Enterprises across Asia Pacific have embraced AI with enthusiasm, but the real test is no longer building pilots. It is operationalizing AI to deliver sustained business value.
According to an IDC FutureScape 2026 prediction, by 2030, 50 percent of new economic value generated by digital businesses in Asia Pacific will come from organizations investing in and scaling their AI capabilities today, as enterprises embed autonomy, data intelligence, and responsible governance into their strategies to deliver measurable business impact.
Yet for many enterprises, what is holding them back is not a lack of ambition. It is the absence of a repeatable way to move from isolated experiments to production systems that can be trusted, governed, and scaled across teams.
AI experimentation creates momentum, but that momentum rarely translates into enterprise-wide value. Teams end up with fragmented tools, uneven controls, and inconsistent outcomes.
Deloitte’s Asia Pacific CFO Pulse survey reflects the same pattern. Fifty-four percent of CFOs report only pockets of AI use, while just 13 percent have scaled extensively, with data quality emerging as one of the biggest constraints.
In practice, unreliable inference becomes a critical point of failure. Once outputs cannot be trusted, people verify everything manually, and the productivity gains AI promised begin to disappear. Moving forward requires enterprises to engineer trustworthiness into the system, starting with data quality.
To close this gap, enterprises can use a four-quadrant AI factory implication matrix to determine what to prioritize, what to standardize, and where IT must take a more active role.
Beyond the model myth
Organizations need to treat data as the main production asset rather than focusing primarily on models.
In the first wave of enterprise AI, many assumed success would come from selecting increasingly sophisticated, trillion-parameter models. Experience has shown that delivering repeatable business outcomes depends far more on organizing and connecting data through a robust, secure, and flexible hybrid-cloud data architecture.
This thinking is pushing enterprises toward the AI Factory model.
In the past, organizations often treated AI as a one-off project designed to answer a single question. Increasingly, AI needs to operate as an always-on production line, where proprietary enterprise data flows continuously through secure, governed pipelines, much like raw material, and is converted into real-time business decisions.
On this production line, the AI factory implication matrix helps enterprises assess the models and data they hold so they can make more informed investment and governance decisions.
Decoding AI implications
Before debating platforms or funding the next pilot, enterprises can pressure-test their foundations with two simple questions. These should be considered for each specific AI use case, workload, or System of Intelligence.
First, how much of the intelligence being used, including the model itself, is truly proprietary?
Second, how much of the input data is uniquely owned by the enterprise, whether from customers, operations, or the market?
The answers shape platform architecture and governance requirements while also influencing where competitive advantage may emerge.
If neither the model nor the data is unique, the outcome may depend primarily on governance and speed of deployment. Where data is proprietary, the advantage lies in connecting it end to end to support trustworthy, scalable inference.
These questions should eventually be answered across all relevant AI use cases and Systems of Intelligence, creating a heatmap that helps organizations identify and prioritize the quadrants most important for scaling AI effectively.
The four quadrants
The AI factory implication matrix maps those ownership questions into four patterns. Each quadrant has different implementation implications. They are not mutually exclusive, and most enterprises will operate across more than one at the same time.
The Run approach enables organizations to access external pretrained models through interfaces or APIs without owning either the model or the underlying data. Success here depends on robust security, governance, and a center of excellence that establishes and communicates clear policies for AI adoption.
Retrieval augmented generation, or RAG, builds on this by combining external pretrained models with proprietary enterprise data to generate context-aware, business-specific insights. The priority is connecting enterprise data streams with inferencing capabilities so users can access integrated AI capabilities quickly and reliably.
Riches focuses on training custom models using enterprise data to unlock competitive differentiation and deeper business insights. This typically requires scalable, energy-efficient, high-performance infrastructure capable of supporting production-scale AI.
Finally, Regulate involves using custom models trained on external data. It requires the same scalable foundation as Riches, but with greater emphasis on legal, regulatory, and governance requirements. Because the data may be sensitive and is not owned by the organization, it must be handled with strong compliance and control.
How these patterns appear across Asia Pacific
Across Asia Pacific, these patterns are already emerging in familiar enterprise contexts.
A regional financial institution may consume external AI services within a trusted governance framework, while a large retailer may augment customer, inventory, and supply-chain data through RAG to improve decision velocity.
At the same time, an advanced manufacturer may train domain-specific models using proprietary operational data to generate differentiated insights.
A public-sector organization may apply stronger controls around compliance and data sovereignty when its AI systems depend on regulated third-party datasets.
The appropriate approach depends on the relationship between the model, the data, and the business use case.
Moving beyond shadow AI
As adoption scales, different teams often begin adopting tools and models independently. This creates shadow AI environments with inconsistent governance, security, and cost controls.
The answer is not to shut down experimentation. Enterprises need to bring structure to it.
A production path requires a data platform strategy that brings enterprise data together with guardrails, governance, and accessible pathways that can safely power AI.
It also requires continued standardization of infrastructure, protection of data integrity, and safeguards for brand trust, while still providing the speed and flexibility AI applications demand.
A decision blueprint for enterprises
The four-quadrant lens can also serve as a practical decision framework for aligning AI ambition with governance and value creation.
Enterprises should ask where they should run AI and what governance is needed to do so safely.
They should determine where retrieval augmented generation can add value and which proprietary data sources should be connected first.
They also need to identify where investment can produce the greatest business value, including which use cases warrant custom models and production-scale environments.
Finally, organizations should determine where AI requires stronger regulation, compliance, and controls because of the risk profile of the use case or the data involved.
The gap between AI ambition and AI impact ultimately comes down to execution. Enterprises moving fastest are often those with the clearest judgment about where to experiment, where to standardize, and where to invest.
With a full view through the AI factory implication matrix, enterprises can better identify where to build proprietary competitive advantage, where to adopt off-the-shelf solutions, and how to turn valuable data assets into real-world business outcomes.
Once this foundation is in place, AI can move beyond scattered, difficult-to-scale pilots and become a more reliable engine for sustained business growth.
Srikanth Seshadri is Head of Advisory & Enterprise Architects, HPE Asia Pacific.
TNGlobal INSIDER publishes contributions relevant to entrepreneurship and innovation. You may submit your own original or published contributions subject to editorial discretion.
Featured image: Steve A Johnson on Unsplash
How the deepfake arms race is accelerating AI innovation in Asia