, conversation and checkout into a single thread that never requires either party to open a browser.

For a large share of small merchants across these markets, a WhatsApp Business number, not a website, functions as the effective storefront: it is where a customer finds the business, often through a forwarded contact or a social post rather than a search result, asks a question, sees a catalog, and increasingly pays, all inside one conversation. That is a genuine reordering of the West's site-first sequence rather than a faster version of the same sequence, though the specific AI-agent commerce numbers above are recent, company-reported figures from an active product rollout, not yet an independently measured, multi-cycle trend.

Where the leapfrog thesis holds, and where it fails

Read against the evidence assembled above, the leapfrog thesis holds cleanly at the level of infrastructure and payments. It is considerably weaker as a general claim that these markets have closed the digital divide by skipping stages of it.

The strongest academic critique of the thesis predates most of the platforms named in this piece and still holds up. Philip Napoli and Jonathan Obar's working paper, "Mobile Leapfrogging and Digital Divide Policy: Assessing the Limitations of Mobile Internet Access," produced through Fordham University and the New America Foundation and drawing on internet-access and device-penetration data across 34 countries, found that mobile-only internet access is a poor functional substitute for PC-based access. People who connect only through a mobile phone show the lowest frequency of internet use for personal and economic development activities of any access group the study examined, a pattern the authors connect to the narrower, more consumption-oriented experience a small screen and a metered data plan tend to encourage, relative to a keyboard, a larger display and, often, an unmetered connection.

That finding maps onto an older distinction in the digital-divide literature. Eszter Hargittai's 2002 framing of a "second-level digital divide," extended by researchers including Jan van Dijk, separates a first-level divide, who has any access at all, from a second level: differences in the skills, frequency and range of use once a device is actually in hand. Mobile has done the most of any technology to close the first-level divide. The second level does not close on the same schedule, and GSMA's own numbers, cited above, operationalize the gap directly: even with 96 percent global mobile-broadband coverage, a 3.1 billion-person usage gap remains, concentrated in low- and middle-income countries, sub-Saharan Africa's own mobile internet use rate is stuck at 27 percent of the population, and an entry-level device can absorb nearly a full month's income for the region's poorest earners. The World Bank's Global Findex Database 2025 frames the same picture from the account side: 84 percent of adults across low- and middle-income countries now own a mobile phone and some 3 billion have a smartphone specifically, yet 79 percent of adults worldwide, not all of them, have a financial account of any kind. None of this is evidence against a connectivity leapfrog happening; it is evidence that connectivity was never the only barrier in the way.

A language and literacy layer sits underneath the same numbers. A KPMG and Google study published in April 2017, "Indian Languages, Defining India's Internet," found 234 million Indian-language internet users in the country in 2016 against 175 million English-language users, and projected that by 2021 there would be more than 500 million users of Indian languages against fewer than half that many English users, with nine out of ten new internet users over that period expected to use a local language rather than English. Read generously, this is part of why India's internet never needed to be desktop- and English-first to begin with: a market this large and this linguistically diverse was never going to be well served by the West's browser-and-keyboard, largely English web, so it built something else instead. Read less generously, it is itself a persistent gap that a mobile connection alone does not close: content, interfaces and, increasingly, AI systems trained overwhelmingly on English and a handful of other high-resource languages remain thinner in Hindi, Swahili, Yoruba, Bahasa Indonesia and the hundreds of other languages hundreds of millions of these same users speak, a gap that carries forward, largely unresolved, into the forward question this study turns to next. Indonesia's 125-percent mobile-connection-to-population ratio, cited earlier, is a small but telling reminder in the same direction: a count of active connections is not a count of connected people, once multi-SIM ownership is accounted for.

The forward question: an AI-answer era entered from a different start

Everything in this section is a projection grounded in the sequencing evidence above, not a measured outcome, because the AI-answer platforms in question have not operated in these markets long enough to produce one.

The West's sequence is worth restating one more time because the contrast is the point. A desktop-and-browser web, built over decades, trained a search box as the default discovery habit long before generative AI existed. AI Overviews, ChatGPT, Perplexity and similar answer engines have mostly been grafted onto that already-formed search habit, which is one reason so much of the current conversation about AI-search adoption quietly assumes a search box as the starting point.

The leapfrog markets traced in this piece never fully formed that habit in the first place. For a majority-mobile, often vernacular-first user base whose commerce and discovery already run through a chat thread, whether a product photo forwarded on WhatsApp or a transaction completed inside a WeChat Mini Program, the search box was never the primary gateway it became in the West. Projecting from the sequencing evidence assembled here, if discovery keeps moving toward conversational, agentic surfaces, these markets may be adapting a habit they already have, finding and transacting with a business inside a chat thread, into a nearby one, finding and transacting with a business through an AI agent inside that same thread, a shorter conceptual distance than the jump a search-first market has to make from a search results page to a conversational answer.

The one piece of real, if early, adoption evidence bearing on this question comes from a 2026 study, "How Early Adopters Used Generative AI Worldwide: Variation by Country Income and Language," posted to arXiv. Examining a large set of anonymized chatbot interactions, the study found that countries whose predominant languages are well represented in AI training data have adopted generative AI at roughly twice the rate of countries whose languages are not, and that in lower-income countries, use skews more toward schooling than the leisure use that dominates in richer countries. The authors frame the open question directly, and this study borrows their framing rather than resolving it: whether improving model performance across languages lets AI extend the kind of leapfrog mobile connectivity delivered, or instead cuts that leapfrog off behind a language barrier the mobile leapfrog never had to fully clear.

Meta's own WhatsApp Business Agent tests, launched first in India, Mexico and Brazil, the same chat-commerce markets already covered in this piece, are a live signal of where at least one major platform is betting that discovery habit lands. A single company's early product test is evidence of a bet, not of an outcome, and should be read as exactly that.

The symmetry is worth naming plainly. The same literacy, language and device-affordability gaps that limited the connectivity leapfrog, documented in the section above, apply with at least as much force to an AI-agent leapfrog. A population a chatbot cannot address adequately in its own language is no better served by a conversational AI surface than an unconnected household was served by a mobile network it could not afford to use. Nothing in the record available today settles which force, the head start of an already chat-native population or the drag of an uneven language and skills base, wins.

The verdict

The leapfrog thesis holds, on real and heavily documented evidence, at the level of infrastructure: fixed-line telephony was skipped almost everywhere it is measured in this study, from Kenya's 200,000 landlines in 1991 to India's low single-digit fixed-line penetration today. It holds just as clearly at the level of payments: M-Pesa and UPI are not smaller or slower versions of a Western card-and-bank system, they are larger, and in UPI's case, by the IMF's own count, the single largest real-time payments system on earth.

It holds more unevenly at the level of commerce and discovery. Super-apps and chat commerce are real, large, and in WeChat's and WhatsApp's case, enormous by any measure. But Paytm's own user decline after a single regulatory action against one affiliate is a reminder that a bundle concentrates fragility along with convenience, a fast, visible failure mode the West's more unbundled, competitive web has generally been slower to produce at the same speed and scale.

And it is overstated, on evidence GSMA and the academic literature themselves supply, as a claim that emerging markets have closed the digital divide simply by skipping stages of it. A usage gap nine times the size of the pure coverage gap, a device that can cost a month's income for the poorest users, and a vernacular-language and skills layer that KPMG and Google flagged as early as 2017 all persist underneath the headline connectivity and payments numbers this study opens with.

Whether that same reordering repeats at the level of AI-mediated discovery, whether these markets' head start in chat-native commerce becomes a head start in an AI-answer economy, or whether the same language and literacy gaps that limited the connectivity leapfrog limit this one too, is not yet answerable from the record. It is the question this study leaves open, and the one worth tracking as these platforms mature.