Imagine AI not as a single invention but as a giant layer cake. At the bottom is energy, the power that keeps everything running. Then come chips, cloud infrastructure, AI models, and finally the apps you actually use every day. This is the framework NVIDIA CEO Jensen Huang laid out at the World Economic Forum in Davos, and it is arguably the clearest map we have of where the AI economy is being built.
So where does Asia fit into this cake?
The short answer: everywhere. but unevenly.
Layer 1: Energy and data centres – Asia is pouring concrete
The base of the cake is raw power and the buildings that house it. You cannot run AI without electricity, cooling, and server space. And Asia is building fast.
Over USD150 billion in AI and data centre capital was announced or advanced across the Asia Pacific region in the second half of 2025, marking a shift from ambition to actual construction. India’s Adani Group is reportedly investing ten billion dollars in hyperscale data centres. Malaysia’s Johor corridor already holds close to 500MW of live capacity. According to IDC forecasts, the Asia Pacific data centre market is on track to reach 142,600 MW of installed power by 2029, growing at 22-percent per year.
The catch?
Demand is still outrunning supply. Heat, humidity, unreliable power grids, and GPU export restrictions are real obstacles, especially across Southeast Asia.
Layer 2: Chips and compute – The engines, but not everyone has the keys
This is the most geopolitically charged layer of the cake. Chips, especially AI accelerators like NVIDIA’s GPUs, are the engines of modern AI. TSMC in Taiwan and Samsung in South Korea are the backbone of global chip supply chains. NVIDIA’s own CEO publicly asked TSMC for more wafers as demand for Blackwell chips surged.
But outside Taiwan and South Korea, the picture gets complicated. US export controls have blocked Chinese enterprises from accessing advanced NVIDIA GPUs, pushing China to pour nine point three billion dollars into private AI investment in 2024 alongside a USD47.5 billion semiconductor initiative to develop domestic alternatives. Companies like Huawei are filling the gap with their Ascend accelerators, though these remain a workaround rather than a full replacement.
The gap in investment is stark: the US spent about USD109 billion in private AI investment in 2024, nearly 12 times China’s figure.
Layer 3: Data and infrastructure software – The messy middle
Jensen Huang calls this layer the “AI factory”, the software, pipelines, and cloud platforms that turn chips into something useful. This is where a lot of Asian enterprises quietly struggle.
Fragmented data and poor pipelines are widespread across Southeast Asia and South Asia, and bad data doesn’t just slow you down, it wastes every dollar you spend on compute above it. China’s platforms are ahead here: Alibaba’s cloud unit now runs full-stack model operations through its Qwen open-source line, serving tens of thousands of enterprises through Model Studio.
Across Southeast Asia, foreign investment is accelerating, Vietnam now allows hundreds of present foreign ownership of data centres, and Indonesia is offering tax holidays of up to twenty years to attract infrastructure players. But data sovereignty rules and cross-border data restrictions continue to fragment the landscape.
Layer 4: AI Models – China is competing at the frontier
This is where reasoning happens. And China has made surprising leaps. By early 2026, China had over 700 generative AI services officially registered under national requirements. Key models include DeepSeek-V3, Alibaba’s Qwen series, ByteDance’s Kimi K2, and Baidu’s Ernie. Alibaba’s Qwen 2.5 Max outperformed Llama-3.1-405B, DeepSeek V3, and GPT-4o on several benchmarks, and Alibaba was named an emerging leader of AI model provider in the 2025 Gartner Innovation Guide.
Southeast Asia has a different challenge: most available models are English-centric, leaving over 1,200 regional languages underserved. AI Singapore and Alibaba’s Qwen team are tackling this with the SEA-LION model, which topped the multilingual SEA-HELM leaderboard. That kind of local investment matters enormously for a region where language diversity is both a barrier and an opportunity.
Layer 5: Applications — broad adoption, shallow depth
Here is where ordinary people interact with AI, through apps, chatbots, recommendations, and tools. Asia is genuinely leading in some areas. Baidu’s Apollo Go autonomous ride-hailing service has completed over 20 million rides globally and has been operating 100-percent driverless since February 2025. Seventy one percent of Asian enterprises use generative AI in business functions, and over 90-percent of Southeast Asian shoppers encounter AI-powered product recommendations online.
But here is the honest truth: most of that AI use is shallow. Only 12-percent of large enterprises say they have a comprehensive AI strategy, and 55-percent of Vietnamese enterprises cite talent gaps as their primary barrier. Broad deployment is not the same as deep value capture.
The Incomplete Cake
Asia is building hard at the bottom of the stack and deploying fast at the top. But the middle, clean data, governance frameworks, integration with legacy systems, and real technical talent, remains underdeveloped across most of the region.
Advanced AI use cases in ASEAN markets are expected to materialise closer to 2028–2030, which means there is a real 2- to 3-year window right now to invest in the foundations that will determine who captures value and who just deploys tools. The enterprises that close this middle-layer gap will not just use AI, they will build something durable from it.