Dear readers,
Each technology boom is frequently accompanied by the recurring question: Is this a bubble? It is a fair warning, but also an easy argument. Investors get carried away. Companies overspend. Promises run ahead of profits.
The better test is what remains after the easy money goes away. The railway boom ruined investors, but the tracks stayed. The dot-com crash destroyed market value, but the fibre, protocols and engineering talent remained.
AI should be judged in the same way. Although some of today’s spending may not make sense, that does not mean the whole boom disappears. The lasting value may sit in a different part of the system than the market expects.
So far, markets have treated compute as the prize. Chips, data centres, cloud capacity and power have taken much of the money and attention. Goldman Sachs projects annual AI capital expenditure will reach US$765 billion ($977 billion) in 2026 and rise to US$1.6 trillion by 2031.
Southeast Asia has been pulled into that buildout. Cloud providers and technology companies are expanding across the region, while governments court investment tied to AI and digital services. The capacity matters as it gives the region more of the infrastructure needed for an AI economy.
See also: Anthropic calls for AI pause button to let humans take stock
Yet, Southeast Asia also needs companies and governments to trust AI enough to put it into real operations, such as banks using it in risk decisions or hospitals using it in clinical workflows. This is where the next constraint appears. The region does not only need access to AI, but also ways to test which systems are safe, reliable and fit for purpose.
Take Singapore, for example. The city-state wants a place in the compute buildout. Last month, OpenAI announced its more than $300 million commitment to the city-state’s AI ecosystem, while Nvidia will establish a Singapore AI research lab focused on embodied AI and efficient AI computing.
Although those deals keep Singapore close to the global AI buildout, the city-state’s sharper move is to build the trust layer around AI. That is the part Southeast Asia will need once companies move beyond pilots and start giving AI access to customers, payments, records, supply chains and public spaces.
See also: Jensen Huang opens mouth. Out comes US$50 billion
Trust is becoming a market
Besides faster models, companies need to know whether an AI system can be trusted with customer data, payments, medical information or public services. They need to know what happens and who is responsible when the system fails.
Without that layer, AI gets stuck. It can produce impressive demos and still deliver weak returns. It can save time on small tasks without becoming part of the systems that run a business.
To address that gap, Singapore’s AI Verify Foundation has launched an AI Tester Accreditation Programme to recognise third-party firms that test AI systems. The programme builds on the Global AI Assurance Sandbox, which has tested 30 AI applications across 14 sectors since February 2025, including agentic AI systems and prompt-injection risks.
This is where governance starts to look like a market. Many governments talk about AI governance in broad terms. Singapore is trying to turn it into something companies can use, including testing methods, sandboxes, accredited providers and practical guidance before harder regulation arrives.
The timing matters because the next wave of AI is harder to manage. Agentic AI does more than generate text or images. It can plan tasks, use tools and take actions for users, which also makes it riskier. Think about it: A chatbot that gives a bad answer can be corrected. Meanwhile, an AI agent connected to procurement, customer records or payment systems can create damage before anyone notices.
Singapore’s updated Model AI Governance Framework for Agentic AI is aimed at that problem. First launched at the World Economic Forum in January 2026, it was updated with feedback from more than 50 organisations, including AWS, DBS, Google and Salesforce. It now includes more than 10 case studies on agentic AI deployments.
To stay ahead of the latest tech trends, click here for DigitalEdge Section
Google’s work with the Cyber Security Agency of Singapore, the Government Technology Agency and the Infocomm Media Development Authority through a joint AI Agents Sandbox fits the same pattern. Launched in August 2025, the sandbox tested how computer-use agents behave in practice and turned the findings into guidance for governments.
This is more useful than the usual AI safety talk. It treats agentic AI as a business risk. The model matters, and so does what it can access and approve, as well as what data it can touch and who checks its work.
Punggol Digital District’s new robotics testbed belongs in this argument too. It will allow multiple operators to test physical AI in a mixed-use public area, with Certis, DHL, Grab and QuikBot among the first companies working on food and parcel delivery, cleaning and security patrols. Unlike a typical siloed robotics trial, it tests whether AI can enter shared public spaces without turning every footpath, lift lobby and delivery route into a safety or liability problem.
MERaLiON, A*STAR’s Southeast Asia-focused audio language model, adds another piece. Its latest version is designed to understand tone, intent and context across Southeast Asian languages, with possible use in healthcare, field operations and emergency response. This matters because trust in Southeast Asia has to be local. AI cannot assume English-speaking users, Western datasets or tidy company systems. If it is going to work here, assurance has to cover language, context and real-world conditions.
The regional gap
Asean has an AI governance guide that provides practical guidance for organisations designing, developing and deploying AI systems. It encourages alignment and interoperability across the region. Yet, it does not by itself create a binding regional regime. As such, AI could spread across Asean before every market has the same ability to check whether systems are safe, reliable and suited to local use.
The Asean-US AI Ministerial Roundtable held during the Asia Tech Summit (ATxSummit) in May showed where the conversation is heading. The discussion covered small-business adoption, capacity building, skills development, technology partnerships and regional AI assurance mechanisms. That last point is the business story. If assurance becomes part of how AI is bought, insured or regulated, Singapore has an opening. It is already building tools and credibility that others may later need.
However, there’s a risk that this stays optional. Governance without enforcement is guidance. Accreditation without demand is a badge. Customers, regulators, insurers and procurement teams will decide whether assurance becomes a real condition of deployment.
There is also a regional limit. Although Singapore can build assurance capacity, Asean adoption will depend on whether other governments and companies decide that testing has commercial value. A trusted hub only matters if the wider region agrees that trust is worth paying for.
This is why the bubble debate is too narrow. If AI spending slows, Southeast Asia’s issue will not only be whether every data centre, model company or AI start-up deserves its valuation. It will be whether the region has built the assurance layer needed to use AI safely. Compute may power AI, but trust will decide how far it can go.
