Find the profit engine — it isn’t the cart
Go company by company and the same pattern repeats. Amazon’s operating profit comes from AWS and a ~$68.6B advertising business, not from selling goods. MercadoLibre’s engine is fintech (~$278B of payment volume) and high-margin ads. Sea’s Shopee throws off less than 1% of GMV in EBITDA — the real money is Garena games and the Monee credit book. Coupang’s retail gross margin only reaches 32% because a $2.2B+ ad business lifts it. Shopify is ~75% payments revenue. Seven different platforms, one identical confession: e-commerce is a customer-acquisition machine whose purpose is to feed a higher-margin attached business.
Three flavors of the same trick
The attached engine takes three forms, and they sort the cluster. The advertising model — selling visibility back to the sellers on your platform — is the purest and highest-margin (Amazon, PDD, Alibaba’s customer-management revenue, Coupang). The fintech model monetizes the payment and credit flow the marketplace generates (MercadoLibre, Sea’s Monee, Shopify Payments). And the adjacent-business model cross-subsidizes commerce from an unrelated cash machine (Amazon’s AWS, Sea’s games, Alibaba’s cloud). The healthiest platforms run all three; the question for each is whether the engine is big and durable enough to carry a thin, competitive retail core.
The de-minimis wall splits the cheap-goods players
One regulatory event cuts cleanly across the cluster: the end of duty-free de-minimis parcels (US 2025, EU 2026). It is existential for the cross-border discounters — PDD’s Temu was built on duty-free price advantage, and US daily users fell ~52% after the repeal, snapping its growth from +59% to ~10%; Shein is the same bet in fashion, a ~$37B-revenue supply-chain machine whose entire price edge rode that loophole. The domestic-anchored platforms (MercadoLibre, Coupang, Sea, and Meituan, whose real marketplace is local services) are insulated, and the infrastructure player (Shopify) is agnostic. Read together, the studies show that the “cheapest cart wins” thesis had a hidden dependency on a tax loophole — and the platforms whose moat is logistics, fintech or a captive market are the ones the wall doesn’t touch.
Geography is still a moat — and a ceiling
The cross-section also shows e-commerce is stubbornly regional. MercadoLibre owns Latin America, Sea owns Southeast Asia (~53% share, #1 in all six markets), Coupang owns Korea, Amazon the US, the Chinese players their home turf. Local logistics, payments and trust don’t travel — which both protects the regional champion (Coupang’s owned delivery network keeps Chinese rivals under 2% share) and caps it (Coupang earns ~90% from one mature market and trades at ~82× on an unproven Taiwan bet). The lesson the cluster makes concrete: a regional commerce moat is real, but it is also the edge of your map, and the multiple usually prices in an expansion that may not port.
Where they agree — and where they split
All seven accept that pure retail is a thin, brutally competitive business and that the durable profit sits in the attached engine. They split on valuation and on the AI/agentic threat. The cheap-vs-dear spread is enormous — PDD at ~8× and Alibaba at ~19.5× (priced for a maturing, share-losing core and a subsidy war) against MercadoLibre at ~40×, Shopify at ~108× and Coupang at ~82× (priced for the growth engine to keep compounding). And every study now carries an agentic-commerce question: if an AI agent does the shopping, who owns the buyer — the marketplace, or the model? The demand to buy things online isn’t the question. Which attached engine keeps paying, and at what multiple, is.