## The Anti-ChatGPT The AI industry has a blind spot. For all the billions poured into large language models, they remain remarkably bad at the data format that runs most businesses: tables. Spreadsheets. Databases. Financial records. CRM exports. Supply chain logs. The structured, tabular data that constitutes the majority of enterprise information is precisely the data type that LLMs handle worst. They hallucinate numbers, lose track of relationships between columns, and produce non-deterministic outputs — meaning you get different answers to the same question depending on when you ask. Fundamental exists because of this gap. ## Nexus: A Different Architecture Founded by Jeremy Fraenkel (CEO), Annie Lamont, and Gabriel Suissa, Fundamental emerged from stealth in February 2026 with Nexus — a foundation model built specifically for tabular data. The technical choices are deliberately contrarian: **No transformers.** While every other AI lab builds on the transformer architecture that powers ChatGPT, Claude, and Gemini, Fundamental designed a different architecture optimized for the patterns found in structured data. Tables have fundamentally different properties than text — column relationships, data types, aggregation logic — and Fundamental argues they deserve purpose-built models. **Deterministic outputs.** Ask Nexus the same question twice, you get the same answer. This sounds obvious, but it's not how LLMs work. GPT and Claude produce probabilistic outputs — slight variations each time. For financial analysis, compliance reporting, and any use case where consistency matters, non-determinism is a dealbreaker. **Enterprise-scale.** Nexus is designed to ingest massive datasets — millions of rows across hundreds of columns — and draw insights that would take human analysts days or weeks. ## The Funding Signal Fundamental raised $255M in total funding, with the bulk coming from a $225M Series A led by Oak HC/FT, Valor Equity Partners, Battery Ventures, and Salesforce Ventures. Angel investors include Perplexity CEO Aravind Srinivas, Brex co-founder Henrique Dubugras, and Datadog CEO Olivier Pomel. At $1.4B post-money, this is among the largest Series A rounds in AI history. The investor profile tells you who the customers are: Oak HC/FT focuses on healthcare and fintech, Battery on enterprise software, Salesforce on CRM. The angel investors — Srinivas (search), Dubugras (finance), Pomel (monitoring) — all run companies that deal with massive structured datasets. The round signals that enterprise buyers are ready to pay for AI that actually works on their data, not just their documents. ## The Market Gap The structured data problem is large and underserved. According to estimates, over 80% of enterprise data is structured — stored in databases, data warehouses, and spreadsheets. Yet virtually all AI investment over the past three years has targeted unstructured data: text, images, audio, video. Companies have tried to bridge this gap by feeding tabular data into LLMs as text prompts. The results are inconsistent at best. A financial analyst asking GPT to analyze a quarterly revenue table might get a reasonable answer once and a hallucinated number the next time. That inconsistency makes LLMs unusable for the high-stakes decisions that structured data typically informs. Fundamental's bet is that the market for a reliable, deterministic structured data model is at least as large as the market for language models — possibly larger, given that structured data drives most business decisions. ## The Competitive Landscape Fundamental's direct competitors are not OpenAI or Anthropic — it's the business intelligence tools that currently serve the structured data market. Tableau, Power BI, Looker, and the SQL query layer that data analysts use daily. The pitch to enterprise buyers: instead of hiring data analysts to write SQL queries and build dashboards, ask Nexus a question in natural language and get a deterministic, auditable answer. The model understands table schemas, column relationships, and aggregation logic natively — it doesn't need a human to translate between business questions and database queries. If this works at scale, it's not a point tool — it's a replacement for an entire workflow. ## Why It Matters Fundamental represents the first serious challenge to the assumption that transformers are the right architecture for all AI problems. By building a different model for a different data type, they're testing whether the "one architecture to rule them all" thesis is correct — or whether the future of AI is a collection of specialized models, each optimized for its domain. The $255M Series A at a $1.4B valuation says the market is willing to bet on the latter. **Watch for:** Enterprise pilot results. If Nexus can demonstrably outperform LLMs on structured data tasks — with deterministic outputs and auditability — the enterprise sales cycle should be short. CFOs and compliance officers don't need to be sold on why consistent answers matter.