Cubox has a subtle allure. You wouldn’t believe that one of Asia’s most fascinating AI stories is taking place inside an ordinary office building a few blocks from the subway if you were to stroll through Gangnam on any weekday. The business lacks both Naver’s marketing prowess and Samsung’s swagger. However, every time a traveler enters Incheon Airport without presenting a passport to a human, its fingerprints are literally on the gates.
The models that Cubox is renowned for are located on either side of the two halves of the company’s story. The older, hardware-focused lineup, which includes the compact ARM-based CuBox machines that enthusiasts adored years ago, is at one end of the spectrum.
| Detail | Information |
|---|---|
| Company Name | CUBOX Co., Ltd. |
| Headquarters | Gangnam District, Seoul, South Korea |
| CEO | Un Sung Nam |
| Founded | Early 2000s (biometrics-focused) |
| Industry | AI, Facial Recognition, Biometrics, Generative AI |
| Flagship Hardware Partner | HPE Cray XD6500 Supercomputer |
| TOP500 Ranking | 156th globally, with 8.136 petaFLOPs peak performance |
| Notable Deployments | Incheon International Airport e-Gates, four major government complexes |
| Business Focus | B2G, B2B (financial sector), B2C (emerging) |
| Key Products | CuBox i4Pro, Facial Recognition Terminals, Generative AI Models |
| Cloud Partner | Cloud4C (AWS Consulting Partner) |
| Public Announcement | April 30, 2024, via Hewlett Packard Enterprise |
| Target Markets | APAC, MENA, Europe, North America |
| Emerging Verticals | Healthcare, Robotics, Digital Twin |
Most people remember the CuBox i4Pro, a two-inch cube that was hardly bigger than a coaster and had gigabit Ethernet, eSATA, and an HDMI port. Although Android 4.4 was pre-installed, a serious user would switch to Debian in an hour. The microSD slot’s lack of spring loading meant that each time you wanted to change operating systems, you had to fish the card out with your fingernail. It’s a small detail, but it’s the kind of thing you remember.
Those early CuBox models outperformed their contemporaries in terms of performance. When compared to the BeagleBone Black and the Cubieboard, the standard suspects of the time, the IO speeds for SSDs were truly remarkable.
CPU clock speeds ranged from 0.8 to 1.2 gigahertz, which may seem insignificant today, but at the time, it was sufficient to power a media center, a small NAS, or a quiet home server hidden behind a TV. No MacBook would be embarrassed by Octane benchmarks, but that wasn’t the point.
The second chapter follows, which completely alters the topic of discussion. Cubes are no longer the main focus of the more recent Cubox. It has to do with supercomputers. The company’s HPE Cray XD6500 system ranked 156th on the global TOP500 list, a ranking that prompts clients in the financial sector to return calls. There’s a feeling that Cubox favored the almost seamless transition from B2G facial recognition contracts to B2B AI services for banks. When CEO Un Sung Nam discusses the company’s plans to expand into robotics, digital twins, and healthcare, it’s difficult to ignore how purposeful the timing seems.
Investors appear to think that the next phase of growth will come from the generative AI angle. The supercomputer is the engine for training large language models and large multimodal models, which Cubox needs if it wants to compete outside of Korea. It is not a vanity purchase. It’s still unclear if the strategy is effective. The partnership with Cloud4C on AWS is the kind of action that businesses take when they are serious about scale, not when they are hedging, and the APAC and MENA expansion plans are ambitious.
It’s difficult to ignore the similarities to Tesla ten years ago, when there was uncertainty about the viability of the electric vehicle wager. Cubox is not yet well-known. However, the supercomputer continues to operate, the banks continue to sign contracts, and the airports continue to use its gates. A larger narrative is emerging somewhere in that subdued consistency.
