Artificial intelligence (AI) could add close to US$1 trillion ($1.28 trillion) to Asean’s collective GDP by 2030, according to research by EDBI and Kearney. That economic promise is adding urgency to an infrastructure race already underway. Tech giants, including Microsoft, Google and Amazon Web Services, have each deepened their Southeast Asian commitments with multibillion-dollar plans as governments race to attract the cloud and data-centre capacity needed for the next wave of AI.
Yet, infrastructure alone does not make an AI economy. The harder question is whether Southeast Asia can build the companies, talent and rules needed to capture that value, or whether it becomes a well-capitalised base for profits made elsewhere.
The smile and its limits
Governments often cast cloud and data centre investments as evidence of progress in building digital and AI capacity. Pushan Dutt, a professor of economics and political science at Insead, is not so easily persuaded. He points to the smile curve, a concept often used in semiconductor economics. It shows that the most value in a production chain is usually made at the two ends: research and design on one side, and branding, sales and distribution on the other. The middle, where manufacturing and assembly usually sit, tends to capture less value. Drawn as a chart, the curve looks like a smile.
Apply that lens to the current AI buildout and the picture for Southeast Asia is uncomfortable.
The big returns will accrue to the companies who create the models, the ones who create the architectures (like Nvidia and SK Hynix), and the ones who distribute, which are the cloud companies and the AI labs. In these three layers, Asean is missing.Pushan Dutt, a professor of economics and political science, Insead
See also: Anthropic calls for AI pause button to let humans take stock
What the region is offering sits closer to the flat bottom of that curve. Malaysia has drawn chip-packaging operations to Penang that carry margins meaningfully higher than the legacy assembly plants they are replacing. However, packaging is still contract manufacturing. The architects of the system capture the economics. The landlord collects the rent.
Dutt is careful not to dismiss the opportunity entirely. The direct gains from hosting AI infrastructure are easy to grasp. Data centres bring construction work, demand for power, operating jobs and tax revenue.
See also: Jensen Huang opens mouth. Out comes US$50 billion
The more important question is whether spillover effects follow. These are harder to measure and accumulate over time. For instance, a global AI lab may arrive as a foreign investor but its wider value depends on what local engineers, universities and suppliers absorb from it. Over time, some of that talent and know-how may move into domestic firms, start-ups or research teams, helping to build the next layer of the ecosystem.
Silicon Valley offers the classic example. Dutt points to Fairchild Semiconductor, the early chipmaker whose former employees helped seed the Valley’s start-up ecosystem, as evidence that the largest gains from technology investment often come after talent and know-how move beyond the original company.
The test for Southeast Asia, then, is not the size of the investment announcements alone. It is whether those investments lead to patents co-authored with local universities, engineers leaving multinationals to start companies, suppliers moving into higher-value work, and a thicker base of researchers and founders. “Unfortunately, most governments tend to talk about the top-line dollar numbers which are being invested, because it’s hard to measure these spillovers. Not impossible [to measure], but hard,” says Dutt.
Even when spillovers do materialise, the gains are not automatically shared. Ng Weiyi, an assistant professor from the Department of Strategy and Policy at NUS Business School, argues that the economic benefits of AI investment depend heavily on whether firms set up locally engage with the workforce, transfer skills and contribute to a domestic innovation ecosystem — rather than simply extracting value and repatriating it.
The firms that set up shop within each nation should engage with the local workforce…and in turn, the local workforce should be afforded valid protections.Ng Weiyi, an assistant professor, Department of Strategy and Policy, NUS Business School
His concern is that a locally incorporated company enjoys grants and tax incentives and then relocates its most valuable work elsewhere — leaving little behind but a registered address in a sector where physical geography is increasingly non-limiting.
To stay ahead of the latest tech trends, click here for DigitalEdge Section
Where AI actually has to work
The gap between AI investment and AI value is not merely theoretical. In a February 2026 survey by McKinsey, the Singapore Economic Development Board and Tech in Asia, nearly half of 330 executives across six Southeast Asian economies say their companies had moved beyond AI pilots. Yet, roughly six in 10 reported less than a 5% earnings impact, while 18% saw no discernible effect at all.
Source: AI in Southeast Asia report by McKinsey, the Singapore Economic Development Board and Tech in Asia
Only around 6% of organisations globally are realising significant bottom-line gains from AI, defined as deriving 11% or more of earnings before interest and taxes directly from the technology. These companies share a pattern. They redesign workflows around AI instead of bolting it onto old processes, allocate more than 20% of their digital budgets to AI and treat governance as part of scaling up instead of an obstacle.
Source: AI in Southeast Asia report by McKinsey, the Singapore Economic Development Board and Tech in Asia
That is where Murli Ravi, co-founder and managing partner of Singapore-based venture firm Tin Men Capital, sees Southeast Asia’s opening. His view is that the region’s best chance is not to compete at the infrastructure layer but to build enterprise applications that solve measurable business problems.
Tin Men’s portfolio of Southeast Asian B2B enterprise and frontier-technology start-ups shows how that plays out. Ailytics uses computer vision layered on existing CCTV infrastructure to detect safety hazards in construction and industrial facilities. It is deployed across more than 300 projects in over 10 countries, with clients including Changi Airport and ST Engineering, and claims reductions of up to 50% in manual safety inspections and up to 73% in total recordable incident rates.
Meanwhile, Groundup.ai uses acoustic monitoring to predict equipment failures before they cause unplanned downtime, with production deployments at the Republic of Singapore Navy and Hamad International Airport in Qatar. As for Graas, the agentic-commerce company has processed more than US$1 billion in gross merchandise value across seven countries, using AI agents to handle tasks such as SKU analytics and B2B order processing.
The common thread across them is simple.
The enterprise buyer is paying for a measurable operational outcome.Murli Ravi, co-founder and managing partner, Tin Men Capital
Buyers are more willing to pay than they were three years ago as volatility, inflation and the end of zero-interest-rate conditions push companies to do more with less. AI is therefore no longer seen as experimental and boards increasingly expect management teams to show how they are using it, he adds.
Moreover, the demand is not confined to Southeast Asia. Ravi shares that Tin Men’s portfolio companies are finding customers in India, the Middle East, Australia, Hong Kong and Japan. This means the region’s stronger AI start-ups are not just serving small home markets. They are using Southeast Asia as a base to build products for industries with similar problems across borders.
In contrast, start-ups selling general-purpose AI tools face a much tougher market. Without a specific industry or workflow to own, they are up against global competitors with deeper pockets and software firms adding AI to products customers already use. “A start-up competing at the infrastructure or general-purpose layer without a clear wedge is entering a fight it cannot win,” says Ravi.
What sovereignty is actually about
Building and owning the application layer calls for more than venture capital. Several Southeast Asian governments are backing sovereign AI initiatives, including regional language models built for populations largely under-represented in the English-language datasets on which global systems are trained.
AI Singapore’s SEA-LION large language model is the most prominent example. Now in its fourth iteration, it has drawn on publicly available open-source models that developers can adapt, including Alibaba’s Qwen, Google’s Gemma and Swiss AI’s Apertus. That mix raises an awkward but useful question: how sovereign can an AI system be if it still depends on global building blocks?
Sovereign capability should be understood in practical terms. It does not preclude the use of widely adopted global technologies — from Nvidia hardware to frameworks such as PyTorch — which underpin much of today’s AI development. What is critical is developing a deep understanding of the models, the data, and how performance is evaluated.Leslie Teo, senior director of AI products, AI Singapore
The more pressing challenge is staying competitive. Frontier models from tech companies, such as OpenAI, Google and Meta, have improved their multilingual capabilities substantially since SEA-LION launched. Teo’s answer is to reframe the contest. “What remains is not primarily a model gap — it is a data gap. Southeast Asia’s languages, contexts and cultural nuances are still under-represented in the datasets used to train these systems. While model capability can scale rapidly with compute, building high-quality, representative regional data is inherently a longer-term effort that requires local leadership.”
For some use cases, the best model is not always the most powerful one. Cost, data governance and local-language coverage can matter as much as capability, especially when sensitive information is involved. SEA-LION is meant to fill that gap. Teo states that the model has performed better than expected in legal and public-service settings, where local context and nuance matter, and in education use cases such as AI literacy. Others are also building specialised country-level models on top of it, including Sahabat-AI for Bahasa Indonesia and WangChanLION for Thai.
SEA-LION also sharpens the sovereignty argument. “This is not an either-or proposition. Frontier models and SEA-LION are complementary. Frontier AI systems continue to push the limits of capability but SEA-LION is focused on making those capabilities work in practice — in Southeast Asian languages, within local infrastructure constraints and for real-world regional use cases,” says Teo.
In that telling, sovereignty is not technological self-sufficiency. It is the ability to understand, adapt and govern the systems on which local institutions will depend. Putting it in commercial terms, Ravi explains: “The commercial case for sovereign AI is not in the model. It is in the companies that deploy AI into the region’s enterprises using that infrastructure. The question is not whether Southeast Asia builds its own [AI model] but whether the companies turning AI into enterprise value are owned here and built here.”
The geopolitical gift
One structural advantage the region did not engineer but is working hard to capitalise on is the US-China technology decoupling. The two countries together capture roughly 65% of global AI investment, according to Tin Men Capital’s research.
US export controls on advanced chips and China’s restrictions on rare earths are making technology supply chains harder to navigate. That gives Southeast Asia a useful role as what Tin Men calls a “third space” — jurisdictions with strong rule of law, existing technology infrastructure and enterprise demand markets where both Western and Chinese ecosystems can operate without mutual entanglement.
Both powers are courting Asean for that role. China has launched the Asean-China AI Application Cooperation Center and upgraded trade ties with the bloc. The US has put AI infrastructure, undersea cables and cybersecurity among its 2026 priorities. Within Asean, Singapore is the clearest example of the balancing act. It hosts more than 60 AI centres of excellence from companies across both ecosystems, including Alibaba Cloud, IBM, Nvidia and Oracle.
Despite seeing the geopolitical tailwind lasting, Dutt is careful not to treat neutrality as a strategy. Geography may prove more durable. For example, the Johor-Singapore Special Economic Zone combines Singapore’s infrastructure, legal system and intellectual property protections with Malaysia’s land and power supply — these structural advantages can outlast any political configuration. “Regardless of how these geopolitical things shift, the geography still remains attractive,” he says.
The funding gap
While geography may still work in Southeast Asia’s favour, the capital flowing to local AI start-ups is not keeping pace. Of approximately US$20 billion in venture investment across Asia Pacific in 2024, Southeast Asia’s AI start-ups received as little as US$1.7 billion, according to the February 2026 report by McKinsey, the Singapore Economic Development Board and Tech in Asia. The region accounted for just 122 AI funding deals that year, against 1,845 across the broader Asia Pacific.
The picture is also tightening on the venture capital side. Tin Men Capital says Southeast Asian venture capital fundraising fell to a historic low in 2025, with only four funds reaching final close, down from 33 in 2023. No first-time manager closed a fund in the second half of the year. Capital has not disappeared, but it has become more selective. Late-stage funding rose 140% in the first half of 2025, while seed funding fell 50%, as investors favoured companies with proven economics over earlier bets.
Ravi reads the reset as a necessary correction and as a warning. The cautionary tale is recent. Between 2019 and 2021, consumer technology in Southeast Asia attracted massive capital inflows. The result was inflated valuations, unsustainable burn rates, and a generation of companies that raised aggressively but could not return capital to investors. “That is the residue of a market that funded narratives before fundamentals,” he says. The risk now is that the same pattern repeats in enterprise AI. If the capital flooding in goes to companies without deep regional integration, the region could end up hosting the cycle without owning what comes out of it.
Dutt is less sanguine about the corrective power of markets alone. The spillover effects he regards as Southeast Asia’s best long-term bet — such as the engineers who stay, the firms they build, and the institutions they strengthen — are precisely the things that do not show up in a venture capital fund’s portfolio or a government’s investment tally. They accumulate slowly, are hard to attribute, and require a policy environment that consistently rewards building over hosting.
Whether the smile curve bends in Southeast Asia’s favour will depend on those choices, says Dutt. Incentives and land can bring in investment. However, they do less to ensure that the skills, suppliers and companies created around that investment remain in the region.
Ng puts the matter more plainly. Capital inflows are only one measure of success. If foreign firms set up in the region, the test is whether local workers gain skills, jobs and protections too. “The idea of prosperity and wealth will only make sense to the nation/region if it’s well distributed,” he says.
