[Edaily Reporter Kim Hyun-ah ] “We’re already a company that sells systems—not chips. Rather than focusing on chips, our goal is to become a full-stack infrastructure systems company.”

For Shin Dong-ju (36), CEO of Mobilint, the next battleground in the AI semiconductor market is not limited to data centers. He believes that “on-premises AI”—which involves installing AI servers at sites where data is generated, such as companies, factories, and control centers, to perform inference—is emerging as a viable market.

Although Mobilint has been known as an on-device and edge AI semiconductor company, its current flagship products are servers. Products equipped with “ARIES,” its first NPU, are also primarily servers.

CEO Shin said, “Servers have become the opposite of the edge,” adding, “We have both servers for data center clouds and servers for on-premises or edge environments.” He continued, “Since the main product for ARIES is the server, our current focus is on servers.”

In on-premises environments, what matters most is not the sheer performance of the NPU, but how easily it can be integrated into existing infrastructure.

CEO Shin emphasized, “From the customer’s perspective, when replacing existing GPUs with NPUs, they need a solution that can be integrated into the various form factor environments they already operate,” and stressed, “It’s crucial to strike the right balance between efficiency and versatility.”

The Reality of On-Premises AI as Seen Through CCTV

CCTV is a prime example of the growing demand for on-premises solutions.

It is difficult to immediately replace all existing CCTV systems with AI cameras, given the large number of cameras already installed nationwide. At the same time, sending all that vast amount of video footage to cloud data centers is also a burden.

A realistic alternative is to leave existing CCTV systems in place and attach AI servers to the backend—such as control centers where footage is aggregated—to analyze the data.

CEO Shin explained, “We can’t replace all the CCTV systems that are already widely deployed right now, so ultimately, we need to integrate AI at the back end to make them intelligent.” He added, “Are we going to send everything to cloud data centers? That’s difficult… Instead, the processing happens on servers at the edge.”

The same applies to fields requiring real-time processing, such as detecting anomalies or fires in factories and industrial sites. He also predicted that companies’ adoption of generative AI would drive demand for on-premises solutions, as businesses are often reluctant to send data to external clouds due to concerns over cost, information security, and regulations.

CEO Shin emphasized, “Just three years ago, ‘on-device’ was a completely unfamiliar concept,” adding, “Personally, I believe the era of on-premises has already arrived.”

Mobilin is expanding its AI semiconductor business—from on-premises to on-device—based on its Eris and Regulars platforms. In the long term, the company plans to expand into the physical AI sector, including robots and drones.

“The reason on-device AI seems easy is because the market hasn’t opened up yet”

He cited on-device AI as the more challenging market in the long term.

On-device AI is a technology that processes AI on the devices themselves—such as smartphones, PCs, cars, and robots—rather than on cloud servers. While it offers advantages in terms of fast response times and security because data isn’t sent to external servers, it requires integrating various semiconductor IPs—such as CPUs, GPUs, and NPUs—within limited power and space constraints, and then incorporating them into actual products for mass production.

CEO Shin said, “Personally, I think the on-device—or edge—market is more challenging. To be honest, I think the data center market is much easier.”

He added, “The reason the edge market seems easy right now isn’t because it actually is—it’s just that markets that haven’t opened up yet tend to seem easier.”

This is because once the market fully opens up, companies will have to compete with global giants like SamsungElectronics, Qualcomm, and Apple in sectors such as smartphones, PCs, and automobiles.

CEO Shin remarked, “Qualcomm, Samsung, and Apple are all major players, and it’s been nearly 10 years since all three began developing their own NPUs,” adding, “There is no company in Korea that has properly mass-produced NPUs to the extent thatSamsungElectronics(005930)has.”

In particular, OnDevice cannot compete based on the performance of a single NPU alone. It requires the capability to integrate various IPs—such as CPUs, GPUs, and NPUs—into a single SoC, apply them to DEVICES, and mass-produce them.

CEO Shin explained, “Accelerator chips are huge, but their chip architecture is simple,” noting that SoCs or APs for robots “consist of dozens of different types of chip IP.”

“What we lack most right now is time… We’re definitely targeting the global market.”

CEO Shin began his NPU research in 2013–2014 at the laboratory of Professor Yoo Hoe-jun in the Department of Electrical and Electronic Engineering at KAIST and received his Ph.D. in 2018 with a thesis on “Low-Power, High-Performance AI Semiconductor Processors.” Oh Jin-wook, CTO of Rebellion, is his senior. He founded his own company in 2019 at the age of 30.

What he emphasizes in the domestic NPU industry is time and reference designs.

“What we lack most right now is time.”

He argues that, given how rapidly the AI market is changing, companies cannot afford to limit themselves to the domestic market.

CEO Shin said, “The AI market is changing so rapidly that we cannot limit ourselves to the domestic market; we must unconditionally target the global market.”

He also urged the government to create a market where companies can quickly secure actual customer references, rather than providing mere research and development (R&D) support.

This is because “when customers adopt a chip, what they ultimately value most is how many meaningful references there are at the moment.”

“The Lower the Valuation, the Better… Teamwork Over Individual Talent”

CEO Shin also shared his management philosophy regarding Mobilint’s corporate valuation, following the company’s successful Series C funding round this year, which raised 70 billion won.

“If a company’s intrinsic value remains the same, I believe the lower the valuation, the better.”

He explained that if a company’s valuation is set too high at once, new employees who join later will find it difficult to reap the rewards of growth if the growth curve subsequently flattens.

“If you set the valuation too high all at once and the growth curve flattens out, employees who join at that point are at a significant disadvantage when it comes to reaping the upside.”

He said, “I am truly committed to ensuring that this company and its individual employees grow together and that the valuation resulting from this growth is shared in a truly substantive way,” adding, “That takes priority over securing my own share.”

In terms of organizational management, he prioritizes collaboration over scale. Mobilint currently has a workforce of 120 to 130 employees.

This is because AI semiconductors require the simultaneous optimization of various technologies—including chip design, cores, compilers, software, and algorithms.

CEO Shin remarked, “It’s difficult for two teams of 100 people each to collaborate that closely,” adding, “It’s crucial that communication and collaboration—both within and between teams—are extremely tight.”

He explained that if each department solves problems independently, technical issues may be resolved, but overall product performance could suffer.

“While it’s important to recruit outstanding individuals, we’ve focused from the very beginning on bringing in people whose teamwork can have a positive ripple effect on those around them…”

CEO Shin, who has been growing the AI semiconductor business since the company’s founding in 2019, envisions a next phase that goes beyond simply manufacturing NPU chips.

In the short term, the plan is to secure real-world references in the on-premises market by integrating AI servers with existing CCTV systems, industrial sites, and corporate infrastructure; in the long term, the vision is to expand “Mobilint Inside” to smartphones, PCs, automobiles, and robots.

CEO Shin’s vision for Mobilint’s next strategic move is to move beyond the competition over chip performance to create systems actually used in the field, and to leverage the references secured in the on-premises market as a springboard to expand into on-device solutions.