SoftBank, Grab and PETROS bet on Malaysian AI infrastructure with 5 GW target
Photo: panumas nikhomkhai / Pexels
A $1 billion seed commitment kicks off a staged plan to build AI compute capacity in Sarawak
SoftBank, Grab, and Petroleum Sarawak Berhad (PETROS) are teaming up to build AI infrastructure in Malaysia. The long-term goal is 5 GW of compute capacity in the state of Sarawak.
The partners have been described as committing $1.85 billion to the effort. The binding paperwork so far locks in a smaller number: $1 billion in seed capital, released in stages as milestones are hit.
The money, and the fine print
The partnership has moved in two steps. On October 8, 2026, the three companies signed a framework agreement in Kuching, the Sarawak state capital.
A Joint Development Agreement followed on November 9, 2026. That second document is where real money gets committed.
The split is lopsided. SoftBank is putting in $800 million, while Grab and PETROS each contribute $100 million.
The funds won’t arrive in one lump sum. Capital gets released in tranches, each tied to specific milestones.
The $1.85 billion figure and other references to roughly $2 billion are likely rounded or broader estimates, according to the research findings. The confirmed commitment under the JDA is $1 billion.
Then there’s the headline capacity number. The partners are aiming for 5 GW of AI compute capacity over the long term.
If reached, that target would position the project among Asia’s largest initiatives of its kind. It is a target, though, not a completed build.
Why Sarawak, why now
Each partner brings a distinct piece of the puzzle. SoftBank contributes its experience in AI and semiconductors.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
Grab, the Southeast Asian super-app, offers its sprawling regional digital ecosystem. That’s a ready-made source of demand for whatever compute gets built.
PETROS covers energy. The state-owned company ties the project to Sarawak’s natural gas resources and the Low-Carbon Hub in Kuching.
The initiative also lines up with Malaysia’s National AI Action Plan for 2026 to 2030. That plan aims for the country to reach “AI nation” status by 2030.
The partners have also emphasized workforce development and community engagement.
Malaysia has been collecting big tech commitments for a while. Google has pledged $2 billion, Microsoft $2.2 billion, and AWS $6.2 billion by 2038.
The country also recorded $94 billion in ringgit-denominated data center investments from 2021 to mid-2026.
Prime Minister Anwar Ibrahim added an even bigger number to the conversation. He disclosed a discussion with SoftBank founder Masayoshi Son about Malaysia attracting up to $50 billion in AI and data center investment.
That figure has not been verified in official SoftBank announcements. Treat it as an ambition voiced by a head of government, not a signed deal.
What this means
For SoftBank, this is another sign of how aggressively Masayoshi Son is chasing AI infrastructure. An $800 million seed contribution makes Sarawak a real line item, not a side project.
For Grab, the deal is a step beyond rides, food delivery, and payments. A $100 million stake gives it a seat at the infrastructure table without betting the company on it.
For PETROS, the project offers a new customer for Sarawak’s energy resources. Turning natural gas into AI compute is a different kind of value chain than shipping it abroad.
For Malaysia, the agreement reinforces a clear strategy: become the region’s go-to hub for AI and digital infrastructure.
Several things are worth watching from here. First, whether the staged tranches beyond the initial $1 billion actually materialize.
Second, how quickly physical capacity comes online in Sarawak. A 5 GW target set without a public timeline leaves a lot of room for interpretation.
Third, whether the $50 billion figure Anwar discussed with Son turns into anything formal. Until SoftBank confirms it, that number sits firmly in the aspirational column.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.

