Fundamental AI emerged from stealth with $255 million in Series A funding and a unicorn valuation of $1.4 billion — all to solve a problem that LLMs consistently fail at: understanding tabular data. Their product, Nexus, is a Large Tabular Model (LTM) designed specifically for the 70-80% of enterprise data that lives in spreadsheets, databases, and structured tables. Unlike transformer-based models that hallucinate on numerical data and struggle with cross-schema reasoning, Nexus is deterministic — it gives consistent answers every time.

The team, led by CEO Jeremy Fraenkel with former DeepMind researchers, already has seven-figure contracts with Fortune 100 clients and a strategic partnership with AWS. Nexus handles predictive modeling, cross-schema reasoning across disparate datasets, autonomous data cleaning, and fraud detection on billion-row datasets. Backed by Oak HC/FT, Battery Ventures, Salesforce Ventures, and angel investors from Perplexity and Datadog, Fundamental is building the foundation model for structured data.

But Nexus solves one critical piece of enterprise data intelligence while leaving another untouched: it can analyze what your data says right now, but it can't remember what it said last time, what changed, or why those changes matter.

Why Tabular Data Needs Its Own Model

LLMs are terrible at tabular data. Feed a spreadsheet into GPT-5 or Claude, and you get approximate answers, inconsistent calculations, and hallucinated statistics. The transformer architecture was designed for sequential token prediction, not numerical precision across millions of rows and hundreds of columns. Context window limitations make it impossible to process enterprise-scale datasets. Even when the data fits, the model's probabilistic nature means the same query can produce different answers.

Nexus takes a fundamentally different approach. It's not a transformer — it's a purpose-built architecture for structured data that operates deterministically. The same query always returns the same answer. Cross-schema reasoning connects data across tables that traditional models can't relate. Billion-row analysis happens without truncation or sampling. For enterprises where financial accuracy, regulatory compliance, and data integrity are non-negotiable, deterministic analysis isn't a feature — it's a requirement.

The market opportunity is massive because nearly all enterprise AI investment has gone toward LLMs that handle unstructured data — text, images, audio. The 70-80% of enterprise data that's structured has been underserved. Nexus addresses this gap directly, and the seven-figure Fortune 100 contracts prove the demand is real.

Analysis vs. Analytical Memory

Nexus excels at point-in-time analysis: given a dataset, it can extract insights, make predictions, and identify anomalies. But enterprise data analysis is rarely a one-time event. The same datasets are analyzed repeatedly — quarterly reports, monthly audits, weekly dashboards. The value isn't just in each individual analysis but in understanding how the data has changed over time.

A fraud detection system that identifies suspicious patterns in January's data is useful. A fraud detection system that remembers January's patterns and can compare them to February's is transformational. It can detect evolving fraud strategies, identify seasonal patterns, and flag anomalies that are only visible in the context of historical baselines. Without memory, each analysis is isolated — accurate in the moment but blind to temporal patterns.

Cross-schema reasoning benefits even more from memory. Nexus can connect data across tables within a single analysis session. But connecting data across analysis sessions — recognizing that a pattern in the sales data correlates with a change in the supply chain data from two months ago — requires persistent memory that spans both datasets and time.

How MemU Adds Analytical Memory

MemU provides the temporal memory layer that transforms point-in-time analysis into longitudinal intelligence. Every Nexus analysis generates memories: discovered patterns, anomaly baselines, cross-schema correlations. Before each subsequent analysis, relevant memories are retrieved, enabling comparisons with previous findings and detection of temporal trends.

For enterprise data teams, the combination is compelling: Nexus for precise, deterministic analysis of current data; MemU for remembering what previous analyses found. Together, they deliver the full spectrum of data intelligence — accurate in the moment and aware of the trends that only emerge over time.

Fundamental built the model for enterprise data. MemU provides the memory that turns each analysis into accumulated organizational intelligence.

Get Started

Add analytical memory to your enterprise data workflows. Explore MemU at memu.pro and on GitHub.