Nums AI Secures First Investment of $2.7M to Advance Tabular Foundation Model

Nums AI, a Korean startup building a tabular foundation model (TFM) for structured data, has raised $2.69 million (KRW 4 billion) in its first institutional funding round. The round was led by Stonebridge Ventures, with participation from SBVA, KT Investment, and Bass Ventures.

Structured data refers to numerical information organized in rows and columns — sales records, transaction logs, and customer profiles are common examples, and much of the data enterprises handle today falls into this category. Nums AI’s TFM is a foundation model built specifically for this kind of data. Just as large language models predict the next word in a sequence, a TFM can predict missing values in a single inference pass. Where companies previously had to spend months building separate models for tasks like demand forecasting, anomaly detection, or credit scoring, a single TFM can now handle all of these tasks instantly — sharply cutting the dedicated headcount and cost typically required to build and maintain task-specific models.

Tabular foundation models have gained rapid traction in the global AI market. In February, US-based Fundamental became the first structured-data AI company to reach unicorn status, with a valuation of $1.4 billion. In May, SAP acquired Germany’s Prior Labs and announced plans to invest more than €1 billion over the next four years. In June, NVIDIA acquired Kumo AI for roughly $400 million, and Google Research unveiled its own tabular foundation model, TabFM — drawing even the largest tech companies into the race. The foundation-model shift that began with language is now extending quickly into structured data.

Nums AI was founded by CEO Jaemin Yoo, currently an assistant professor in the Department of Computer Science and Engineering at Seoul National University, together with fellow researchers. Yoo received a Google PhD Fellowship during his doctoral studies, was appointed assistant professor at KAIST at age 29, and has spent more than a decade researching structured-data AI, including graph and time-series models. Co-founder Doo-ho Lee published three first-author papers at top AI venues including ICML and KDD within 18 months of starting his master’s degree, while co-founder Minyong Jo, after completing his master’s at Seoul National University, worked as a machine learning engineer at DeepingSource, covering the full AI development cycle from modeling and research to deployment and service operations.

With the new funding, Nums AI plans to expand its GPU infrastructure and research team and accelerate development of its TFM. The company will offer both an API service and an on-premise solution for enterprises unable to export their data externally, and intends to expand across industries where structured data is central to operations, including manufacturing, finance, commerce, and healthcare.

Hyeongsu Park, senior investment manager at Stonebridge Ventures, who led the round, said, “Most industrial decision-making happens not through language but through tables and numbers. The shift LLMs brought to language is now beginning in structured data as well. In an early-stage market where research capability determines the winners, we saw Nums AI as a team with the potential to grow into a global leader.”

Jaemin Yoo, CEO of Nums AI, said, “Going forward, how effectively a company can leverage structured data will be a key factor determining its competitiveness. Building on our research capabilities and expertise, we aim to grow into a core AI company leading the global race in model development.”