I attended Shanghai’s Inclusion Conference on the Bund in September as a fintech and financial-inclusion practitioner, not as someone paid to rank countries on AI leadership—which is probably why it felt like the question everyone kept asking was the wrong one.

Is China ahead on AI?

That framing sends you looking for the smartest model. But there is another contest underway right now that tends to get less attention: who can embed AI fastest into the services and habits people already know and trust — a payments app, a WhatsApp chatbot, a voice call, an agent network that was already carrying cash before anyone called it AI.

By that measure, China is one of several labs running this experiment at scale right now. It is not the only one, and in some respects, it is not even the most instructive one.

At the Bund, the pattern was unmistakable. Alipay’s AI agent, Ah Bao, launched in June 2026, offers access to more than 10,000 everyday services, according to Ant Group, including utility payments, ride-hailing, retail orders. Ant Group’s health app, AQ, has reportedly crossed 150 million users, tracking sleep, nutrition, and weight for entire households. None of this required a frontier model. It required getting AI into something people already opened every day.

The same logic is running across Africa’s agricultural finance sector. Apollo Agriculture, a Nairobi-based fintech company that supports over 350,000 farmers, uses satellite imagery and machine learning to underwrite credit for smallholder farmers with limited formal credit histories. Ghana’s Farmerline reaches farmers through Darli AI, a WhatsApp chatbot available in more than 27 languages—embedding AI into a messaging app people already use, rather than asking them to download something new.

India’s version of this race centers on language rather than distribution. Nandan Nilekani, the architect of India’s digital public infrastructure, called voice AI the key to digital equality, betting that voice interfaces, not better screens, are what finally bring India’s Universal Payment Interface (UPI) to citizens who cannot navigate a menu in English or Hindi script. Backed by NVIDIA’s infrastructure and India’s own sovereign language models from Sarvam AI, the goal is explicitly to embed AI into UPI’s existing rails, not replace them.

Brazil offers perhaps the clearest proof that embeddedness, not model sophistication, is the determining variable. Nubank reports that its nuFormer underwriting model has delivered a 70 percent reduction in risk for comparable borrower populations compared with earlier models. The bank’s AI layer on Pix, Brazil’s instant-payment system, has already passed 10 million monthly active users.

During my conversations in Shanghai, it became clear that the real contest is over who gets AI embedded into existing, trusted daily infrastructure fastest, and the Global South is not far behind in this race. In agricultural credit and alternative underwriting, parts of it are arguably ahead, because the needs of populations with little formal credit history have pushed firms to develop new approaches.

There is one important caveat: dependence on compute, cloud infrastructure, or foundation models controlled by a small number of firms headquartered in the United States or China can limit who captures the value. Winning the embeddedness race on the ground does not guarantee a Global South institution captures the value it creates, if the layer beneath it is rented rather than owned.

The frontier-model race gets the headlines because it is legible to a Western tech press trained to cover Silicon Valley and now Beijing. The embeddedness race is harder to see, because it is happening inside WhatsApp chatbots, agent terminals, and voice calls in most of that press doesn’t speak.

It is, nonetheless, a race that will actually determine who benefits from AI first—and right now, it is a Global South story as much as a Chinese one.

Neha Mehta is the Founder of FemTech Partners and author of the book One Stop.