We have spent the last two years watching artificial intelligence write poetry, generate video, and debug code. Yet if you hand those same powerful models a massive spreadsheet and ask for a reliable forecast, they often choke. They hallucinate numbers, lose track of the logic, or simply run out of memory. A new lab called Fundamental just emerged from stealth with a $1.2 billion valuation to fix this specific blind spot, betting that the future of enterprise AI is not about language, but about the rigid rows of a database.
Key Takeaways
- Fundamental raised $255 million in Series A funding at a $1.2 billion valuation.
- The company launched Nexus, a Large Tabular Model designed for structured enterprise data.
- Fundamental secured seven-figure contracts with Fortune 100 clients and a strategic partnership with AWS.
Fundamental has introduced a system called Nexus. They call it a Large Tabular Model (LTM) to distinguish it from the Large Language Models (LLMs) like GPT-4 or Claude. While LLMs are trained on the open internet—text, images, and audio—Nexus is built to handle structured enterprise data.
The company has secured $255 million in funding, mostly from a Series A round led by firms including Salesforce Ventures and Battery Ventures. High-profile individual investors include the CEOs of Perplexity, Brex, and Datadog. The goal is to bring the power of modern foundation models to the unglamorous backend of business data.
The big deal
Most of the world’s business data lives in tables, not in chat logs. Banks, hospitals, and logistics companies run on structured databases. The problem is that current AI models are terrible at analyzing this data. LLMs are probabilistic; they guess the next word in a sentence. When you ask them to analyze a spreadsheet with a billion rows, they often struggle to maintain accuracy or simply cannot process that much information at once.
This creates a disconnect. Companies have powerful AI for writing emails but still rely on older, manual methods for their core analytics. Fundamental claims Nexus can bridge this gap. By applying the scale of a foundation model to tabular data, they aim to automate complex reasoning tasks that currently require teams of human data scientists.
If this works, it solves a reliability problem. Businesses cannot afford “hallucinations” when forecasting revenue or tracking inventory. They need an AI that understands math and logic as well as it understands English.
How it works
Fundamental built Nexus using a different architecture than the standard models we see today. It does not use the “transformer” design that powers OpenAI or Google’s Gemini.
Think of a standard Large Language Model as a brilliant improv actor. They are great at conversation and creativity, but if you ask them to memorize a phone book, they might invent a few numbers just to keep the scene flowing. Nexus is like a forensic accountant. It does not improvise. It looks at the rows and columns, understands the strict relationships between them, and reports exactly what the data says.
Because it is built this way, Nexus is deterministic. This means if you ask it the same question five times, you get the same answer five times. It combines the reliability of old-school predictive software with the flexibility of modern AI. It can reason over datasets that are far too large for a standard LLM’s context window.
The catch
The biggest unknown is the “how.” The source text notes that Nexus does not use transformers, but it does not explain what architecture it actually uses. We know it involves pre-training and fine-tuning, but the underlying mechanics remain a black box. For a system that promises transparency in results, the method is currently opaque.
There is also a barrier to entry. Fundamental is targeting the high end of the market, securing seven-figure contracts with Fortune 100 clients. This is not currently a tool for small businesses or individual developers. It is an industrial-grade product with a price tag to match.
Finally, while the model is deterministic, it is competing against decades of established statistical tools. Data scientists trust their current methods because they understand the math behind them. Fundamental will have to prove that its “black box” is as trustworthy as the manual calculations it aims to replace.
What now?
Fundamental has launched a strategic partnership with AWS. This allows AWS customers to deploy Nexus directly from their existing cloud instances, which is a major distribution advantage.
If you work at a large enterprise with massive, messy datasets, you may see this tool integrated into your analytics stack soon. For the rest of the industry, the thing to watch is whether other AI labs begin to pivot away from transformers for specialized tasks. We may be seeing the beginning of a split between “creative” AI and “logical” AI.
