APAC Is Done With AI Running on Someone Else’s Rules

Let's be frank: For most of the AI era, enterprises handed their data to a handful of global hyperscalers, got back impressive inference outputs, and called it a strategy. The data residency question — where does this actually live? — was something you left to Legal to sort out, quietly, after the fact.

That arrangement is unraveling fast across the Asia-Pacific. A new Accenture study of 756 senior leaders across 10 APAC countries finds that more than 60% of enterprises plan to increase sovereign AI and cloud investments over the next two years. It means this story is going mainstream, and it's asking CDOs some uncomfortable questions about where their intelligence actually lives.

The numbers: 64% of Southeast Asian enterprises plan to increase sovereign AI investment over the next two years. Only 25% of APAC orgs currently extend sovereignty to AI models — not just data. 57% prefer a “hybrid sovereignty” model. And roughly one-third of workloads actually need to remain sovereign — the rest can flex.

The intelligence gap hidden in plain sight

While 60% of APAC organizations apply sovereignty controls to their data, only about 25% extend those controls to their AI models. Accenture calls this the “intelligence sovereignty” gap — and it's a gaping one.

Think about what that means operationally. You've locked down your data residency. You've done the PDPA compliance work. You've got data localization policies that your DPO is proud of. And then your LLM — the thing actually generating decisions, summarizing contracts, scoring customers — is a black-box model trained on someone else's corpus, governed by someone else's values, optimized for someone else's market. Congratulations on your sovereign data lake feeding a non-sovereign brain.

“Enterprises recognize that full technological independence is neither practical nor desirable, and where and how infrastructure and workloads are managed and hosted matters more than who does it. Beyond data protection, sovereign AI can drive growth and competitiveness through improved performance of AI loads, greater relevancy for better customer experience, and it can accelerate national industrial AI agendas,” said Kunal Shah, sovereign AI lead for APAC at Accenture.

Shah's point about relevancy is poignant. A foundation model trained predominantly on English-language, Western-market data will hallucinate Thai idioms, misread Bahasa Indonesia colloquialisms, and produce customer-facing outputs that feel — at best — slightly off. At worst, reputational and regulatory liability. “Intelligence sovereignty” isn't just a geopolitical principle but a data quality and model performance argument.

Hybrid sovereignty: The pragmatist’s escape route

Nobody is seriously arguing that APAC enterprises should abandon AWS or Azure and rebuild the stack from scratch. As Shah highlighted earlier, that's neither practical nor desirable. What's emerging instead is what Accenture labels “hybrid sovereignty.” This is a model in which you strategically dial sovereignty up or down based on risk, regulation, and value.

The key insight: only about one-third of enterprise workloads actually require sovereignty. Your customer analytics for a retail loyalty programme? Probably fine on a global hyperscaler. Your health insurance claims processing or energy grid optimization data? That's where local governance, local infrastructure, and locally-trained models start to matter a lot.

“Enterprise investment in sovereign technologies is driven by the resilience agenda, with an expected focus on protecting and defending data and infrastructure in a rapidly evolving tech and policy landscape. What's equally important to long-term resilience is an organization's ability to innovate... sovereign AI presents tremendous innovation potential. We see this opening up a unique opportunity for regional players, especially telecom companies and neo-cloud operators," said Ryoji Sekido, the chief executive officer for Asia Oceania and the chief executive officer for APAC at Accenture

Sekido's reference to telcos and neo-cloud operators is worth flagging. Indonesia's Indosat Ooredoo Hutchison is already building the country's first sovereign AI cloud with Accenture and NVIDIA. The biggest blockers right now? Cost of sovereign-grade infrastructure and foundation models (cited by 40% of organizations), and limited availability of local solutions (29%). Both are problems that money and time can solve.

CDO playbook: 5 things to do

1. Audit your intelligence layer, not just your data layer. Most CDOs have mature data governance frameworks. Far fewer have applied equivalent scrutiny to AI model provenance — where it was trained, on what data, and with whose values baked in. Map your AI models the way you map your data flows.

2. Run a workload sovereignty matrix. Not everything needs sovereign treatment — and pretending it does is expensive. Categorize your AI workloads by regulatory sensitivity, data residency requirements, and performance relevance. The ~30% that genuinely need sovereign infrastructure will become obvious. Govern those hard. Let the rest stay flexible.

3. Reclassify sovereign AI as a value driver, not a cost center. The compliance framing is killing budget conversations. Locally-governed, locally-tuned models deliver measurably better performance for local-language tasks and local customer contexts. That's a product-quality argument,  and it lands differently in the boardroom.

4. Get ahead of the policy curve. APAC governments are moving fast on sovereign AI frameworks, and execution maturity varies wildly by market. If you operate across multiple APAC markets, you need a modular governance architecture that can flex per jurisdiction without rebuilding your entire data stack each time a new regulation lands.

5. Evaluate regional alternatives seriously. Neo-cloud operators and regional telcos building sovereign AI infrastructure are no longer a consolation prize. Start building those vendor relationships now — before you need them and are negotiating from a weak position.

The bottom line is that sovereign AI has moved from the geopolitics seminar room to the enterprise budget cycle. CDOs who frame it purely as a compliance overhead will get outmaneuvered by peers who recognize it as a data quality, model performance, and competitive differentiation play. The intelligence gap is real. The hybrid path is practical. The window to get ahead of it is also fast disappearing.

Image credit: iStockphoto/AlexKalina