Last month’s Asia Tech x Singapore did more than bring technology leaders into one room. It captured where the wider Asia-Pacific (APAC) region’s AI debate is heading. The takeaway was hard to miss: APAC has moved past the question of whether enterprises should use AI. The harder question is how they deploy it securely, reliably, and at scale.
That conversation matters, yet it still leaves one cost under-discussed.
Across APAC, boards are scrutinising AI spend with growing intensity. They are looking at model pricing, token bills, GPU costs, software licences, cloud consumption, and the latest invoice from another AI vendor. These costs are real. They are also the easiest to measure.
The bigger cost is rework.
Here’s the problem: A cheap model on a messy stack is still an expensive AI strategy. The enterprises that win with AI will not be the ones that negotiate the lowest token price. They will be the ones that remove the waste sitting around AI projects: duplicated data, brittle integrations, stale embeddings, and governance gaps.
It is like negotiating a cheaper electricity rate for a building with poor insulation and the windows left open. The bill may look better for a while, but the real waste is still built into the infrastructure.
Production AI exposes weak foundations
This is becoming urgent because APAC is moving from AI experimentation to AI production. Deloitte’s 2026 research found that 32% of Singapore respondents had already moved 40% or more of their AI pilots into production, while 72% plan to deploy agentic AI in several operational areas within two years.
That shift puts architecture under pressure. A proof of concept can run on curated data, a limited workflow, and a small team of experts. Production AI has to work with the business as it exists: fragmented systems, real-time transactions, customer records, documents, service histories, risk signals, compliance rules, and regional data requirements.
AI agents raise the stakes again. A chatbot retrieves information. An agent takes action. It creates records, updates states, triggers workflows, and leaves an audit trail. Once AI starts operating parts of the business, stale data and brittle integrations stop being an inconvenience. They become operational risks.
IDC’s “Modernizing Legacy: Winning in the Age of AI” study found how deep the issue runs. Across 1,400 organisations in eight Asia-Pacific markets, 95% reported project delays and 90% had encountered failed modernisation initiatives, with poor data quality consistently appearing as a key issue. The same research found that 89% of APAC organisations acknowledge technical debt as a major obstacle to modernisation, while 43% say their existing architecture makes it impossible to build new applications without extensive modernisation.
Look at what connects these failures and the answer is usually data. Models and agents are only as useful as the data feeding them, and how quickly they can reach it. An agent reasoning over a stale customer record, last week’s inventory, or a price that has already changed will not just return a wrong answer. It could take a wrong action.
That is why data architecture increasingly decides who succeeds with production AI. Without a foundation that can bring different data types together and serve current operational data when decisions are made, every new AI use case becomes another round of rework.
Leaders are building for production
The companies moving fastest are the ones cutting that rework at the source. IDC identified a cohort of leading APAC organisations generating 71% of revenue from digital sources, compared with 23% for mainstream organisations. The gap comes down to leaders treating modernisation as a sustained capability, not a one-off migration. They are investing in data foundations that can absorb change, support different data types, and give teams a cleaner path from AI idea to production application.
That is the real AI cost conversation. Not “How do we make the model cheaper?” but “How do we reduce the rework required to make AI useful?”
For technical decision-makers, this means looking beyond the visible AI stack. The issue is not only which model to use, which cloud service to select, or which AI assistant to trial. The deeper question is whether the organisation can connect AI to trusted, current operational data without rebuilding the foundation every time a new use case appears.
Three moves for APAC businesses now
The path forward is practical.
First, treat data quality and governance as hard requirements. AI systems need consistent, trusted operational data, not a patchwork of stale extracts and hand-curated datasets. If teams cannot explain where data came from, how current it is, and who is allowed to use it, they are not ready for production AI.
Second, modernise the architectures that block change. Lift-and-shift migration often moves old constraints into a new environment. It looks like progress, while the same rigidity stays in place. Organisations should prioritise architectures that support structured, semi-structured, unstructured, and vector data without forcing developers to stitch together a new stack for every AI use case.
Third, make modernisation part of how the business runs. Leaders invest in skills, change management, technology platforms, and clear business objectives. They do not wait for a perfect transformation window. They build the muscle to keep modernising as markets, regulations, customer expectations, and AI capabilities change.
That is the point APAC’s AI conversation needs to land on after Asia Tech x Singapore. Scaling AI securely and reliably is not just a matter of model economics. It is a test of whether the enterprise has removed enough architectural waste for AI to work in the real world.
The real cost problem is the rework sitting underneath.
