Africa and Southeast Asia are making aggressive moves to build homegrown AI infrastructure, but they're running into the same hard constraint: not enough graphics processing units (GPUs) to power their ambitions. From Kenya's proposed $1.5 billion data center to Vietnam's national AI strategy, governments across two continents are treating computing capacity as critical national infrastructure, much like electricity or telecommunications. The challenge is that building sovereign AI requires massive upfront investment in hardware that remains tightly controlled by a handful of Western chipmakers.

What Does Sovereign AI Infrastructure Actually Require?

Sovereign AI means more than just adopting AI tools. It means a country developing, training, and running AI systems using domestic infrastructure, data, and workforces rather than relying entirely on foreign companies. Vietnam's comprehensive AI law, which took effect on March 1, 2026, spells this out explicitly. The law requires national AI infrastructure to include computing capacity, shared data platforms, training and testing facilities, and homegrown foundation models, with a particular focus on Vietnamese-language large language models (LLMs), which are AI systems trained on vast amounts of text to understand and generate human language.

Vietnam is not alone in this ambition. Across Africa, governments are moving quickly. Nigeria unveiled a National Digital Cloud Policy targeting $750 million in private investment over two years to build data centers and AI computing infrastructure. Kenya's Amaco Energy Group is seeking approval for a $1.5 billion AI data center in Mombasa that would generate its own power through liquefied natural gas (LNG), making it independent of the national grid. Morocco has formally established four Jazari institutes to support sovereign computing, high-performance infrastructure, and skills development under its Maroc IA 2030 roadmap.

Why Is GPU Capacity the Real Bottleneck?

Here's where the ambition hits reality. Nvidia, the dominant chipmaker supplying GPUs to the world, estimates that Vietnam currently has between 10,000 and 15,000 GPUs in operation, compared with 50,000 to 60,000 in South Korea. That's a four-to-one gap. To close it, Vietnam would need to match South Korea's projected capacity of 200,000 to 250,000 GPUs over the next five years, according to Nvidia's estimates presented to Vietnam's Minister of Science and Technology on August 20.

The GPU shortage is not unique to Vietnam. Vietnam's own science ministry identified three critical hurdles in July: a shortage of skilled talent, fragmented data systems, and a shortfall in physical hardware, chiefly GPUs, needed to run advanced AI models. This is the infrastructure problem that no amount of policy can solve overnight. GPUs are the core hardware used to train and run complex AI models, making access to large clusters of advanced chips a strategic resource for governments and technology companies alike.

"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," noted the analysis of Vietnam's policy shift.

Entrepreneur Asia Pacific reporting on Vietnam's AI strategy

How Are Governments and Companies Building Computing Infrastructure?

• Public Data Centers: Nigeria is creating a National Digital Marketplace and phased government migration to cloud infrastructure, while Kenya's Technopolis Development Authority partnered with Amazon Web Services (AWS) to establish an AWS Outpost at Konza and a Startup and Innovation Centre of Excellence.
• Corporate AI Factories: Vietnam's largest IT company, FPT (formerly Financing and Promoting Technology), announced a $200 million investment in an AI factory built with Nvidia technology, including thousands of H100 GPUs, and began offering computing services in 2025.
• Regional Partnerships: Nvidia is collaborating with Vietnamese companies such as FPT, Viettel, and GreenNode to enhance AI infrastructure, while also proposing that Vietnam develop a national-scale computing plan serving government agencies, companies, universities, and researchers.
• Shared Infrastructure Models: Nigeria's AI Analytics Intelligence and Open Access Data Centres announced a partnership to offer locally hosted AI and cloud infrastructure across Africa, combining enterprise software with high-performance computing including AI Processing Units for businesses seeking local data residency and regulatory compliance.

The financing challenge is enormous. Global tech companies are already borrowing heavily to fund AI infrastructure. Advanced Micro Devices (AMD), Nvidia, and Alphabet are ramping up debt offerings to fund escalating AI capital expenditure plans. Hyperscalers like Meta, Amazon, Alphabet, Microsoft, and Oracle are projected to spend between 2.5% and 2.8% of U.S. GDP on capital expenditure in 2026 and 2027, with 2026 capex estimates revised upward from $515 billion to $775 billion. For smaller nations, the capital requirements are proportionally even more daunting.

What Does This Mean for Developing Nations' AI Sovereignty?

The gap between ambition and execution is widening. Vietnam's AI law prioritizes not just computing infrastructure but also mastering AI hardware and semiconductor technologies, with a separate government program aiming to master at least four strategic technologies by 2030 and commercialize 15 products. Yet without sufficient GPUs, that vision remains constrained.

For Africa, the challenge is compounded by the need to build multiple layers simultaneously. Nigeria is scaling autonomous security systems through Terra Industries, which closed a $52 million seed round and plans to open a Ghana factory in the fourth quarter of 2026. Ethiopia's Artificial Intelligence Institute completed a two-week bootcamp for startups developing AI-based solutions addressing social and economic challenges. Egypt launched BelMasry, a sovereign platform supporting Egyptian dialects and Modern Standard Arabic with speech-to-text, translation, and text-to-speech capabilities across 50 foreign languages.

But these initiatives depend on access to computing power. Nvidia's proposal to Vietnam effectively puts scale against ambition. The company told Vietnam's ministry that South Korea's installed GPU base could rise to 200,000 to 250,000 over the next five years, and that South Korea made major public investments in computing infrastructure, including the government's purchase of 13,000 Nvidia GPUs in 2025. Vietnam and African nations are now facing the same question: can they afford to build the computing infrastructure that sovereign AI requires?

The answer will likely determine whether sovereign AI remains a policy goal or becomes operational reality. Having decided they want greater technological sovereignty, these nations now have to pay for the infrastructure that makes it possible.