Prepared with AI assistance from the linked primary sources. Company and government claims are attributed to their issuers. How we use AI.

Asia’s AI economy is easier to understand as a set of connected activities than as a single race. Computing providers supply resources. Model developers create reusable capabilities. Software firms connect those capabilities to tasks. Businesses change their operations around the resulting systems. Progress at one level creates possibilities for the next, but it does not guarantee that those possibilities become economic value.

India’s government reported 38,231 GPUs onboarded through 14 providers under the IndiaAI Compute initiative in March 2026. This describes a mechanism for expanding access to computing resources; it is not a count of successful commercial applications. IndiaAI programme update.

Each layer solves a different problem

At the computing level, the challenge is making resources available with suitable performance and terms. At the model level, it is producing behaviour that fits a task. At the application level, it is connecting that behaviour to relevant information and usable interfaces. At the operating level, it is deciding who acts on the output and who is responsible when something goes wrong.

Regional adaptation illustrates the middle of this chain. AI Singapore’s Gemma-SEA-LION-v4-4B-VL model builds on Gemma 3 and targets Southeast Asian language and cultural contexts. Its model card documents both regional evaluation and limitations. It represents an adaptation layer built on an existing foundation. Gemma-SEA-LION model card.

The distinction matters for commercial claims. A model can perform well on a published evaluation and still need substantial integration work before it helps a company. The company may need to clean records, control access, redesign approvals and establish quality checks. Those costs belong in the business case even when the underlying model is inexpensive to obtain.

Value appears where the layers connect

DBS reported about S$1 billion of economic value from data analytics and AI/ML in 2025. That is a company-reported aggregate covering a broad portfolio, not a measure of generative AI alone or independent proof of causation. DBS’s 2025 chairman and CEO letter.

The example points to a more useful question than which model a business uses: how does a technical capability change a measurable process? A document assistant might reduce time spent finding information, but the benefit depends on whether employees trust the answer, can inspect its basis and can finish the task without repeating the work elsewhere.

Suppliers can capture value at several points in this chain. Infrastructure providers sell dependable access. Model developers can differentiate through performance or specialisation. Application builders can own a well-understood workflow. Buyers should examine those roles separately, because the strongest supplier at one level may still depend heavily on partners at another.

Follow the handoffs

For readers tracking the regional economy, the revealing events are often agreements and products that connect these activities. The evidence to watch is specific: available compute, documented model capabilities, working integrations and measured operating results. Treating those as distinct milestones produces a clearer account of economic progress than adding every AI-related investment into one undifferentiated total.

Sources

See something we should correct? Contact the newsroom.