Vietnam’s AI Strategy: Why Experts Say Building Foundational Models is a Losing Battle

The global rush to build sovereign, large-scale artificial intelligence models has reached a critical turning point. As tech giants in the United States and China spend tens of billions of dollars on massive data centers and specialized silicon, smaller, developing tech ecosystems are facing a hard choice. At the GStar 2026 forum held in Ho Chi Minh City on May 29, 2026, a panel of prominent domestic and international artificial intelligence experts delivered a clear and unified message: Vietnam must not attempt to build its own foundational AI models. Doing so, they argue, is not only economically unviable but also a misallocation of the nation's unique engineering talent.

Quick summary

• Pragmatic Positioning: Rather than wasting resources on giant foundational AI models, Vietnam should adopt a "fast-follower" strategy, leveraging open-source and proprietary platforms from pioneers like Google, Meta, and OpenAI to build specialized, localized applications.
• Shift to Edge AI: Industry veterans recommend focusing on hardware-optimized, "small AI" models designed to run directly on consumer devices like smartphones, wearables, and smart glasses, bypassing the need for massive cloud GPU infrastructures.
• Human-Centric Advantage: Experts emphasize that Vietnam's true competitive edge lies in the resilience, ambition, and critical thinking of its developers, with the most crucial future skill being the ability to ask the right questions rather than merely generating answers.

Why it matters

This strategic pivot has profound economic implications for Vietnam's emerging tech ecosystem. Attempting to build a rival to GPT-4 or Gemini requires thousands of cutting-edge GPUs, vast energy grids, and billions of dollars in capital—resources that are extremely scarce and better spent elsewhere. By shifting the focus to application-level development and on-device execution (Edge AI), Vietnamese startups and enterprises can build highly profitable, low-latency, and privacy-focused solutions tailored to Southeast Asian markets. This approach democratizes AI adoption, allowing local firms to deliver immediate commercial value without incurring unsustainable cloud infrastructure costs.

Background

Vietnam's AI community has matured rapidly over the last decade, transitioning from academic interest to commercial viability. The GStar forum—organized by NTI (formerly known as VietAI)—has been a central hub for this evolution since 2018, previously drawing global heavyweights such as OpenAI CEO Sam Altman and Stanford professor Christopher Manning in 2023, followed by Google Chief Scientist Jeff Dean in 2024.

Historically, prominent Vietnamese research hubs attempted to push the boundaries of pure generative research. For instance, before its integration into Qualcomm AI Research, VinAI heavily focused on pioneering complex generative frameworks like GAN-VAE (Generative Adversarial Networks combined with Variational Autoencoders) to produce highly realistic synthetic data. However, as the hardware requirements for training modern Large Language Models (LLMs) scaled exponentially, a stark reality set in: local players could not compete with the sheer volume of compute power controlled by global hyperscalers. This realization triggered a systemic shift toward optimization and localized execution.

The 'Fast-Follower' Strategy: A Faster Route to Value

According to Curtis S. Chin, the Global Market Director at the Milken Institute, trying to duplicate the foundational work of Silicon Valley is a strategic misstep. "Vietnam should not try to build another giant foundational AI model like the major corporations have," Chin explained at GStar 2026. Instead, he advocated for the "fast-follower" approach.

In business terminology, a fast-follower does not bear the massive financial risks, trial-and-error costs, and research overhead associated with creating a new market. Instead, they closely monitor the first-movers, identify successful technological breakthroughs, and quickly adapt, optimize, and scale those innovations to capture specific markets. For Vietnam, this means taking existing open-weights models developed by Meta or Google, fine-tuning them with localized Vietnamese data, and deploying them to solve immediate regional problems in agriculture, logistics, finance, and public administration.

From Cloud to Device: The Power of Resource-Constrained AI

Dr. Bui Hai Hung, Vice President of Technology at Qualcomm AI Research and former CEO of VinAI, provided concrete technical backing for this pragmatic approach. Reflecting on his team's journey, Dr. Hung recalled the transition from chasing raw model size to solving the hard constraints of limited physical resources.

"We quickly realized that Vietnam cannot compete directly with organizations that own massive GPU clusters," Dr. Hung stated. "Consequently, we shifted our focus to resource-constrained environments, developing solutions that run on smaller GPUs or even directly on mobile phones."

This engineering shift has unlocked several major advantages:

• Data Privacy: Processing sensitive personal or corporate data directly on-device prevents leaks to external cloud servers, a key selling point for healthcare and financial applications.
• Zero Latency: Applications running locally do not suffer from network delays, which is critical for real-time systems like robotics, wearable tech, and smart glasses.
• Cost Efficiency: On-device processing eliminates the continuous operational costs of cloud hosting and API call fees, making AI deployment economically sustainable at scale.

Redefining Human Capital in the Age of Automation

As AI tools become highly democratized, the technical skills required by developers are also undergoing a paradigm shift. Dr. Luong Minh Thang, Research Director at Google DeepMind, highlighted this changing dynamic. "The most crucial skill in the future might not be answering questions, but knowing how to ask them," Dr. Thang noted, pointing out that while AI models have become highly proficient at retrieving and structuring answers, they remain inherently limited in creative and critical questioning.

This human element is precisely where Vietnam holds its greatest advantage, according to Wendy Nguyen, co-founder of Pacific Gateway Partners. Nguyen emphasized that the cultural traits of the Vietnamese people—tenacity, ambition, and a relentless drive to improve—are invaluable assets in the AI era. Even as standard of living indicators rise, the hunger to innovate remains strong, providing a resilient foundation for the country's technology sector.

Qnews24h insight

Vietnam’s refusal to chase the sovereign LLM trend is not a sign of technological weakness; it is a highly calculated, market-driven strategy. The global AI infrastructure landscape is currently suffering from a capital-expenditure bubble, where companies are buying silicon faster than they can monetize the underlying models. By focusing on Edge AI, localized software layers, and user-experience integrations, Vietnam positions itself to win the application layer of the AI revolution. Building wrappers, custom fine-tunes, and deploying low-latency models to localized hardware will generate immediate, sustainable economic returns far quicker than any attempt to train a trillion-parameter model from scratch.

Sources

This article is based on reporting and expert panel discussions from the GStar 2026 forum in Ho Chi Minh City, as documented by VnExpress.