Global AI race down to control over hardware, algorithms, data: Flitto CEO

Simon Lee, CEO of Flitto, gives a lecture on the current state and future of AI technology during the Corea Image Communication Institution's (CICI) Korea CQ Forum at the Qatari ambassador's residence in Seoul, Tuesday. Courtesy of CICI

In just a handful of years, artificial intelligence (AI) has reshaped how we think and engage with the world. Generative AI has already transformed our digital habits, becoming for many a primary, one-stop source of information. It is remapping the workplace, the classroom, the halls of politics and even the fabric of culture itself.

“AI didn’t just change technology; it changed human habits. And once human behavior changes, society moves at an entirely different speed,” said Simon Lee, CEO of Flitto, an AI-powered translation and live interpretation provider. His remarks came Tuesday evening at the Qatari ambassador’s residence in Seoul, during the Corea Image Communication Institution’s (CICI) Korea CQ Forum.

Today, the world has moved beyond fascination with the technology’s promise. Leaders now want their companies to stake early claims in the global AI race, with the hope of harnessing its power. The contest to develop and command AI is no longer looming, but is fully underway.

This race spans every layer of technology, Lee explained: hardware, algorithms and data.

At the ground level is hardware, such as the chips that make complex AI models possible. Graphics processing units, or GPUs, enable AI to perform massive calculations simultaneously. They are what drove Nvidia to become a chip making giant valued at $5 trillion today.

“Countries saw Nvidia’s success, and they are now trying to develop their own versions of this technology — not full GPUs, but smaller, more efficient chips called NPUs,” he said. These NPUs can run smaller AI models on compact devices like phones with greater speed and a fraction of energy consumption, while GPUs remain essential for powering giant models.

In Korea, two startups — Rebellions and Furiosa AI — have emerged as strong contenders in this race.

As large AI models have swept across the globe, a new concern has surfaced: many countries now depend on massive systems built by two superpowers, the United States and China.

“There is always the fear that user data might flow back to those countries,” the CEO said. “And because these models are trained within a specific culture and value system, their answers naturally contain cultural or political bias.”

That is why many governments are trying to build their own national AI models that reflect their languages, cultures, social norms and geopolitical realities.

“In short, countries don’t just want to use AI anymore. They want to shape it, train it on their own data and make sure it reflects their own values and national priorities,” he added.

But the reality is that not every nation can develop its own foundation model, which demands staggering computing power, billions of data points and immense financial resources.

For that reason, some countries, including Korea, are shifting their focus. Rather than pouring all resources into giant foundation models, they are investing in vertical AI, or systems tailored to specific industries and use cases.

“We may not invent every foundational technology, but we can build specialized applications and services. And this is exactly why vertical AI is becoming so important,” Lee said, noting that Flitto’s own real-time multilingual AI interpretation service, designed specifically for international conferences, is one such example.

Lastly, there is raw data, which Lee calls “the one element that actually determines whether an AI succeeds or fails, yet strangely, the one thing people talk about the least.”

The supply of high-quality, human-generated data is shrinking fast. Some experts estimate that by 2027, we will have exhausted nearly all digital data available for AI training. What will we feed AI in the years to come, if it has already consumed the bulk of humanity’s knowledge? What remains before it begins cannibalizing its own outputs and collapsing under the weight of recycled data?

It is an ongoing problem with no clear solution, but the CEO argued that this data shortage should reframe how we think about AI development. “The goal of building AI is to generate new data from human beings so that we can train that data into our solutions,” he noted. “Flitto provides real-time interpretation for people. Sure, it helps break down language barriers, but our main goal is to collect massive amounts of human pronunciation, accents [and speech patterns] coming from all kinds of people.”

That growing reservoir of data can then feed other AI-powered services, from voice recognition to machine translation, that people interact with in their daily lives.