Fundamental, founded by DeepMind veterans, has raised $255 million ($30M seed, $225M Series A) at a $1.2 billion valuation for NEXUS, what it calls a Large Tabular Model. The company is betting enterprises need something fundamentally different from LLMs to handle structured data.
The pitch: LLMs struggle with tabular data because tables aren't sequential text. NEXUS uses a deterministic, non-transformer architecture designed for spreadsheets, financial records, and relational databases with billions of rows. The target use cases are predictive forecasting and enterprise analytics where legacy ML approaches currently dominate.
This is significant because structured data underpins most enterprise decision-making, from inventory management to financial planning. If the approach works, it could replace existing ML pipelines in retail, finance, and operations. That's a big "if" for a company still in stealth.
Fundamental has partnered with AWS for deployment infrastructure, a practical choice given the scale requirements. What's less clear is how quickly enterprises will adopt a new category of model when existing tools work, even if imperfectly.
The valuation and funding size are aggressive for a company without a shipped product. The Series A alone is larger than most enterprise AI companies raise across multiple rounds. Investors are clearly banking on the team's DeepMind credentials and the thesis that tabular data represents an underserved market.
Three things to watch: First, whether NEXUS can demonstrate clear advantages over existing approaches in production. Second, how enterprises respond to adding another model type to their stack. Third, whether the deterministic approach actually scales better than preprocessing data for standard LLMs.
The timing matters. Enterprise AI budgets are under pressure to show ROI. A specialized model for tabular data could be exactly what's needed, or it could be a solution looking for a problem that better data engineering would solve. We'll see which when they start shipping.