“In the AI Era, Falling Intelligence Costs Mean Corporate Competitiveness Depends on Organizational Design” [A Preview of AI World]

“The marginal cost of intelligence continues to fall. Abilities that were once expensive and scarce are becoming increasingly affordable and abundant factors of production, much like electricity and computing.” Songyee Yoon, a partner at Principle Venture Partners (PVP) who will deliver a keynote address at AI World 2026 on the 9th, told Financial News on the 2nd that the essence of the changes artificial intelligence (AI) will bring lies in “the collapse of the scarcity of intelligence.” Until now, companies needed the limited time of educated workers to secure advanced analytical, reasoning and coding capabilities. But as AI makes it possible to summon intelligence on demand like software, the way businesses and economies function could fundamentally change. Companies have generally had to hire more people to grow, but if AI takes over part of intellectual labor, the relationship between revenue and headcount may no longer remain proportional. Yoon expects corporate competitiveness in the AI era to depend less on adopting the technology itself than on “how fundamentally companies redesign their organizations.” Regarding South Korea’s strategy in the intensifying global AI competition, Yoon advised that the country should identify areas where it has structural advantages. “I believe it is a much stronger strategy for South Korea to become an indispensable key node in the global AI ecosystem than to build a small system separate from that ecosystem,” Yoon noted.
The following is an edited transcript of the interview with Yoon.
—In your AI World 2026 lecture, titled “AI Shift: When Intelligence Becomes Abundant,” what specific changes does the phrase “intelligence becomes abundant” refer to?
△Until now, intelligence has been a scarce resource tied to people. Analyzing complex problems, writing code, reviewing contracts and developing strategies required the time of highly educated individuals. Because people’s time is limited, intelligence commanded a high price. This is precisely what AI is changing. Intellectual abilities such as analysis, reasoning, coding and writing above a certain level can now be used whenever needed, much like calling up software. In other words, the marginal cost of intelligence continues to fall, and abilities that were once expensive and scarce are becoming increasingly affordable and abundant factors of production, like electricity and computing. Of course, the abundance of intelligence does not mean that wisdom or judgment will also become abundant. As the cost of generating 100 answers with AI approaches zero, it becomes even more important to decide what questions to ask, which answers to choose and what to do with them.
—If AI lowers the cost of reasoning, analysis and coding, how will the way businesses and economies function change?
△The most important change is that companies will no longer need to obtain all intellectual capabilities solely by hiring people. Until now, a company’s growth has generally meant hiring more people.
But if intelligence can be supplied like software, that relationship may break down. Revenue and headcount may no longer increase in proportion to each other as they do now. The industries likely to change first are those where work is already digitized, results can be verified relatively easily and the cost of human intellectual labor is high. Software development, customer support, marketing, finance, law, accounting and research are likely to move first.
—What do you see as the biggest obstacle to AI transformation at established companies?
△Technology may actually be the easiest part. Good models will become increasingly easy to use, and their prices will fall. By contrast, changing work processes, compensation systems and reporting structures built over decades is far more difficult. What matters is not “how well a company adopts AI,” but “how boldly it can redesign itself.”
—As massive amounts of capital flow into AI, warnings of an “AI bubble” have emerged. What criteria distinguish necessary upfront investment from excessive investment?
△Two things can be true at the same time. AI is indeed a major technological transition, and it is also highly likely to generate substantial overinvestment along the way. The same thing happened with railroads and the internet. During the dot-com bubble, enormous amounts of capital were invested in fiber-optic networks, and many investors lost money. Yet the fiber-optic infrastructure built at the time later became an important foundation of the internet economy.
That is why socially necessary investment and investment that generates good returns for investors must be considered separately. As an investor, I look at several factors: Is it actually being used? Are customers willing to pay for it? Is unit economics improving? Can the asset be repurposed for other customers or uses? And if a competitor invests the same amount of capital, can it easily achieve the same result?
—Among models, infrastructure and applications, where do you expect the greatest value to be created in the future?
△In the long term, I believe the application layer, which ultimately controls customers’ core workflows, will create the most economic value. Models will remain important, but they are likely to become increasingly commoditized. A small number of companies that build the highest-performing models may create enormous value, but not every model company will have its own economic moat.
There are still major opportunities in infrastructure. Technologies that lower inference costs or improve the utilization of computing resources, enable companies to use multiple models freely, and address data pipelines, reliability and security will become increasingly important as AI usage grows. Conversely, companies that control important customer workflows, deliver visible returns on investment and accumulate unique data and feedback through usage will grow stronger over time.
—You said models are likely to become increasingly commoditized. How should we view efforts at the national level to secure independent AI models?
△I believe South Korea’s efforts to develop its own AI capabilities are important in themselves. But more important is how we define “independence.” The ultimate goal should not be to build “the best Korean model in South Korea.”
True technological sovereignty does not mean self-sufficiency—making everything alone. It means having the ability to understand, verify, modify and operate important technologies, as well as exercising the freedom to choose when necessary.
What we ultimately need to prove is not a single performance score. We must ask whether developers overseas use it voluntarily, whether it connects easily with the global software ecosystem, whether foreign companies adopt it in real products and pay for it, and whether South Korean startups are building new products on top of it. It must be chosen not only for performance, but also for cost, reliability, safety and compatibility. If developers and customers around the world choose it even without support from the South Korean government, that is success. Conversely, if a model is used only because it is protected domestically, it is difficult to call it a global platform, even if it is technically excellent. K should be a starting point, not a boundary.
—South Korea cannot easily compete in the same way as the United States and China. Where should it focus? If you had to identify South Korea’s decisive areas in AI, what would they be?
△I do not believe that putting in the same scale of capital and GPUs as the United States and China to play exactly the same game is necessarily the best strategy. The first area is semiconductors and inference economics. South Korea has world-class capabilities in memory and semiconductor manufacturing. It can play an important role in the infrastructure that lowers the price of intelligence.
The second is industrial AI and physical AI. These include manufacturing, automobiles, shipbuilding, batteries, robotics and logistics—fields where South Korea already has world-class industrial foundations. They contain real-world data and on-site expertise that cannot easily be obtained on the internet.
The third is an open AI ecosystem connected to the world. Technological sovereignty comes not from isolation, but from having choices. I believe it is a much stronger strategy for South Korea to become an indispensable key node in the global AI ecosystem than to build a small system separate from it.
—As intelligence becomes abundant, what will become even scarcer and more valuable?
△I believe judgment and trust will become the scarcest resources. As the cost of producing answers falls, the value of good questions rises. Brands are ultimately another form of trust. In an era when information was scarce, information itself was valuable. In an era of information abundance, the value of trust—knowing what can be believed—may become even greater.
AI can produce answers and content in seconds, but trust cannot be created at the same speed. Finally, there is responsibility. AI can present countless options, but there must ultimately be an actor who decides which choice to make and takes responsibility for the outcome.
[email protected] Yoon Joo-joo Reporter