Today, Superb AI marks its eighth anniversary.

Over the past year, Superb AI was selected as the only Korean company in the NVIDIA Physical AI Ecosystem Partner program at CES 2026. It was also named one of the Top 5 teams in the Korean Ministry of Science and ICT’s National AI Foundation Model Consortium, the only Vision AI specialist included in the group. At CVPR 2025, the world’s leading computer vision conference, Superb AI ranked second in InsDet and fourth in FSOD, making it the only Korean startup to place among the top performers.

These milestones did not happen overnight. They were built on assets we have accumulated steadily over the past eight years. Today, we’d like to share what those assets are and where they are taking us next.

Over the last year, Superb AI entered the video analytics market, accelerated the implementation of Physical AI technologies, and more than doubled its revenue.

That revenue growth means more than the number itself. At a time when much of the AI industry is still wrestling with the same question—the technology exists, but how do you commercialize it?—Superb AI grew by responding to real demand from industrial sites. Our expansion into video analytics was not simply a business extension. It was the result of turning seven years of accumulated Vision AI expertise into an instantly deployable product for real-world operations.

Behind all of this are five core assets we have built, consistently and without distraction, over the last eight years.

5 Assets Built Over 8 Years

1. Full-Stack AI Infrastructure — One Platform Covering the Entire Lifecycle

The first problem companies face when adopting AI is not the model. It is fragmentation.

One tool for data collection. Another for labeling. Another for training. Another for deployment and operations. Stitching these pieces together can take months, and once the system reaches production, many teams find themselves back at the starting line.

Superb AI covers the full AI lifecycle—from data collection and curation to training, deployment, and operations—within a single platform. Customers can build and operate AI in one unified environment without assembling fragmented tools on their own.

This architecture means more than convenience. Because data and models live in one place, a data flywheel begins to work: the more customers use the platform, the stronger the foundation becomes for the next customer.

2. 130 Million Industrial Images — A Defensible Asset

Over the past seven years, Superb AI has accumulated more than 130 million real industrial images across 100+ domains. We estimate that rebuilding the same asset from scratch would cost approximately $68 million.

But the real value of this asset is not the price tag. It is the composition.

This is not lab data. It is data collected from real industrial environments. And it does not come from a single vertical. It spans more than 100 industries and use cases. That diversity allows models trained on this foundation to adapt quickly even in unfamiliar environments.

New Superb AI customers do not start from zero. They start with seven years of industrial learning.

3. VFM ZERO — Korea’s First Industrial Vision Foundation Model

VFM (Vision Foundation Model) ZERO is a 622 million-parameter industrial Vision Foundation Model and the first Vision Foundation Model developed in Korea.

Its significance can be summarized in three numbers.

22x to 111x greater data efficiency: ZERO was trained on just 900,000 high-quality images, yet outperformed DINO-X, which was trained on datasets ranging from 20 million to 100 million images, on benchmark tests. It is proof that a data-centric philosophy—quality over sheer quantity—still holds at the foundation model level.

One-tenth the training resources: ZERO completed training in just 10 days on 8×A100 GPUs. Compared with competitors using 64 to 128 GPUs for multiple weeks, it achieved comparable or better performance with roughly one-tenth the resources.

Top-tier global results at CVPR 2025: ZERO ranked No. 2 in InsDet (70.3 AP) and No. 4 in FSOD (47.2 AP). The companies above it were big tech firms or major Chinese research labs. Superb AI was the only Korean startup to place at that level. On our internal industrial benchmark—covering 50 domains, 200 object categories, and 150,000 instances—ZERO ranked first.

Less data. Less compute. World-class performance. That is the clearest expression of Superb AI’s technical philosophy: smart data and smart architecture can outperform scale alone.

4. An 81-Patent Portfolio — Technology Customers Can Trust

For a technology company, patents are not just numbers. They are both a legal shield that protects core innovation and a signal that customers can adopt the technology with confidence.

Superb AI currently holds a patent portfolio of 81 cases in total: 31 registered patents and 50 pending applications. We have completed filings across Korea, the United States, Japan, the EU, and the PCT international system. The portfolio covers all four generations of our technology stack, and we have also completed Freedom to Operate (FTO) verification.

FTO matters, especially for enterprise customers. It means they can adopt and expand our technology with confidence, without concern over infringement risk from third-party patents.

5. National and Global Partnerships — Validation from the Outside World

Technical excellence is not something a company declares on its own. It is something the market and the ecosystem validate.

Over the past year, Superb AI has been validated across three levels.

Validation from the global technology ecosystem

NVIDIA selected Superb AI as a CES 2026 Physical AI Ecosystem Partner, the only Korean company in the group. At GTC 2026, Superb AI secured both a booth presence and an official invited session. AWS also named Superb AI its Partner of the Year.

Validation from the government

Superb AI was selected as one of the Top 5 teams in the Korean Ministry of Science and ICT’s National AI Foundation Model Consortium, again as the only Vision AI specialist in the group. Within the LG-led consortium, Superb AI is also the only non-LG affiliate responsible for foundation model development.

Validation from the industry

Superb AI is participating in the K-Humanoid Alliance as the company responsible for visual intelligence, alongside Samsung, LG, SK, POSTECH, Seoul National University, and KAIST.

These validations are independent, but they all point to the same conclusion: Superb AI’s technology has reached a level recognized by global big tech, national research initiatives, and major Korean industrial players alike.

8 Years, 1 Principle

Over the past eight years, the AI industry has moved through wave after wave of change. Every year, a new keyword is introduced. Many technologies surged into the spotlight and then faded.

Through all of that, Superb AI held on to one principle:

Build AI that works in the customer’s real environment.

That principle became our data-centric philosophy. It became our full-stack infrastructure. It became ZERO VFM. And now, it is becoming the foundation for our next chapter in Physical AI.

Superb AI is evolving on both the business and technology fronts. The challenges ahead are clearer than ever, and so is the change we want to create. In the era of Physical AI—when AI moves beyond the data center and into factories, logistics centers, and everyday physical spaces—Superb AI aims to stand at the center as the company that builds visual intelligence best.

To everyone who has been part of the last eight years, thank you again.

We hope you will help us write the next chapter, too.