Essay
Two-Tiered Marketing and the 2.1 Billion People the AI Sector Cannot See
The most important market in the history of artificial intelligence does not appear on a single investor slide. It generates no ARPU worth reporting, occupies no enterprise seat count, and files no earnings. It is roughly 2.1 billion people — the global lower-middle class (LMC) — and today it is legible to the AI sector only as a source of raw data and cheap labor, never as a customer.
This essay lays out the case for two-tiered marketing: the argument that AI will not reach the LMC through the same commercial machine that serves the affluent world, that no company will build the second tier on its own, and that catalyzing it in the next eighteen to twenty-four months is the decisive task of AI governance in the Global South. The figures below include rough, illustrative estimates. I would rather be approximately right about a 2.1-billion-person market than precisely silent about it.
Two-tiered marketing is a deliberate go-to-market architecture with two structurally different tiers rather than one price list applied to everyone.
Tier One is the market the industry already serves: enterprises and affluent consumers, sold AI at global prices for immediate return on investment. Essentially all of the sector's marketing energy, product design, and pricing lives here.
Tier Two is the LMC mass market — people above extreme poverty but below the consuming middle class — who cannot be reached by Tier-One economics at all. Serving them requires a different apparatus entirely: localized pricing (rupiah and peso micro-tiers, not dollar subscriptions), embedded microcredit, subsidized activation, blended finance, vernacular-first design, and educational scaffolding.
Why the second tier is required to activate the LMC. Tier-One economics don't merely underserve the LMC; they structurally exclude it. A $20/month subscription is a rounding error to a Global North professional and a week's income to a warung owner in Central Java. Without a purpose-built second tier, the LMC never becomes a market. It becomes only a data commons to be enclosed and a labor pool to be tapped.
How is the LMC receiving AI today — ChatGPT or a Chinese app? Overwhelmingly not through anyone's paid subscription. It arrives free — and increasingly, it arrives Chinese. Microsoft's AI-diffusion tracking for the second half of 2025 put global generative-AI use at roughly 16% of the world's population, with adoption in the Global North growing nearly twice as fast as in the Global South, widening the divide rather than closing it. Within that gap, DeepSeek — open-source, free, no credit card — has surged fastest precisely in underserved markets, with reported usage in parts of Africa several times higher than elsewhere, and it ships as the default assistant on Chinese-made handsets, promoted alongside Huawei's infrastructure. So the LMC's first taste of AI is often a free Chinese model, pre-installed on an inexpensive device, or an AI feature embedded invisibly inside an app they already use (CapCut alone reaches hundreds of millions this way). It is not a US product they chose and paid for.
Has two-tiered marketing ever happened before? Yes — repeatedly, everywhere except AI. It is the standard mechanism by which a frontier technology reaches the poor:
Two-tiered marketing is not exotic. It is how humanity has always delivered the new thing to the many. It has simply never been built for AI — and the window to build it is closing.
Still concentrated, and by some measures concentrating further. The North's adoption is growing roughly twice as fast as the South's, and the population-normalized gap has been widening. ChatGPT's ~900 million weekly users skew heavily toward affluent, English-fluent, urban populations.
India is instructive — simultaneously the largest single opportunity and the clearest warning. India has built the world's most advanced public digital rails (Aadhaar identity, UPI payments, the wider India Stack) sitting under a vast, price-sensitive, multilingual user base. That combination is the one place a genuine second tier could form almost organically. But India is also where the extractive default is most visible: foreign models trained on Indian data and monetized elsewhere, with little of the value returning to the users who generated it. So the honest answer to "when does AI reach the Global South at scale?" is: unevenly, and on someone else's terms — free Chinese models arriving first, US products arriving late and priced out. "At scale and on sovereign terms" is not the default trajectory. It is precisely the outcome the second tier has to be built to produce.
The second tier is not a slogan; it has a working shape, and Indonesia is where its two halves are already visible. It takes both a horizontal layer and a vertical one, and neither works without the other.
The horizontal layer — financial and civic rails. This is the Sri Mulyani contribution: the fiscal architecture, treasury discipline, and public payment and identity rails (QRIS, digital ID, the BRI microcredit system) that let value move cheaply and reach people the formal banking system never underwrote. Rails are what turn a dispersed, cash-poor population into an addressable one.
The vertical layer — platform integration that formalizes. This is the Nadiem contribution: the Gojek pattern of drawing informal workers into formal economic structures by giving them tools worth adopting, plus an education reform (Merdeka Belajar) aimed at the critical-thinking and just-in-time skills the AI era demands. Verticals are what convert access into livelihood.
Put the horizontal rails and the vertical platforms together and you have the template for a second tier: national infrastructure that lowers the cost of reaching the LMC, and platforms that turn that reach into income. Indonesia is Tier One not because it is finished, but because both halves of the mechanism already exist there and can be shown to work.
A tempting shortcut is to serve the whole Global South from a few regional AI "hubs" — one for Southeast Asia, one for Latin America, one for Africa. It is cheaper to build and easier to sell. It is also self-defeating, because the thing the LMC most needs is the thing a hub cannot provide: local language, local data, local institutions, and local ownership of both.
A pan-regional hub optimizes for the average and erases the particular. It trains on the dominant regional language and underserves the rest; it centralizes data and governance offshore; it makes every nation a customer of infrastructure it does not control. National anchor institutions do the opposite — they hold the data, adapt the models to the vernacular, and keep sovereignty where accountability lives. Anchors localize; hubs homogenize. Sovereignty is national or it is nothing.
It is worth being precise about why 2.1 billion people go unseen, because the invisibility is structural, not accidental. A single-tier industry measures itself in the units Tier One produces: enterprise seats, paid subscriptions, average revenue per user. By those instruments the LMC reads as a near-zero — high volume, negligible per-head revenue — and so it never appears as a market on any slide that matters.
But the absence is an artifact of the measuring apparatus, not of the population. The demand is real; the willingness to pay (in small, local increments) is real; the productive upside is enormous. What is missing is a second set of instruments — local pricing, microcredit, activation economics — capable of registering it. Invisibility is not an accident of measurement. It is the natural output of a one-tier industry.
The uncomfortable fact for anyone who thinks the second tier can wait is that someone is already building it. Huawei is laying AI-ready infrastructure across dozens of Global South markets — cloud, data centers, managed connectivity, even AI-managed solar power for telco sites where the grid can't be assumed — and DeepSeek is arriving free on the handsets that ride on top of it. Cheap devices, free models, financed infrastructure: to the LMC user, that is a functioning, purpose-built go-to-market.
Read carefully, this is China already building a de facto second tier for the Global South. But it is built on terms of data dependency, not data sovereignty. The infrastructure is generous precisely because it captures the data, the standards, and the long-term dependency. The question is not whether the second tier gets built. It is whether the Global South builds a sovereign one in time, or inherits a dependent one by default.
Left alone, none of this corrects itself. A one-tier industry keeps measuring the LMC as a near-zero; a dependency-based second tier keeps it as a data source. Recognition — becoming a stakeholder rather than a resource — is not something the current system drifts toward. It has to be built, and built in a specific order.
Recognition requires ownership; ownership requires data sovereignty; data sovereignty requires the second tier to be built before lock-in hardens. Miss the window and the LMC is frozen in its current status — a source, never a stakeholder. Data colonialism is not a metaphor here; it is the default outcome of doing nothing in time.
Seven gates stand between the LMC and activation. The second tier has to fund them together, because any single unopened gate stalls the whole:
Each is a gate. Two-tiered marketing means funding all seven as a package rather than pretending a translated app is enough.
Rough estimates, built up from a per-capita activation cost. The anchor figure from the coalition's Indonesia modeling is roughly $40 per person to activate an LMC member — meaning onboarding, localized applications, microcredit enablement, and digital literacy — not the credit itself.
| Nation | Est. LMC population | Activation capital |
|---|---|---|
| Indonesia | ~140M | ~$5.6B |
| Brazil | ~70–90M | ~$2.8–3.6B |
| Mexico | ~45–55M | ~$1.8–2.2B |
| Thailand | ~25–30M | ~$1.0–1.2B |
| Tier One total | ~280–315M | ~$11–12.5B |
That ~$11–12.5 billion is activation capital, and its leverage is the point: it catalyzes a far larger pool of revolving microcredit. In Indonesia alone the credit unlocked is on the order of $10–20 billion — plausible when you note that BRI already disburses roughly $12 billion a year in subsidized KUR microcredit to more than 4 million enterprises, and its Ultra Micro (UMi) Holding serves some 34.5 million active borrowers. So the activation dollar leverages roughly 2–4x its value in credit.
Straight commercial investors won't, at first — and that's the entire diagnosis. The bottom tier is a genuine market failure, not an oversight. Blended-finance investors, however, will, for five reasons:
Slower ROI is a rational trade when the alternative is ceding the largest market in the world to someone else.
Because the bottom tier is the classic missing middle: too risky and too slow for commercial capital, too large and too commercial for pure aid. Philanthropy plays the role no other capital can — the catalytic first-loss layer.
Grants fund the public goods the market won't: vernacular datasets, digital-literacy programs, open governance frameworks, critical-thinking curricula. Program-related investments (PRIs) provide concessional, first-loss capital that de-risks the commercial tranche above it. The result is the blended-finance stack:
Philanthropy (first-loss) → Development finance (concessional) → Sovereign/patient equity → Commercial (senior, return-seeking).
Remove the philanthropic catalyst and the stack never assembles; the tranches above it have nothing to sit on. This is what I've called the missing mechanism: not missing money, but the missing catalytic layer that lets the rest of the money flow.
Activation gets the LMC in the door; graduation is what makes it a durable market. A first microloan is not the goal — moving a borrower up a ladder is. AI-assisted underwriting matters here precisely because better cash-flow data lets a lender advance a borrower from ultra-micro to micro to small far faster and at lower cost than a loan officer ever could.
The Indonesian precedent is concrete. BRI's Ultra Micro Holding already reaches ~34.5 million active borrowers, and BRI's own graduation program, which walks a borrower from ultra-micro up to small enterprise, is the working template for moving the LMC up a tier rather than trapping it at the bottom.
There is a labor-market reason the leading labs should care, quite apart from the moral one.
The most mission-driven AI talent — and especially the most ethically motivated researchers — increasingly want their work to matter for humanity, not only for enterprise margins. A lab with no credible answer for the 2.1-billion-person LMC will steadily lose its best and most ethical people to competitors, or to mission-first startups, that do offer that purpose. Purpose is the retention mechanism for scarce ethical talent.
Building the second tier is therefore a talent-attraction and talent-retention strategy. It gives the ethical core of the workforce a reason to stay, and it hands recruiters the single largest available mission on the planet. Labs that have signaled mission-first commitments have already shown this dynamic draws high-intent talent; the LMC is the largest such mission on offer.
Rough, illustrative projections — the orders of magnitude matter more than the decimals.
If Tier-One activation succeeds and begins scaling toward the broader LMC by 2032, assume even a modest 5–10% of LMC members gain AI-assisted livelihoods:
Each new micro-enterprise needs working capital, so microcredit demand surges. Using BRI as the template: it already serves ~34.5 million ultra-micro borrowers and disburses ~$12 billion a year. AI underwriting could plausibly double reach while halving unit cost — and, crucially, accelerate the graduation ladder, because better cash-flow data lets a lender move a borrower from ultra-micro to micro to small far faster. BRI's graduation model is exactly the mechanism for converting microcredit into middle-income enterprise. A reasonable 2032 scenario for Tier One: the AI-assisted microcredit pool grows 2–3x, and perhaps 20–30% of ultra-micro borrowers graduate up a tier. That graduation — not the initial loan — is where the LMC actually becomes a durable market.
To lend to the LMC at scale you need underwriting, and the LMC has no traditional credit history — no collateral, no formal salary, no bureau file. China solved this with pervasive alternative-data credit scoring, but at the cost of a surveillance apparatus fused to the state.
The Global South needs the underwriting capability without importing the surveillance model. The AI Middle Way answer is alternative-data, AI-assisted underwriting built on data the borrower owns and consents to share — a sovereignty-preserving, consent-based credit assessment, transparently governed, rather than a social-scoring regime run from the center. So yes: a form of AI-assisted credit scoring is unavoidable if the LMC is to be served. The whole question is whether it is built on consent and ownership or on surveillance and control. This is exactly why data sovereignty is the linchpin of the entire architecture, not a side concern.
There are two paths to scaling the LMC.
The autocratic path achieves scale through centralized control: surveillance-based credit, state-directed platforms, mandated adoption. It is fast — and extractive of human agency.
The democratic path scales through distributed ownership, consent-based data, plural institutions, and critical-thinking education. It is slower to start and self-reinforcing thereafter.
Democracy is the optimal outcome for three reasons. First, it generates the next wave of jobs. Just-in-time livelihoods emerge from critical thinking and creativity — exactly the human capacities autocracy suppresses. A scaled-but-controlled population produces compliance; an empowered citizenry produces invention. Second, consent builds trust, and trust is the real currency of adoption — a population that owns its data adopts faster and deeper than one that fears it. Third, and most fundamentally, only the democratic path is consistent with the coalition's core principle: consciousness must direct intelligence, not the reverse. Autocratic scaling inverts that relation — intelligence is used to direct and control consciousness. The LMC scaled under autocracy becomes a managed population. Scaled under democracy, it becomes both a free citizenry and a genuine market. Those are not equivalent outcomes.
Development banks and sovereign wealth funds are not interchangeable pools of "cheap money." They occupy different tranches and do different work.
Development banks (the World Bank/IFC, ADB, IDB, and national development banks) supply concessional, patient capital, absorb country risk, and fund the infrastructural gates from Section 8 — connectivity, power, data centers, digital ID — that no private investor will finance alone. They open the gates.
Sovereign wealth funds (Indonesia's Danantara, which now controls BRI; Temasek; the Gulf funds) supply anchor equity and strategic patience. Their mandate is national and generational rather than quarterly, so they can hold long-dated positions in a second tier that won't return capital on a venture timeline — and they align the second tier with the nation's own sovereignty goals.
Placed in the blended stack, each plays a distinct part: philanthropy is catalytic first-loss; development banks provide concessional infrastructure capital; sovereign funds provide patient anchor equity; commercial capital sits senior and return-seeking on top. The architecture only works when each tranche does its own job.
Rough timeline, and a crucial distinction between architecture and capital.
If Tier One is capitalized between 2026 and 2028 (~$9–12B) and proves the model, Tier Two and Tier Three nations follow from roughly 2028 to 2032. Full global LMC activation — all 2.1 billion at ~$40/head — implies on the order of ~$84 billion of activation capital, catalyzing perhaps $200–400 billion in revolving credit, realistically phased across 2026–2035.
But the binding constraint is not the money; it is the lock-in window. The architecture — data sovereignty, governance frameworks, consent-based credit, national anchor institutions — must be set by roughly 2027–2028, even though the capital fills in over a decade. Get the architecture right in that window and the later capital funds a sovereign system. Miss it, and the same later capital simply funds a more efficient data-colonial one. So: architecture by 2028, Tier One funded by 2028, global fulfillment around 2033–2035 — with everything hinging on the 2027–2028 architecture window.
Today there is almost no incentive for the LMC to adopt AI, because AI as currently marketed offers them a subscription they can't afford to do a job they don't have. Incentives have to be built on four legs at once — remove any one and the structure collapses.
The coalition's role is to build the mechanism that makes all four incentives real — the blended finance, the governance, the education, the anchor institutions — and to build it before the window closes. The digital-divide movement's unfinished work was always this: not merely connecting people, but ensuring that when the technology arrived, they owned their place in it. AI is the last and largest chance to finish it.
Rough estimates throughout are illustrative and intended to convey orders of magnitude, not precision. Empirical figures on global AI diffusion draw on the Microsoft AI Economy Institute's 2025 diffusion tracking and reporting from Rest of World and the China–Global South Project; microcredit figures draw on Bank Rakyat Indonesia's published disbursement and Ultra Micro Holding data.
I would rather be approximately right about a 2.1-billion-person market than precisely silent about it. Introduction
Two-tiered marketing is not exotic. It is how humanity has always delivered the new thing to the many. §1 · What is two-tiered marketing
The LMC’s first taste of AI is often a free Chinese model, pre-installed on an inexpensive device. It is not a US product they chose and paid for. §1 · How the LMC receives AI today
Anchors localize; hubs homogenize. Sovereignty is national or it is nothing. §4 · Why regional hubs prevent localization
Invisibility is not an accident of measurement. It is the natural output of a one-tier industry. §5 · Why the LMC is invisible
Read carefully, this is China already building a de facto second tier for the Global South. But it is built on terms of data dependency, not data sovereignty. §6 · The Huawei case
Data colonialism is not a metaphor here; it is the default outcome of doing nothing in time. §7 · Why the LMC stays unrecognized
This is what I’ve called the missing mechanism: not missing money, but the missing catalytic layer that lets the rest of the money flow. §11 · Why philanthropy is needed
A scaled-but-controlled population produces compliance; an empowered citizenry produces invention. §15b · Why democracy is the optimal outcome
AI is the last and largest chance to finish it. §18 · Building the incentives
Empirical sources
Intellectual antecedents (§1)