Nums AI's Causilo gains visibility in NVIDIA's tabular model showcase
The AMW Read
The keynote appearance raises Nums AI's visibility among tabular foundation-model developers but supplies no new evidence of deployment, revenue, or independently verified performance.
Nums AI's Causilo gains visibility in NVIDIA's tabular model showcase
South Korean startup Nums AI's Causilo appeared in NVIDIA CEO Jensen Huang's keynote at GTC Berlin on October 21, according to Platum. Huang discussed foundation models for structured tabular data and presented Causilo alongside Kumo-Tabular and Prior Labs' TabPFN-3.5. Causilo predicts outcomes such as sales volumes and customer churn from enterprise tables, including transaction histories and customer records. Nums AI released the model in September, making its code and trained weights available for evaluation while applying a separate license to commercial use.
The appearance gives a smaller model developer visibility in a foundation-model market whose applications extend beyond language and media generation. Tabular models target business prediction tasks, with Causilo aiming to let companies make predictions without separately training a model for every task. That puts its competitive proposition at the model layer, even when the end use is an enterprise workflow. The keynote mention establishes exposure; it does not establish a commercial partnership or customer adoption. Nums AI says Causilo ranked first overall and in classification and regression on TabArena at release; the report provides no independent validation of that claim.
For enterprise builders, the concrete next step is to evaluate Causilo on their own prediction tasks and review its commercial license. The relevant investment question is whether benchmark performance translates into useful predictions and less task-specific training work on customer data. Code and weight access support evaluation, but the keynote appearance alone leaves deployment economics and repeatable commercial demand unproven.
