NextFin News - In the intensifying global race for artificial intelligence supremacy, Malaysia is at a critical crossroads. As of February 8, 2026, the nation has firmly established itself as a regional infrastructure hub, securing multi-billion dollar investments from global tech titans. However, industry experts are now urging a fundamental shift in strategy: Malaysia should abandon the ambition of building foundational large language models (LLMs) from scratch and instead master the art of AI deployment and "post-training."
According to The Edge Malaysia, Ren Ito, co-founder and chief operations officer of Tokyo-based Sakana AI, argues that for nations like Malaysia, attempting to emulate the giants by training foundational models is a strategic misstep. The capital requirements are astronomical, with US and Chinese firms pouring billions into R&D annually. Ito suggests that the utility of a marginally better foundational model rarely justifies the cost for a mid-sized economy. Instead, the focus should shift to refining open-source models to solve specific societal and economic problems, such as accelerating mortgage approvals or optimizing public services.
This call for a strategic pivot comes as Malaysia enters a decisive phase of its National AI Action Plan 2026-2030. U.S. President Trump’s administration has recently unveiled "America’s AI Action Plan," a policy roadmap that has further accelerated the global data center surge and demand for AI components. In response, Malaysia has already attracted significant infrastructure plays, including Oracle Corp’s US$6.5 billion public cloud region and a US$4.3 billion partnership between Nvidia Corp and YTL Power International Bhd to build AI supercomputers in Johor. Furthermore, the Malaysian government’s Budget 2026 has allocated RM5.9 billion for R&D and innovation, with RM2 billion specifically directed to the Malaysian Communications and Multimedia Commission (MCMC) for a sovereign AI cloud.
The economic logic behind this urged shift is rooted in the diminishing returns of "pre-training." While OpenAI and Google dominate the phase of feeding models the entire internet to create raw intelligence, the real value for Malaysia lies in "post-training"—the process of fine-tuning these models with local data and specific industry parameters. Ito notes that for most users and businesses, the difference between a GPT-4o and a GPT-5 is negligible compared to the value of a bank being able to approve a loan faster and more accurately through a specialized local model.
Data from the National AI Office (NAIO), which received RM20 million in the latest budget to become fully operational, suggests that Malaysia’s path to becoming an "AI Nation" by 2030 depends on widespread adoption rather than just hardware accumulation. The government’s RM53 million Digital Acceleration Grant is a step in this direction, aimed at boosting AI adoption among local companies. However, the broader RM20 billion investment roadmap through the end of the decade still leans heavily on infrastructure.
The impact of this strategic choice will be felt across Malaysia’s industrial landscape. If Malaysia successfully pivots to a deployment-first strategy, it could see a rapid transformation in its manufacturing and services sectors. According to a 2026 Manufacturing Industry Outlook by Deloitte, agentic AI—autonomous agents that can reason and act—is poised to elevate smart manufacturing by identifying alternative suppliers during disruptions and capturing institutional knowledge from retiring workers. For Malaysia, applying these "agentic" layers to existing open-source models would be significantly more cost-effective than building the underlying intelligence from zero.
Looking forward, the trend suggests a bifurcation in the global AI market: a few "foundational" powers providing the raw intelligence, and a multitude of "application" powers that dominate specific vertical markets. Malaysia’s geographical advantage and existing data center footprint in Johor and Cyberjaya provide the perfect laboratory for this application-centric approach. By focusing on digital trust and data security—areas Digital Minister Gobind Singh has identified as priorities for 2026—Malaysia can create a secure environment for the "post-training" of models that respect local linguistic nuances and regulatory requirements.
Ultimately, the success of Malaysia’s AI ambition will not be measured by the number of H100 chips within its borders, but by the integration of AI into its GDP. As Ito concludes, the goal should be utility over raw power. For a nation aiming to escape the middle-income trap, the ability to deploy AI to solve local inefficiencies will yield a far higher return on investment than a vanity project to build a sovereign OpenAI competitor.
Explore more exclusive insights at nextfin.ai.
