[For more diverse corporate information on the startup featured in this article, you can view it on the Unicorn Factory big data platform 'Data Lab'.]
"When it comes to AI infrastructure, what customers are buying is not the data center or the GPU (Graphics Processing Unit) itself. They are purchasing 'finished computing' that enables them to stably and efficiently execute the large-scale training and inference they wish to run."
Ahn Jae-man, CEO of Vesel AI, said this at the "2026 NeoCloud Strategy Announcement" held on the 15th at Dream Plus in Seoul Gangnam-gu, stating that "operational capabilities to resolve bottlenecks in large-scale computing infrastructure, such as communication between GPUs, fault recovery, data input/output, and inference optimization, are the essence of the NeoCloud business."
NeoCloud refers to a business operator that rents out GPU resources specialized for AI computing. Unlike existing cloud services centered on general-purpose servers, it provides high-performance GPUs bundled with the development and operational environments required to run them. CoreWeave and Nebius in the United States are representative companies.
Vesel AI plans to embark on "AI computing exports" by combining domestic data centers and its own software (SW) to absorb AI computing demand from Silicon Valley. The company set targets to expand its current GPU operation scale of approximately 7,000 units to 10,000 units within the year, 50,000 units in 2027, and 200,000 units by 2030.
Furthermore, it aims to secure AI data center hubs with a total power capacity of 500MW (megawatts) across eight regions nationwide by 2028, and to leap forward as a global top-tier NeoCloud company based on this foundation.
Targeting 2 trillion won in revenue by 2028… "Sales within 1-2 weeks if GPU supply is confirmed"
Founded in 2020, Vesel AI started as an MLOps (Machine Learning Operations) platform supporting the development, training, deployment, and operation of AI models. It subsequently accumulated technology to operate multiple GPU clusters on a single platform and expanded its business area into NeoCloud, directly supplying GPU computing.
Revenue based on contracts through the first half of this year is 130 billion won. The company projected annual revenue of 250 billion won and set targets of 700 billion won in 2027 and over 2 trillion won in 2028. Revenue generated overseas has also exceeded 50 billion won.
Regarding GPU acquisition, the company is pursuing not only direct ownership but also joint construction with data center and cloud partners as well as long-term supply methods. The potential GPU demand currently received by Vesel AI from customers exceeds 30,000 units. Of this, 70-80% is generated overseas.
An (CEO) stated, "US AI computing demand can be up to 100 times larger than that of Korea," adding, "Our business strategy is to become a company that supplies AI computing to Silicon Valley companies based on large-scale domestic AI factories (integrated infrastructure for AI development, training, deployment, and operation) and earns foreign currency."
Domestically, the company has secured over 1,000 'B200' GPUs from SK Telecom and NVIDIA, and is promoting GPU supply and AI factory construction with NHN Cloud, Samsung SDS, and others. It also plans to expand its business into AI factories that operate NPUs (Neural Processing Units) and GPUs together through cooperation with Rebellions and FuriosaAI.
Competitiveness lies in 'Fluid Computing'… "Treating GPUs across multiple data centers as one"
The technical differentiator highlighted by Vesel AI is 'Fluid Computing.' This is a technology that integrates and operates GPUs with different regions, data centers, and types as a single large-scale computing resource.
Jeon Ji-hwan, CTO of Vesel AI, said, "Due to power, space, and cooling issues, it is difficult to place hundreds of thousands of GPUs in a single data center," adding, "The core of Fluid Computing is utilizing scattered GPUs as a single computing resource and automatically selecting the GPU suitable for the task."
Users do not need to directly select each data center or individual GPU; they only need to specify the required task and GPU scale. The system analyzes the characteristics of training and inference tasks, GPU types, and network/storage environments to automatically deploy suitable resources.
Vesel AI also provides a CLI (Command Line Interface)-based interface that allows AI agents to directly operate infrastructure and deploy workloads. Through this, the company explained that it has reduced manual human work by approximately 30% by enabling AI agents to directly create and deploy GPU tasks.
The company is also automating responses to GPU failures. It observes the status of GPUs, servers, networks, and storage in advance, isolates and replaces nodes where abnormalities are detected, and moves running workloads to normal resources. Through this, it proactively detects and responds to over 80% of all failures.
CTO Jeon stated, "GPUs are equipment where failures occur routinely as the scale grows," adding, "To operate a fleet of hundreds of thousands of units without continuously increasing personnel, every process from GPU acceptance testing to operational status observation, failure determination and isolation, and workload migration must be automated."
Promoting 400 billion won in 'GPU Finance'… Securing expansion leadership through direct purchases
Vesel AI is also actively launching fundraising for large-scale GPU acquisition. It is currently promoting 400 billion won in GPU finance and is in discussions with investment banks and financial institutional investors in the United States, Singapore, and Korea. The company plans to expand the scale of GPU finance to approximately 2 trillion won next year based on already contracted data centers.
Series C fundraising is underway with the goal of concluding by the end of this year or early next year. To continuously secure capital needed for growth, it is also pursuing Series D in parallel, targeting the first half of next year.
In addition to joint construction with partners, Vesel AI will begin direct GPU purchases using its own capital starting next year. This decision stems from the judgment that customer demand has exceeded the volume and supply speed of GPUs that can be secured through partners.
Regarding the difference from global NeoCloud companies, An (CEO) stated, "While CoreWeave and others started with large-scale capital and data center investments, Vesel AI is a company that started with software," adding, "The differentiator is understanding what training and inference customers intend to perform and flexibly supplying GPUs accordingly."
He then emphasized, "GPU prices per hour may be similar to or slightly higher than competitors', but the total cost borne by customers can be lower," stating, "We will secure competitiveness in the global AI computing market by combining Korea's HBM supply chain and data centers with Vesel AI's operational technology."
[MoneyToday startup media platform 'Unicorn Factory']
in FUTURE
