FuriosAI chief says ambitious goals are key to attracting top talent

Baek Jun-ho, CEO of FuriosAI, said Thursday that companies must set ambitious goals — not just achievable ones — if they want to attract the most talented people.

Baek made the remarks during a lecture on the second day of the 2026 KFEZ CEO Jeju Summer Forum, held at Lotte Hotel in Seogwipo, Jeju Island. His session was titled "The AI Infrastructure Power War: The Era of Data Centers Has Already Begun."

Baek, who previously worked at AMD and Samsung Electronics, founded FuriosAI in 2017 as a domestic AI semiconductor fabless startup. The company drew wide attention last year after Meta, the parent company of Facebook, offered more than 1 trillion won ($672 million) to acquire FuriosAI — an offer Baek turned down.

At the forum, Baek credited FuriosAI's rapid rise in global markets to two key factors: exceptional chip design talent and the company's strength in low-power efficiency.

Having set out with the goal of developing the world's best AI inference chip, Baek recalled that "many people advised us that making smaller chips — rather than chips for data centers — would be more realistic. But we set our sights high," he said. "Thanks to that, people with MIT PhDs and backgrounds at Meta came to join us."

He went on to say that the company focused on data center chips from the outset "because we believed it would be a high-difficulty but high-value, high-demand space." He added: "If your goals are too realistic, it becomes hard to attract outstanding talent. Instead of focusing on reasons something can't be done, you have to focus on reasons it can."

As Baek predicted, the training and inference workloads of major AI models — including Anthropic's Claude, OpenAI's ChatGPT and Google's Gemini — are now handled through large-scale AI data centers.

"It is expected that 100 GW of AI data center capacity will be added worldwide by 2030," Baek said. "At that scale, the demand for AI chips translates to hundreds of quadrillions of won in purchasing terms."

Reducing power consumption, however, remains the biggest challenge facing companies that operate AI data centers.

"Running AI draws enormous amounts of power. Virtually all costs associated with AI and data centers come down to energy," Baek said. "Ultimately, AI semiconductors that use less power will be worth more."

FuriosAI is currently mass-producing its second-generation Renegade domestic AI semiconductor, equipped with fourth-generation HBM (HBM3).

While Nvidia dominates the global AI ecosystem with GPU-based semiconductors, FuriosAI is leading the push for domestic AI chip independence through inference-focused neural processing units (NPUs). NPUs are cheaper and more power-efficient than GPUs.

Baek noted that startups that emerged to challenge Nvidia around 2016 and 2017 have now, after eight to ten years of research and development, begun releasing genuinely competitive products. "It took time for AI semiconductors like ours to become competitive and reach the market, and we are now entering the phase where that is happening in earnest," he said.

He added that FuriosAI is focused on assembling top talent and minimizing energy use. "Not only large conglomerates like Samsung and LG Group, but also global companies are adopting our chips. In final testing, we are showing higher efficiency than GPUs," he said.

Baek closed by expressing hope that "companies will embrace our AI semiconductors and adopt them widely, so that together we can build an ecosystem that becomes a new engine of growth."