As AI continues to scale, the race to build the infrastructure across APAC that supports AI has intensified dramatically. The region is today underpinned by strong digital economies, with large internet-using populations and growing businesses adopting AI.
The region is also becoming the centre of this race. But how ready is our tech infrastructure to meet the staggering growth and energy demands?
In a first of our two-part special report, iTNews Asia speaks with key players in the industry to examine the state of play and the pressure points that the region’s tech provides face.
For the better part of five years, the tech industry sold artificial intelligence (AI) as a software problem. Better architectures. Cleaner datasets. Refined model weights. The implication was clear: intelligence scaled in the cloud, frictionlessly, limited only by engineering ingenuity.
That story is now colliding with a very different kind of constraint, one measured in gigawatts, not gradient descent.
In 2026, the AI boom has outrun the physical world's ability to support it. Data centres are queuing for grid connections that won't materialise for years. Power utilities are buckling. Hardware supply chains are stretched to breaking point. And it’s rearing it ugly head in Asia-Pacific, where a generation of enterprises is trying to move AI from boardroom experiment to production reality and discovering the infrastructure simply isn't there.
"AI has hit a physical constraint," says Lionel Yeo, CEO of Southeast Asia for ST Telemedia Global Data Centres (STT GDC).
Compute, power, and cooling have officially overtaken algorithms as the primary bottleneck of technological progress.
- Lionel Yeo, CEO, Southeast Asia, ST Telemedia Global Data Centres.
The numbers back him up. STT GDC's research finds that 71 percent of APAC organisations are stuck in what it calls the "builder phase", unable to push their models into live production because their legacy infrastructure was never built for high-density, always-on AI workloads. Only 17 percent of organisations in the region are considered genuinely future-ready.
The cracks first started in the US, and is now spreading fast to APAC
To understand APAC's predicament, start by looking at the United States, where the infrastructure crisis first became impossible to ignore.
OpenAI throttled public access to its Sora video generator, quietly redirecting those GPUs toward revenue-generating enterprise APIs. Anthropic weathered a string of outages and service degradations, eventually imposing session limits on subscribers to keep operations stable during peak hours.
Neither company's problems are isolated; they're symptoms of an ecosystem-wide supply chain crisis where capital expenditure requirements for next-generation data centres and high-bandwidth memory have triggered component shortages and pricing volatility across the global hardware market.
The same pressure points have fractured infrastructure worldwide. Europe's electricity grid is buckling under rapid industrial electrification. AI data centres across Western Europe are frequently operating at half-capacity, caught in decade-long grid connection queues. In water-stressed regions, think Arizona, Texas, Spain, the Netherlands, the cooling demands of modern AI clusters are competing directly with agriculture and residential consumption.
APAC’s infrastructure challenges
But it's APAC that has become the ultimate proving ground for this resource war.
The APAC data centre market is scaling at a pace fundamentally untethered from the region's underlying power infrastructure. The IEA projects global data centre demand crossing the 1,000 terawatt-hours threshold, and a massive portion of that load is concentrating in regional grids that weren't designed for it.
Across Southeast Asia and India, the pattern repeats: severe grid congestion, minimal cross-border interconnectivity, and underdeveloped energy storage systems are forcing tech giants to choose data centre locations based entirely on where they can secure an electrical connection, not where their users actually are.
Sumner Lemon, Senior Director of Data Centre and AI Go-To-Market for APJ at Intel, breaks the crisis into three vectors: supply chains for data centre CPUs and specialised silicon stretched to their limits; thermal and power density requirements that traditional facilities simply cannot accommodate; and construction lead times measured in years, not quarters.
AI-driven demand for compute continues to outpace available supply, particularly in data centre CPUs. This is starkly evident in Asia-Pacific as customers rapidly accelerate investments in AI infrastructure and inference workloads.
- Sumner Lemon, Senior Director of Data Centre and AI Go-To-Market, APJ, Intel.
Even AWS, which added 3.9 gigawatts of global power capacity in 2025 alone, can't keep pace with regional demand. Saji PK, Director of Asia Pacific, Japan, and China Data Centre Operations at the company, says the appetite for AI infrastructure is unlike anything the cloud computing industry has ever seen.
OpenAI's Managing Director of International, Oliver Jay, says the company is scaling available compute roughly 3x year-over-year, growing from 0.2GW in 2023 to 1.9GW in 2025. Its Stargate initiative targets 10GW long-term, with over 8GW already identified.
"Durable access to compute is now the essential strategic advantage for any technology company," Jay says.
Our approach has been to build aggressively ahead of demand, rather than react after the bottleneck has already crippled performance.
- Oliver Jay, Managing Director, International, OpenAI.
The production reality is messier than the demos
Here's the problem enterprises are actually living with: AI in production fails, a lot. Observability platform Datadog's State of AI Engineering 2026 report found that nearly 1 in 20 AI requests fail in production environments, and 60 percent of those failures stem from infrastructure capacity constraints (e.g. rate-limiting, backend timeouts) rather than model limitations or software bugs.
"As organisations transition toward complex multi-model pipelines and autonomous agent-based architectures, engineering teams are running straight into major reliability issues," warns Yadi Narayana, Field CTO for Asia Pacific and Japan at Datadog.
Organisations are facing blind spots in visibility, highly inconsistent performance, and skyrocketing token costs because they lack robust optimisation disciplines.
- Yadi Narayana, Field CTO, Asia Pacific and Japan, Datadog.
For enterprises that have built core operations around these models, the current environment introduces systemic risks that go beyond inconvenience: sudden access restrictions, deprioritised product lines, and the very real possibility that their AI vendor redirects compute capacity away from their use case without warning.
The infrastructure crunch isn't a temporary supply hiccup. It's a structural reckoning that's reshaping how the entire APAC technology ecosystem is being built.
*Look out for part 2 of our report next week on how enterprises are mapping out their strategies to cope with the demands on the AI infrastructure.
For related stories, see our earlier feature last year exploring if there is enough power to sustain the industry’s growth.
