The 39-year-old has spent nearly a decade at the company, helping develop AI models including Gemini. He also co-founded the nonprofit AI organization New Turing Institute (NTI).

Ahead of the GStar Summit 2026 on "AI & Humanity," scheduled for May 29 in Ho Chi Minh City, he speaks with VnExpress about Vietnam’s AI development, its weaknesses and opportunities during the next phase of the AI race.

- What does Vietnam’s AI ecosystem lack the most?

- The biggest gap in Vietnam’s AI sector is the shortage of computing infrastructure and world-class AI talent. While major global technology hubs possess massive supercomputing clusters to train advanced models, access to computing power in Vietnam remains limited.

Vietnam also lacks a culture of long-term fundamental research. Most resources are focused on short-term applications rather than nurturing generations of talent capable of creating system-level breakthroughs on a global scale.

- What are the weaknesses of Vietnam’s AI workforce? And how can the country attract talented engineers back home?

- It must be acknowledged that Vietnamese students and engineers have very strong foundations in mathematics and computer science. Our intelligence and technical adaptability are not inferior to the rest of the world. However, what we truly lack to compete globally is long-term vision and leadership thinking. Many Vietnamese engineers possess excellent technical foundations but lack product-thinking abilities, critical reasoning skills, and the capacity to lead projects on a global scale.

To address this, we need programs that both deepen core technical expertise in large language models and rigorously train leadership, independent thinking, presentation skills, and collaboration abilities to cultivate genuine AI leaders.

Regarding the issue of attracting overseas talent back home, there are practical barriers related to the research environment and resources. Outstanding engineers need a sufficiently large playground. The shortage of investment for moonshot projects and the absence of a strong peer ecosystem where they can continuously learn make it difficult for them to fully realize their potential if they return. To attract them, Vietnam needs to build organizations and programs that meet international standards.

Luong Minh Thang, principal scientist & director of research at Google DeepMind. Photo courtesy of WeChoice Awards

- What are the differences between the global AI research environment and Vietnam’s?

- The clearest difference lies in vision and patience. At leading institutions, teams are given resources to tackle problems that may take five to 10 years before yielding results. Failure is treated as valuable data.

Vietnam’s environment is more pragmatic, and pressure for short-term profits unintentionally limits breakthrough innovation efforts. In terms of AI capabilities, the gap also lies in ecosystem density and the ability to solve problems on a massive scale.

At leading global hubs, ecosystems are tightly integrated combinations of enormous capital, international talent, and strong academic backing. They are not merely solving current problems but constantly pushing the boundaries of technology. Meanwhile, Vietnam’s AI capabilities remain fragmented and are still largely focused on fine-tuning or building applications on top of foreign-developed models.

- What concerns you most about Vietnam’s AI community?

- It is the gap between surface-level excitement and core technological depth and the intersection between AI and humanity. Technology is advancing very quickly, but awareness of its social impact, changes in labor structures, and how people should prepare for the new era have not kept pace. Surface-level excitement is when AI is treated as a magic wand or trendy buzzword used for marketing and attracting attention.

In reality, many businesses and startups rush to integrate application programming interfaces, or APIs, from large AI models such as OpenAI’s ChatGPT or Anthropic’s Claude, but end up creating only very thin application layers. These products may initially appear impressive, such as basic question-and-answer chatbots or automated content-generation tools. However, they lack long-term competitiveness. Once global AI companies roll out new features, those surface-level application layers immediately lose value or become obsolete.

In contrast, core depth requires understanding the essence of the technology and solving foundational problems. This depth is reflected in two key aspects.

The first is mastering architecture and building foundational models. It is not enough to simply issue prompts or use APIs. True depth requires the capability to train large-scale foundation models independently, similar to what companies such as DeepSeek and Moonshot AI have done successfully.

Achieving that level requires mastery of the core layers, from optimizing computing infrastructure and leveraging proprietary datasets to deeply understanding architectures to build effective autonomous-agent control frameworks.

The second is human foundational thinking. Depth is not only about machines but also about the people who create and use them. When AI becomes capable of handling surface-level tasks such as basic coding or report summarization, the core capability of engineers becomes system architecture vision, while leaders must possess multidimensional judgment abilities.

This gap reflects the boundary between remaining merely a consumer of technology and becoming a creator and master of technology. This concern was one reason NTI partnered with Pacific Gateway Partners, an organization focused on connecting U.S. policymakers and Silicon Valley innovators through the GStar Summit.

I hope those interested in the broader AI landscape can participate and discuss how technology can be directed to serve humanity sustainably.

- Should Vietnam focus on building its own foundation models or leverage global open-source infrastructure and models? What enables an AI company to survive over the long term?

- Leveraging open-source models and global infrastructure is the most practical and intelligent approach. Nevertheless, Vietnam should not attempt to do everything and instead focus on refining models for sectors with high impact. For example, companies could specialize deeply in programming or cybersecurity, similar to the breakthrough approach recently demonstrated by Anthropic’s Mythos model.

To survive long-term, an AI company must possess proprietary data and use AI to solve real-world problems thoroughly, rather than merely showcasing technology.

- What opportunities do you see for Vietnam’s AI sector over the next five years?

- The next five years will be strongly shaped by autonomous AI, or AI agents. Vietnam needs to seize this trend quickly.

There will soon be enormous demand for understanding and building "agent harnesses," as well as mastering advanced systems such as Google Antigravity and OpenAI Codex.

Understanding how autonomous agents operate and interact with one another will reshape the entire software industry.

- AI is rapidly changing labor structures. Who will be affected the most?

- Jobs involving repetitive information processing or pattern recognition will be replaced the fastest. But this is also an opportunity to reshape education and work.

When AI becomes highly capable in mathematics, basic programming, and routine logical thinking, humans must focus on asking the right questions, thinking creatively beyond conventional limits and building human connections. Businesses must stop evaluating employees based on mechanical tasks and instead train them to become AI coordinators.

AI will replace people who do not know how to use AI. That is the reality. In the age of widespread AI adoption, the skills that will help humans maintain their unique value are taste, creativity, and critical judgment.

The last factor is particularly important. With AI now producing enormous amounts of content, code, and information at unprecedented speed, people without strong evaluative and critical-thinking abilities risk becoming overwhelmed or manipulated by AI.

Teaching young people how to judge what is right and wrong and good and bad, and how to create original perspectives is the central message.

- If you could send one message to Vietnam’s AI community, what would it be?

- Be curious, be ambitious, and be collaborative. Be curious so that you never stop learning and continue exploring technology deeply rather than skimming across trends.

Be ambitious. Do not limit yourself to local problems or become discouraged by early failures. I personally felt devastated and gave up mathematics after missing the opportunity to compete in the International Mathematical Olympiad in 2005.

But my ambition kept me in computer science. Twenty years later, that unfinished dream was realized when the Gemini Deep Think AI system achieved a gold-medal-level performance at IMO 2025. Always aim high and believe that we can conquer technological frontiers.

Finally, be collaborative; progress is never achieved alone. My sincere advice is for Vietnamese people around the world to unite, support one another, and work together. Only by setting aside personal barriers and combining our strengths toward a larger mission can Vietnam’s AI ecosystem truly take off and leave a lasting mark on the global map.