The global business landscape is gradually moving past the first phase of the AI boom, which focused predominantly on models and applications. Instead, conversations surrounding AI in this second phase of the AI boom are moving lower down the technology stack. Governments and large companies are thus increasingly looking at who controls the data centres, chips, cloud environments and software layers that AI depends on.
More than simply trying to solve performance issues, this shift signals how companies are making greater effort to ensure that critical workloads can keep running even if access to a supplier changes, regulations tighten or geopolitical tensions affect hardware availability.
Malaysia, for example, has allocated RM2 billion towards building a Sovereign AI Cloud. At the same time, the country recorded RM385.7 billion in data-centre-related investment between 2021 and the first half of 2026, as companies including AWS, Microsoft and Google expanded their presence. This rise in investments reflects a broader shift across Southeast Asia as the region now accounts for roughly half of Asia Pacific’s data-centre capacity under construction in the first half of 2026. A decision that might once have looked like an ordinary hardware purchase is now tied to wider questions about national infrastructure, technology dependence and the competing US and Chinese technology ecosystems.
Here are 6 signs Southeast Asia’s AI boom is entering its next phase
Sovereign AI is about control rather than complete independence
The term “sovereign AI” can sound as though every country needs to build its own chips, cloud provider and foundation model. In practice, however, the idea is more flexible. Sovereign AI is more about maintaining meaningful control over data, computing capacity, model deployment and critical digital infrastructure. This means that countries can still adopt foreign technology while deciding where sensitive data is stored, who has access to it and how dependent essential systems become on a single provider.
This distinction is important as Southeast Asian countries do not operate independently of global technology supply chains. AI infrastructure can involve US cloud providers, Taiwanese chip manufacturing, Chinese equipment and local data-centre operators. The question is therefore not whether governments can remove every external dependency, but which dependencies they are comfortable accepting. Thus, questions of data residency, control over infrastructure and operational resilience inevitably lead to sovereign AI moving higher up the agenda for founders and executives.
Malaysia shows why hardware choices now carry wider consequences
Malaysia’s sovereign AI initiative makes these trade-offs particularly visible. As mentioned earlier, the country already has the physical infrastructure needed to become a larger regional computing hub driven by data-centre-related investments in key states such as Greater Kuala Lumpur and Johor.
Bloomberg’s report that Malaysia was evaluating Huawei’s Ascend AI hardware for the project illustrates how chip decisions can influence much more than computing speed or price. While different accelerators operate most efficiently with different software tools, libraries and cloud environments, once an organisation builds its large systems around one stack, changing suppliers can become expensive.
Geopolitics adds another layer for consideration. US restrictions continue to shape access to advanced semiconductor technology and how certain hardware can move between countries. Those policies have changed repeatedly as Washington tries to control access to high-end AI computing. For Malaysia, which maintains strong commercial ties with both China and the US, preserving flexibility could be particularly important.
Malaysia’s hardware considerations thus encapsulate the dilemma faced by digitalising nations around the region. After all, hardware decisions can affect procurement, software compatibility, cybersecurity and the country’s freedom to change direction in the future.
Southeast Asian businesses increasingly have several infrastructure choices
The same issues extend beyond government projects as businesses across ASEAN increasingly have access to infrastructure from US, Chinese and regional providers. Microsoft, for example, has expanded local cloud capabilities designed around data residency requirements. In Singapore, Microsoft Fabric Go Local allows certain workloads to store core customer data inside the country. Its Indonesia Central cloud region similarly supports in-country data residency, including for highly regulated sectors such as financial services. Chinese providers are similarly expanding across Southeast Asia. Huawei has been promoting its Atlas AI computing infrastructure in Malaysia alongside local industry applications and model partners.
More competition gives companies in the region additional options and potentially reduces their dependence on any single supplier. The difficulty, however, comes when technologies do not move easily between ecosystems. An application built tightly around one cloud service or chip architecture may require significant work before it can be deployed elsewhere. What begins as a technical decision can therefore become a long-term commercial commitment.
Startups will need to plan for infrastructure flexibility
For startups, the commercial consequence is straightforward: enterprise customers may increasingly care not only about what an AI product does, but where and how it can be deployed. For example, an AI startup may naturally build its first product around whichever cloud provider, model and chip architecture is easiest to access. However, when a large customer asks for a different arrangement, the product may run into some limitations. After all, some government agencies may require local hosting while banks may want sensitive information and workloads stored inside a private cloud. Startups targeting regulated or government-linked industries could therefore be asked not only what their AI does, but where it can run.
Nonetheless, supporting every environment from the beginning would be expensive and unnecessary. A more practical approach is to avoid dependencies that are difficult to reverse. To do so, companies can keep model layers relatively modular, separate customer data from proprietary cloud services and document deployment processes so workloads can be moved more easily.
This flexibility may become particularly valuable across ASEAN, where governments are unlikely to choose identical sovereign AI strategies. Products that can operate across several cloud and computing environments could eventually gain an advantage over one locked tightly to a single technology stack.
Data-centre growth needs to create capability, not only capacity
Southeast Asia’s data-centre boom shows why this question is becoming urgent. Approximately 50% of APAC’s data-centre capacity under construction in H1 2026 was located in Southeast Asia, with Malaysia leading the region at 1,039MW, followed by Thailand at 859MW.
However, building more data centres does not automatically lead to a robust domestic AI industry. Countries can instead capture more long-term value if local companies and startups can harness that computing capacity to build products and intellectual property while engineers develop skills around it.
Again, Malaysia serves as a good example of how long-term value can be created. MIDA has shifted its data-centre discussion towards local suppliers, AI adoption, skilled jobs and home-grown innovation rather than simply the amount of infrastructure being built. Its September 2026 Data Centre Nexus event brought together global operators with 51 Malaysian vendors, compared with 17 participating suppliers the previous year.
That is ultimately the bigger test for sovereign AI. If a country is able to import hardware and host foreign workloads, it may gain investment and construction activity. If the same infrastructure also supports domestic research, startups, AI services and skilled employment, that infrastructure can begin to create a wider technology ecosystem.
Infrastructure flexibility could become an underrated advantage
Sovereign AI is unlikely to develop in the same way across Southeast Asia. Singapore, Malaysia, Indonesia, Thailand and Vietnam have different infrastructure strengths, regulations and relationships with global technology providers.
While fragmentation often creates complexity, it can also reward businesses that remain flexible and adaptable. Governments need enough control over critical infrastructure without cutting themselves off from global innovation. Enterprises need to avoid technology decisions that become unnecessarily expensive to reverse. Startups face the same problem on a smaller scale. The ability to move workloads, support different deployment environments and work across several technology ecosystems may become increasingly valuable as sovereign AI strategies develop.
While the first phase of Southeast Asia’s AI boom was largely about accessing models, the next may be about controlling what sits underneath them. As Southeast Asia progresses towards this second phase, infrastructure flexibility could become just as important as computing power itself.