Asia-Pacific has no shortage of ambition when it comes to artificial intelligence. Governments are announcing AI strategies, hyperscalers are expanding infrastructure, and enterprises are moving generative AI from experimentation into customer service, software development, finance and operations.
But ambition is increasingly colliding with reality.
The region’s AI challenge is no longer simply whether organisations can access powerful models or GPUs. It is whether the infrastructure, energy systems, data foundations and operating models can support AI at scale.
McKinsey estimates that APAC could account for roughly 34 percent of global data-centre demand by 2030, up from its current position as a major growth market. More than 70 percent of current APAC data-centre demand comes from traditional compute, storage and cloud workloads, while AI training and inference account for around 30 percent. By 2030, McKinsey expects the mix to move closer to 50/50.
The scale of investment reflects that trajectory. AWS, Google, Microsoft and Oracle have committed more than US$160 billion between January 2024 and May 2026 to AI infrastructure in APAC, according to McKinsey.
Alibaba has announced at least US$52 billion in global cloud and AI infrastructure investment over three years, while ByteDance could spend around US$30 billion on AI infrastructure in 2026 alone.
China is expected to remain the region’s largest market, potentially accounting for more than 70 percent of APAC data-centre demand by 2030. Outside China, new data-centre corridors are emerging across East and Southeast Asia, including Johor in Malaysia, Chonburi in Thailand, Jakarta in Indonesia and Osaka in Japan.
Yet building capacity is proving harder than expected.
The power dilemma
AI infrastructure is ultimately constrained by physical infrastructure. Data centres require enormous amounts of electricity, cooling capacity and grid connectivity.
The International Energy Agency estimates that global data-centre electricity consumption increased 17 percent last year, while AI-focused data centres grew even faster. Data-centre electricity use is expected to double by 2030, with AI-focused facilities potentially tripling their electricity consumption. At the same time, shortages of transformers, gas turbines, advanced chips and other components, alongside delays in grid connections and approvals, are creating new bottlenecks.
For enterprises, this changes the equation. Adding more compute is not necessarily the solution when the constraint is electricity, cooling, network capacity or the availability of suitable facilities.
From experimentation to economics
The economics become even more complicated when AI moves into production.
A company can experiment with a few copilots or internal chatbots at relatively modest cost. Deploying AI across thousands of employees or millions of customer interactions is different. Inference costs, latency, model selection, data movement and security can quickly become significant operating expenses.
This is why enterprises across APAC are increasingly looking beyond a simple cloud-first approach. Distributed and edge architectures can make sense for workloads where latency, data sovereignty or connectivity matters. The question is no longer simply “Where can we get more compute?” but “Where should each workload run, and what is the most efficient way to run it?”
Data presents another less visible constraint
Many organisations have large volumes of data but struggle to turn that data into usable AI inputs. Fragmented systems, inconsistent data and governance requirements can undermine AI projects before model performance even becomes the issue.
The result is a paradox: organisations can invest heavily in AI infrastructure while still failing to achieve enterprise-scale adoption because the underlying data and operating environment are not ready.
Adoption is not the same as scale
Japan provides an important reality check.
A Reuters/Nikkei Research poll published in August found that more than 80 percent of Japanese companies had yet to fully integrate AI into their operations. Around 60 percent were using AI only in limited areas, while just 16 percent reported company-wide use.
That distinction matters.
Having employees use an AI assistant is not the same as operationalising AI across an enterprise. Real scale requires governance, secure data pipelines, workflow integration, monitoring, skills and clear accountability.
There are examples showing what this can look like.
Singapore’s DBS has moved considerably beyond isolated AI experiments. In 2025, the bank deployed more than 2,000 AI models across 430 use cases, generating approximately SGD 1 billion in economic value from data analytics and AI/ML initiatives. Its technology platform and governance frameworks are designed to make AI deployment more repeatable and scalable.
The bank is now extending this approach into customer-facing generative and agentic AI. In July 2026, DBS announced that its AI-enabled virtual assistants had reached more than 10 million customers across Singapore, Hong Kong and Taiwan.
The lesson is not that every enterprise needs thousands of models. It is that scaling AI requires an operating model around AI, not simply access to AI.
The APAC region faces an ecosystem problem
Across APAC, there will not be one formula.
Singapore is focused on efficiency, digital infrastructure and sovereignty. Japan is balancing AI ambitions with cautious enterprise adoption. Southeast Asia is attracting significant data-centre investment, while Australia, Taiwan and South Korea occupy important positions in the region’s energy, semiconductor and technology ecosystems.
What connects these markets is that AI infrastructure is becoming an ecosystem problem.
Power, chips, cooling, networks, cloud capacity, data, cybersecurity, sovereignty, skills and governance increasingly have to work together. A bottleneck in any one layer can undermine an otherwise strong AI strategy.
For APAC enterprises, therefore, the next phase of AI will not be won simply by buying more GPUs or adopting the newest model.
The organisations most likely to succeed will be those that identify where AI creates genuine business value, build flexible architectures around those priorities, strengthen their data foundations and understand the real cost of deploying AI at scale.
The AI race in Asia-Pacific has entered a more difficult phase.
The question is no longer who has the biggest AI ambition. It is who can make that ambition operational, affordable and sustainable at scale.
Nivash Jeevanandam is a contributing writer to iTNews Asia. He is also an Advisory Member at the Commonwealth Educational Media Centre for Asia.
