Japan, Singapore and Malaysia Team Up on ASEAN AI Compute
2 min readThree companies from three different Asian markets have agreed to build AI data centers across Southeast Asia together, with each one taking a different layer of the stack. SoftBank’s Singapore telecom arm, Singapore’s SC ZEUS Data Centers and Malaysia’s Robust HPC Group signed a memorandum of understanding effective 1 August 2026.
Why the region is short on compute
Southeast Asia has plenty of AI demand and not much of the infrastructure to serve it. Most regional workloads still route through Singapore, which has limited land and a tight grip on power allocation for data centers. Neighbouring markets have space and electricity but lack the operators, GPU supply chains and network engineering to turn that into usable AI capacity. The result is a bottleneck that no single company in the region is equipped to clear alone.
How the three-way split works
According to TNGlobal, the partners divided the job by specialty rather than by territory. SC ZEUS Data Centers will source the land and design, build and operate the core facilities. Robust HPC Group handles procurement, setup and management of the high-performance GPU clusters that sit inside them. SB Telecom Singapore, the SoftBank subsidiary, takes data hall fit-out, ICT infrastructure integration, localised deployment support and connectivity.
Initial deployments are aimed at Malaysia, Thailand, Indonesia and Vietnam rather than Singapore itself. That is the strategic point of the arrangement: capacity gets built where land and power are available, and Singapore contributes operating expertise and demand instead of hosting the racks.
Why it matters
Most large AI infrastructure announcements in Asia have been national in shape. Japan is funding Noetra, South Korea is backing its own sovereign models, and India is bankrolling homegrown foundation models. This deal runs the other way. It is a cross-border commercial arrangement in which a Japanese carrier, a Singaporean operator and a Malaysian GPU specialist each contribute one piece and share the resulting footprint across four countries.
An MoU is not a construction schedule, and no capacity figures, sites or spending numbers were disclosed. What to watch is whether the first named site lands in Malaysia’s Johor corridor, already the default overflow valve for Singapore demand, or somewhere less obvious like Vietnam. The answer will say a lot about where Southeast Asia’s AI capacity actually settles.