Vietnam Seeks Sovereign AI; Nvidia Says the Country Needs More GPUs

Vietnam has spent the past two years backing its ambition to become an artificial intelligence hub with policy, foreign investment and corporate partnerships. Nvidia has now put a number on one of the harder constraints facing that push: computing power.

The chipmaker estimates that Vietnam currently has about 10,000–15,000 graphics processing units, or GPUs, compared with 50,000–60,000 in South Korea, and has proposed that Hanoi develop a national-scale plan for AI computing capacity serving government agencies, companies, universities and researchers. Nvidia presented the estimates at a meeting with Vu Hai Quan, Vietnam’s minister of science and technology, on August 20.

That makes the latest discussion less about Nvidia entering Vietnam, where it is already deeply embedded, and more about what Vietnam needs to build to gain greater control over the infrastructure, data and models underpinning its AI economy.

The country’s first comprehensive AI law, which took effect on March 1, 2026, goes unusually far in spelling out that objective. It requires national AI infrastructure to include computing capacity, shared data, training and testing platforms and foundation models, while prioritising Vietnamese-language large language models (LLMs), high-performance computing, AI hardware and semiconductors. Apparently, it aims to keep the country’s AI capabilities under domestic control instead of relying on systems built elsewhere.

Vietnam is also aiming to be among Southeast Asia’s top three countries for AI research and development by 2030. Its latest national data strategy calls for a unified data ecosystem alongside shared computing infrastructure. A five-year science and technology plan envisages a national supercomputing centre and a network of specialised, high-performance computing facilities.

The problem is that sovereign AI is compute-intensive.

GPUs have become the core hardware used to train and run increasingly complex AI models, making access to large clusters of advanced chips a strategic resource for governments and technology companies alike. Vietnam’s own science ministry named three hurdles facing the country’s AI development in July: a shortage of skilled talent, data systems that don’t yet talk to each other and a shortfall in the physical hardware, chiefly GPUs, needed to run advanced models.

Nvidia’s latest proposal effectively puts scale against that ambition.

The company told the ministry that South Korea’s installed GPU base was already four times Vietnam’s at the upper end of the two estimates and could rise to 200,000–250,000 over the next five years. Those comparative figures are Nvidia’s estimates, not official inventories. Nevertheless, South Korea has made major public investments in computing infrastructure, including the government’s purchase of 13,000 Nvidia GPUs in 2025.

From Foreign Investment to Domestic Capacity

Of course, Vietnam is not starting from scratch.

Nvidia agreed to set up its first research and development centre in the country in December 2024 as well as an AI data centre. At the time, the chipmaker said it was already working with more than 100 Vietnamese AI startups and 65 universities. It also acquired VinBrain, the Vingroup-backed health technology company, as part of the expansion.

FPT (formerly Financing and Promoting Technology), Vietnam’s largest and most valuable listed IT company from Hanoi, separately announced a $200 Mn investment in an AI factory built with Nvidia technology, including H100 GPUs. The Vietnamese technology company later said its domestic AI factory was equipped with thousands of Nvidia chips and began offering computing services in 2025.

Nvidia announced this week that it is collaborating with companies such as FPT, Viettel and GreenNode in Vietnam to enhance AI infrastructure and capabilities.

But individual corporate AI factories differ from the national computing layer Vietnam is now considering.

Under the AI law, the state is expected to guide and co-ordinate national AI infrastructure while encouraging companies, universities and research institutions to invest in and share computing resources. State-backed infrastructure is also intended to support researchers and startups, a potentially important distinction in a market where access to advanced GPUs can otherwise depend on large technology companies’ budgets.

Nvidia’s proposal, therefore, fits neatly with an architecture Vietnam has already written into law: pooled computing capacity, domestic models, local data and wider access to AI infrastructure.

The company proposed three other areas alongside computing. It wants to support development of a Vietnamese-owned LLM, expand its Deep Learning Institute programmes across the country and deepen support for researchers and AI startups. It also proposed connecting Vietnamese AI companies and products more closely with the wider ASEAN market before expanding their reach globally.

Quan said the four areas were aligned with Vietnam’s science, technology and innovation strategy and instructed ministry units to work with Nvidia on a joint action plan setting out tasks and resources. But no investment figure or timetable for additional computing infrastructure was disclosed.

Who Owns the AI Layer?

That leaves the more consequential question unresolved.

Nvidia has promoted ‘sovereign AI’, describing it as the ability of countries to develop AI using domestic infrastructure, data and workforces. Its commercial position within that shift is clear. The more governments decide computing capacity is national infrastructure, the greater the demand for the accelerators, systems and software on which Nvidia has built its dominance.

For Vietnam, the equation is more complicated. Building sovereign AI does not necessarily mean building every layer of the technology domestically. The country can seek greater control over its data, models and computing capacity while continuing to depend heavily on foreign-designed chips and platforms.

Vietnam’s policy is already trying to bridge that divide. Its AI law not only prioritises computing infrastructure and Vietnamese models but also mastering AI hardware and semiconductor technologies. A separate government programme approved this month wants Vietnam to master at least four strategic technologies by 2030 and commercialise 15 products.

That makes Nvidia’s 10,000–15,000 GPU estimate a useful marker of where the country sits today.

Vietnam has moved relatively quickly from treating AI largely as an investment and adoption opportunity to framing computing, data and foundation models as national infrastructure. The next phase will be more capital-intensive. An AI law, local models and a growing startup ecosystem matter far less if companies and researchers cannot get enough computing power to train and run them.

Nvidia’s latest pitch is, in that sense, also a measure of Vietnam’s next AI challenge. Having decided it wants greater technological sovereignty, it now has to pay for the infrastructure that makes it possible.