signal

Glocal Signal: three moves that test the substrate thesis

UPI crosses more borders, Shopee retreats from Europe, Nubank passes 100M — three moves, one thesis: interoperable substrate travels, walled substrate has to be rebuilt behind you.

Dipesh Majumder·Jul 10, 2026·2 min read·Updated Aug 4, 2026·Signal#signal#upi#shopee#nubank

TL;DR — Three moves, one thesis: what crosses a border is substrate, not apps. A payment rail travels; a walled super-app retreats; a bank that re-founds its stack per market compounds.

Signal 1 — UPI keeps crossing borders

India's UPI is now live in roughly nine countries (Singapore, the UAE, France and more), linked into local fast-payment systems, with cross-border transactions up about 20-fold in a year — and UPI-modeled systems being adopted as far away as Peru and Namibia.

The take: The cleanest counterexample to "substrate stays home." India didn't export an app — it exported an interoperable rail and plugged it into each country's own. If you want a local advantage to travel, ship the protocol, not the walled garden.

Signal 2 — Shopee's retreat proves the reverse

Sea's Shopee pulled out of India and France in March 2022, then Spain, exiting Europe within a year of entering. A ban on Sea's Free Fire game in India wiped roughly $16 billion off the parent in a single day; Sea's market value fell from about $200 billion in 2021 to roughly $27 billion in 2022.

The take: A super-app is the most substrate-dependent thing you can export — it needs distribution, trust, regulation and payments all at once. Blitzscaling into Europe on pandemic capital skipped every one of those layers, and the retreat was as fast as the entry. You can't buy substrate with a marketing budget.

Signal 3 — Nubank passes 100 million by re-founding, not exporting

Nubank crossed 100 million customers across Brazil, Mexico and Colombia in 2024 — roughly 10 million in Mexico and about 2.5 million in Colombia — and became Latin America's most valuable bank, the first digital-only bank to lead a region.

The take: The un-sexy winner. Nubank didn't assume Brazilian trust or credit data would travel; it rebuilt the stack country by country and let each market compound on its own substrate. Slow is the strategy, not the tax.

This week in my build

Shipped the newsroom this publication runs on — an AI-augmented, human-gated content engine where nothing publishes without a real operator take, cited sources, and an honest AI-vs-human provenance line. Same thesis, applied to media: the portable part (drafting) is cheap; the substrate (judgment, sourcing, voice) is the moat. More in the build log.

AI-drafted · human-edited by dipesh · fact-checked · 3 sources cited — held to the Firehoz editorial gate.

teardown

Jumia's Amazon problem: why 'the Amazon of Africa' couldn't find the customers

The most-watched consumer-internet story in Africa is a reverse lesson: not a great local product that failed to travel, but a foreign playbook imported before the local substrate existed.

playbook

The Localization Stack: what actually has to be rebuilt when you cross a border

A reusable framework: one portable product layer sits on four local ones — distribution, trust, regulation, money-movement. The product ships in an afternoon; the substrate has to be re-founded per market.

buildlog

I built a newsroom that can't publish slop

The rebuild after I killed my media site had one rule: make slop structurally impossible to publish. AI does the cheap 80%; a human-attested operator take gates the other 20% — and the agent can't fake the attestation.