Fundamental emerged from stealth today with $255 million in funding to tackle enterprise tabular data with a foundation model approach. The San Francisco startup, founded by DeepMind alumni, built NEXUS to predict business outcomes directly from raw tables in ERP systems, CRMs, and financial ledgers.
The pitch: skip the manual data science work. Traditional predictive modeling requires data scientists to define features, build pipelines, and train algorithms like XGBoost or Random Forest. NEXUS claims to ingest raw tables and identify patterns across columns automatically. CEO Jeremy Fraenkel positions this at a different layer than recent LLM-spreadsheet integrations: "We aren't trying to allow you to build a financial model in Excel. We are helping you make a forecast."
The technical challenge is real. LLMs tokenize numbers like words, breaking "2.3" into three separate tokens and losing numerical distribution. Tabular data is also order-invariant in ways language isn't: swapping column positions shouldn't affect predictions about patient diabetes risk, but it confuses standard transformer architectures.
NEXUS was trained on billions of tabular datasets using Amazon SageMaker HyperPod. Fundamental targets split-second decisions where humans aren't in the loop: fraud detection at point of sale, equipment failure prediction, hospital readmission risk.
This sits adjacent to, not competitive with, the lakehouse movement. Databricks (Delta Lake), Snowflake, and Microsoft Fabric focus on unifying storage and query layers using open formats like Apache Iceberg. Dremio enables SQL queries on lakes without heavy movement. Fundamental is betting on a model-based approach for structured prediction rather than storage optimization.
The real question is whether foundation models trained on tabular data generalize across industries better than purpose-built classical ML. Enterprises already invest heavily in lakehouse infrastructure for analytics. Whether they'll add another layer for predictions, or whether lakehouse vendors will absorb this capability, remains to be seen.
Notably, Fundamental hasn't disclosed customers or benchmarks. The $255 million suggests serious investor conviction, but tabular AI has been promised before. We'll watch for deployment specifics.