The business of AI in Asia

Ask most operators in the region what the business of AI looks like and they describe a software company: a model, a subscription, a price per seat. Very little of the money in Asia moves that way.

The companies earning real revenue from AI here mostly do not sell it on its own. They put it inside something the customer already pays for, such as cloud compute, a delivery platform or an ad budget, and charge for that. AI is the reason to buy. It is rarely the line on the invoice.

Last week I wrote about where the profit sits in the stack. This piece is about the shape of the business model, because it decides what you should be negotiating for.

The largest AI business in Asia is a cloud business

Alibaba reported AI-related product revenue of RMB12,376 million (US$1,824 million) for the June quarter, its twelfth consecutive quarter of triple-digit growth. Revenue from its combined AI Cloud and Compute Services unit was RMB48,437 million (US$7,139 million), up 45 per cent. Group revenue grew 9 per cent. Capital expenditure for the quarter was US$9,975 million, up 75 per cent (Alibaba, 20 August 2026).

Read those numbers together. The AI line is growing fastest, but it is roughly a quarter of the cloud unit's revenue, and the cloud unit is where it gets paid. Alibaba gives away open Qwen models and sells the compute to run them. The model draws customers in; the infrastructure earns the money.

So what: When a hyperscaler offers you cheap or free model access, you are not being sold a model. You are being sold a multi-year compute commitment. Negotiate the cloud contract, exit terms included, as hard as you would any other infrastructure deal.

Grab gives its AI away

Grab is the clearest Southeast Asian example of AI as a feature. On its first-quarter call, Grab said its merchant assistant, Mai, had been adopted by about half of its active single-store merchants, with a 15 per cent uplift in GMV for engaged users. Drivers using its AI-powered Turbo mode earned 23 per cent more per online hour than those who did not (Grab Q1 2026 call, 4 May 2026). When an analyst asked whether these tools would become a separate software revenue line, chief executive Tan Ping Yeow did not give a price, saying only that there was "no reason why our partners should not have access to these tools".

On the second-quarter call, he said the cost per AI interaction with drivers and merchants had roughly halved in a year while monthly interactions grew tenfold. Revenue was US$997 million, up 22 per cent, and full-year guidance rose to between US$4.10 billion and US$4.15 billion (Grab Q2 2026 call, 3 August 2026).

That is the business model. Grab does not need merchants to pay for an assistant. It needs merchants to sell more through Grab, and to keep Grab as their main platform rather than one of three.

So what: If you sell through a platform in Asia, its AI tools are a loyalty mechanism, and they will deepen your dependence on it. Use them, because the uplift is real, but keep your customer data and your own channels alive.

The independent AI company is mostly a Singapore story

For businesses that do sell AI as the product, the capital is thin and concentrated. Tracxn counts US$4.1 billion raised by native AI start-ups in Southeast Asia in 2026 to the end of July, across 23 rounds, against US$2 billion across 41 rounds in all of 2025. One US$2.8 billion round accounts for about 68 per cent of this year's total. Singapore-based companies have raised US$9.3 billion since 2024. Vietnam, Malaysia, Indonesia and Thailand together have raised under US$40 million (TNGlobal, on Tracxn data, 6 August 2026).

Fewer rounds, larger cheques and one dominant hub. Part of the concentration is legal domicile, since regional companies often raise through a Singapore entity. Even allowing for that, the standalone AI vendor market outside Singapore is small.

So what: Before you sign a three-year deal with a regional AI start-up, check its runway as carefully as its demo. Many will be acquired, folded into a platform or shut down, and your workflows go wherever they go.

What this means for the people buying

Put the three together and the business of AI in Asia looks less like a software market and more like a bundling market. The big players fold AI into compute, commerce and advertising. The independents are few and well funded in one city. And the buyer, the ordinary company, increasingly gets AI whether it asked for it or not, inside tools it already pays for.

That is good news on price. It is a trap on capability. When AI arrives as a free feature, nobody budgets for teaching people to use it, and the feature sits idle. In the companies I work with, the gap is almost never access. It is whether the people who know the work have been shown how to redesign it.

The through line

The money in Asian AI follows the bundle, not the model. Cloud providers sell compute, platforms sell loyalty, and the few pure AI vendors sell a bet on being bought. An operator who understands that stops shopping for the best model and starts asking a better question: which bundle am I already inside, and have my people been trained to steer it?