## The Anti-Transformer Bet Every major AI lab — OpenAI, Anthropic, Google DeepMind, Meta — builds on the transformer architecture. Fundamental abandoned it entirely. Their thesis: transformers were designed for sequential data (language), but 80% of enterprise data lives in tables, spreadsheets, and databases. Structured data where column order doesn't matter, where determinism is non-negotiable, and where "close enough" isn't good enough. Their answer is **NEXUS**, a Large Tabular Model (LTM) — a new model category that processes billions of rows of structured data without hallucinations. Same input, same output, every time. ## Why This Matters The AI hype cycle has focused almost exclusively on unstructured data: chatbots, image generators, code assistants. Meanwhile, the spreadsheets and databases that actually run pricing decisions, risk calculations, supply chains, and patient diagnostics have been left behind. Fundamental isn't competing with ChatGPT. They're competing with the army of data scientists manually building XGBoost models and random forests for every new business question. NEXUS pre-trains on billions of tables to capture cross-table patterns that single-dataset ML can't see. ## The Numbers $255 million raised ($30M seed + $225M Series A) at a $1.2–1.4 billion valuation — unicorn at launch. Led by Oak HC/FT with Valor Equity Partners, Battery Ventures, and Salesforce Ventures. Angel investors include Aravind Srinivas (Perplexity CEO), Assaf Rappaport (Wiz CEO), and Olivier Pomel (Datadog CEO). Multiple seven-figure Fortune 100 contracts were signed before the company even emerged from stealth. That's product-market fit validated by procurement departments, not Twitter hype. ## The AWS Play NEXUS is available directly through the AWS marketplace — purchasable like compute or storage. It deploys within the customer's own AWS environment, solving data residency and compliance concerns in one move. This distribution moat puts Fundamental ahead of most AI startups still doing bespoke enterprise sales. ## The Team Co-founded by Jeremy Fraenkel (CEO, UC Berkeley, ex-DeepMind) and Gabriel Suissa. The research team includes alumni from DeepMind, Isomorphic Labs, and JP Morgan. Headquarters in Menlo Park with R&D operations in Israel. ## The Contrarian Signal While every AI company chases the next GPT, Fundamental is building for the data format that LLMs handle worst. Their non-transformer architecture is a structural bet against the prevailing paradigm. And with deterministic outputs — no hallucinations, no probability — they've removed the single biggest enterprise objection to AI adoption. The boring data is where the money is.