Half of Asia's enterprises don't use English for AI. That's the bottleneck

Only 1% of Asian enterprises say AI is not a priority. Everything else in Alibaba Cloud's new survey follows from that number, and most of it reads like a spending plan. NielsenIQ polled 1,000 IT decision makers across eight Asian markets in mid-2025 for the report, published in July 2026. Some 95% intend to increase investment in AI products, and more than half plan increases above 20%. Nine in ten say their organizations are optimistic about adopting AI, and 75% call the technology indispensable to their operations.

The money goes to the base layer first

Budget growth is not spread evenly. The largest planned increases hit foundational infrastructure: 69% of companies expect more than 20% growth in IaaS spending, ahead of PaaS at 61% and MaaS at 58%. Compute, storage and networking come first, with tools and models stacked on top. The same pattern showed up in Chinese markets when Moonshot AI's K3 model helped push C shares up more than 5% on expectations of continued infrastructure spending, with memory chip makers SK Hynix and Samsung among the clearest winners. The K3 bump fits a wider play: Chinese labs are shipping open-weight models at no cost, a strategy that has split Silicon Valley, per our analysis of China's open-weight gambit. Alibaba is applying the same logic at a much larger scale, committing RMB380 billion, roughly $53 billion, over three years to cloud and AI infrastructure, more than it spent on AI and cloud in the previous decade.

Cost cutting is not the headline motive. The top strategic objective for AI adoption is driving innovation and new business opportunities, cited by 64% of respondents, just ahead of cost savings and efficiency at 63%. Revenue growth follows at 48%, competitive advantage at 45% and customer experience at 43%. In practice, the use cases skew toward data analysis and decision-making (66%), customer service chatbots (64%), marketing and content creation (58%) and product development including coding (55%). Alibaba is chasing that first use case with QwenPaw-Data, an agent built to bridge the gap between business analysts and their data.

Privacy, cost and skills cap the enthusiasm

The ambitions run into a familiar wall. Data privacy and security is the top barrier to broader adoption, named by 48% of respondents, followed by implementation costs (42%) and a lack of internal expertise (37%). Regulatory and ethical concerns sit at 31% overall but climb sharply in financial services, education and the public sector. The rest of the list shows how young some of these deployments still are: integration challenges (28%), limited accessibility of suitable solutions (23%), unclear return on investment (22%), a shortage of obvious business cases (19%) and resistance from management (18%).

Availability is not the real constraint. On average 91% of respondents say AI solutions are available in their markets, and around 90% say they have started adopting AI development tools and models. Asked what would unlock wider use, 54% point to more customized, sector-specific solutions, 49% to better talent and skills, and 45% to more affordable offerings. The products exist; the fit does not. The same mismatch runs through agent research: models surge in coding but stall where enterprises actually need them, according to the Messier corpus and its 957,253 benchmark records.

Local-language models are the quiet bottleneck

The most consequential finding is also the easiest to skip. Only about half of respondents select English as their primary language for AI solutions; the rest work in Bahasa, Japanese, Korean, Thai and other local languages. In Japan, South Korea, Indonesia and Thailand, enterprises report that the scarcity of high-quality local-language models is holding development back, even as English- and Chinese-language models proliferate. Japan's answer so far has been orchestration rather than building a frontier model of its own, the bet that defines the country's AI sovereignty play.

That is where Alibaba Cloud's own models enter the story. The survey notes that products natively supporting multiple Asian languages are winning market share, and cites the Qwen family as especially popular in Japan and South Korea, with strong local-language performance and support for more than 200 languages and dialects in its latest generation.

A survey that reads like a product brief

It is worth remembering who paid for the numbers. Alibaba Cloud commissioned NielsenIQ, and the findings line up neatly with its strategy. Asked about provider preferences, 45% of respondents favor fully integrated AI and cloud offerings, 34% prefer hybrid flexibility across multiple clouds and only 16% want standalone models without bundled infrastructure. Alibaba Cloud sells exactly the first option, end to end.

The company's framing is explicit. "As we enter the agentic AI era, our focus is shifting from producing efficient tokens to enabling actionable outcomes," said Dr. Feifei Li, CTO and president of international business for Alibaba Cloud Intelligence Group. Independent assessment is more cautious: Gartner placed Alibaba Cloud in the emerging leaders quadrant, not the leaders quadrant, across all four reports in its November 2025 generative AI series. The gap between the survey's enthusiasm and that analyst placement is where the risk sits.

The survey measures appetite, not market share. It tells you that Asian enterprises are past the pilot phase, that infrastructure spending will keep climbing, and that language support and trust will decide which vendors collect the checks. The $53 billion question of who actually captures this demand is still open, with four very different answers on AI infrastructure control now in play.