Founder / CEO / AI Pioneer
While everyone chased language models, Fraenkel went after the spreadsheet.
CEO & Co-Founder, Fundamental • DeepMind Alum • UC Berkeley ML • JPMorgan • Bridgewater • San Francisco, CA
The Story
Jeremy Fraenkel walked into the enterprise AI conversation and pointed at the elephant in the room: the spreadsheet. Not metaphorically - literally. Billions of rows of pricing data, logistics schedules, medical records, risk models. The stuff that actually runs the world. The stuff that language models, despite all their glory, are bad at.
Fraenkel's path to that insight wasn't straight. It ran through JPMorgan's trading floors, Bridgewater Associates' data-driven investment culture, and two startup founding rounds - both of which he exited. He went back to school at UC Berkeley for a graduate degree in Machine Learning, then found his way to DeepMind, the AI research lab that operates like a permanent revolution. There he met Marta Garnelo, a research scientist who would become Fundamental's Chief Science Officer and write the academic whitepaper that became the company's technical backbone. He also connected with co-founder Gabriel Suissa.
Together they made a bet that looked, at the time, like a contrarian long shot. In 2024, while the AI industry poured its energy into making text generation smoother, Fraenkel, Garnelo, and Suissa set up shop in San Francisco and disappeared into 18 months of stealth. The goal: a foundation model trained specifically on tabular data. Not text. Not images. Tables. The kind that live in databases and spreadsheets and dictate whether a hospital orders the right drugs, whether a retailer overbuilds inventory, whether a bank extends credit at the right price.
What came out of those 18 months is NEXUS - a Large Tabular Model that delivers predictions from raw enterprise tables with a single line of code. No preprocessing. No feature engineering. No army of data scientists hand-crafting models for each use case. Fraenkel describes it as "the OS for business decisions." Salesforce Ventures, in backing the company, put it differently: NEXUS does for structured data what Claude does for natural language.
The analogy is precise. Just as large language models replaced bespoke NLP pipelines with one model that handles everything from summarization to translation, NEXUS replaces the laborious machine learning workflow - months of cleaning data, building features, training and retraining models - with a single foundation model that generalizes across tasks. Customer churn. Equipment failure prediction. Demand forecasting. Traffic analysis. One model, one line of code.
Fundamental emerged from stealth in February 2026 with $255 million in total funding: a $30 million seed round followed by a $225 million Series A led by Oak HC/FT, with participation from Valor Equity Partners, Battery Ventures, Salesforce Ventures, and Israel's Hetz Ventures. The post-money valuation: $1.4 billion. They had seven-figure contracts with Fortune 100 companies already signed. They had a strategic AWS partnership that made NEXUS available as a first-party product on the AWS Marketplace - a distribution channel that most AI startups can only fantasize about.
The investors who backed this included operators who've built category-defining companies: Aravind Srinivas of Perplexity, Assaf Rappaport of Wiz, Henrique Dubugras of Brex, Olivier Pomel of Datadog. These are not passive checks. They are endorsements from people who move fast, pattern-match on founder quality, and don't write seven-figure checks into crowded markets unless they believe the thesis is correct.
Fraenkel's thesis - the one that explains everything - is that the AI revolution solved the wrong half of the problem first. "We've only solved half the brain," he told Jason Calacanis on This Week in AI. "Language and creativity. But not the math, statistical, and structured reasoning side." LLMs transformed unstructured data. The 70-80% of enterprise data that lives in rows and columns? Still waiting. That's Fundamental's market.
Oak HC/FT described Fraenkel as having "a rare combination of technical depth and commercial execution" - with financial literacy that is "uncommon in the world of AI research." That dual fluency - the ability to read a Bridgewater model and a DeepMind whitepaper in the same week - is probably what makes the company's go-to-market as interesting as its technology. NEXUS isn't being sold to AI researchers. It's being sold to CFOs, supply chain leaders, and risk managers who have spent decades throwing data science teams at problems that should have been solved by a model.
Fundamental's R&D team extends to Israel, with engineers who previously built at AI21 Labs - broadening the technical bench and adding a second center of gravity to the company. It's a geographically distributed research operation, which reflects Fraenkel's willingness to build where the talent is rather than where the investors are watching.
The company is 57 people strong and growing. The technology is live. The contracts are signed. And Fraenkel is mid-stride, building what he believes is the missing half of enterprise AI - one table at a time.
The Technology
Traditional machine learning on tabular data is a grind. A data scientist joins a company, spends months understanding the data, engineers features by hand, trains a model for a specific prediction task, and then repeats the entire process for the next task. At enterprise scale, this means dozens of siloed models, maintained by teams of specialists, each built for one purpose. It's expensive, slow, and fragile.
NEXUS rejects that paradigm. Trained on billions of real-world enterprise tables, it learns the deep structure of tabular data - the non-linear relationships, the hidden patterns, the cross-column dependencies - during pre-training. When an enterprise deploys it, the model already knows how tables work. It doesn't need to be re-taught. It just needs the data.
The result: predictions from raw tables with a single line of code. No preprocessing required. No feature engineering. No fine-tuning for each new use case. One model that handles customer churn, demand forecasting, fraud detection, equipment failure prediction, price optimization, and drug discovery from the same underlying architecture.
Unlike transformer-based LLMs that tokenize data and lose precision when applied to structured formats, NEXUS is built natively for tables. It understands columns as types, rows as instances, and relationships as patterns - not as sequences of tokens. This architectural difference is what Fraenkel and Garnelo spent 18 months getting right.
In His Own Words
"While LLMs have been great with unstructured data, they don't work well with structured data like tables. With our model Nexus, we have built the best foundation model to handle that type of data."
"We've built a generalized foundation model specifically to leverage the world's most valuable data: the billions of tables that underpin predictions in every enterprise, across every vertical."
"One model across all of your use cases delivers better performance than what you would otherwise be able to do with an army of data scientists."
"LLMs have transformed unstructured data but barely touched the 70-80% of enterprise data that lives in rows and columns."
"The difference between now and previous industrial revolutions is that we're automating cognition, not just physical labor."
"NEXUS is the OS for business decisions."
Funding Snapshot
Career Arc
Joined JPMorgan - first exposure to mission-critical enterprise data at institutional scale. Where the spreadsheet isn't a convenience tool; it's the infrastructure.
Joined Bridgewater Associates, the world's largest hedge fund and a temple of data-driven decision-making. Ray Dalio's "radical transparency" is, among other things, a philosophy about what structured data can tell you if you let it.
Co-founded first startup. Built it. Exited. Learned what it takes to go from zero to shipped.
Co-founded second startup. Exited again. Two for two. The pattern holds.
Graduate degree in Machine Learning. The technical foundation that turns commercial experience into a genuine research perspective.
Joined DeepMind. Met Marta Garnelo - the researcher who would become Fundamental's Chief Science Officer and write the whitepaper that became the company's technical foundation.
Co-founded Fundamental with Marta Garnelo and Gabriel Suissa. Raised a $30M seed. Went dark for 18 months. Built NEXUS.
Emerged from stealth. $225M Series A. $1.4B valuation. NEXUS public launch. AWS partnership announced. Seven-figure Fortune 100 contracts signed. One morning changed everything.
Spoke at HumanX conference, San Francisco - on behavioral data and enterprise AI's next frontier.
Appeared on This Week in AI with Jason Calacanis to discuss why tabular data is the unfinished business of the AI revolution.
Achievements
Built Fundamental to a $1.4B unicorn valuation at Series A - before most people had heard the company name.
Raised $255M total ($30M seed + $225M Series A) with Oak HC/FT, Salesforce Ventures, Battery, Valor, and Hetz.
Signed multiple seven-figure contracts with Fortune 100 enterprises before Fundamental's public launch.
Brokered a first-party AWS Marketplace partnership - placing NEXUS directly in the buying path of AWS enterprise customers.
Led development of NEXUS, the first major foundation model purpose-built for tabular data - no preprocessing, no feature engineering, one line of code.
Attracted investments from Perplexity CEO Aravind Srinivas, Wiz CEO Assaf Rappaport, Brex co-founder Henrique Dubugras, and Datadog CEO Olivier Pomel.
Assembled a founding team from DeepMind and Isomorphic Labs - including Chief Science Officer Marta Garnelo, whose whitepaper became NEXUS's technical foundation.
Successfully exited two prior startups before founding Fundamental - a track record that explains why serious investors trusted the thesis before the product shipped.
Who Backed It
Technology
"NEXUS was trained on billions of tabular datasets using Amazon SageMaker HyperPod infrastructure - the same kind of scale investment that distinguishes foundation models from fine-tuned experiments."
- Fundamental, Feb 2026 launch announcementDetails That Matter
Marta Garnelo, Fundamental's Chief Science Officer, literally wrote the academic whitepaper that became NEXUS's technical foundation - before the company existed as a business.
Fundamental was valued at $1.4 billion before most of the world knew its name. It operated in complete stealth for 18 months.
Fraenkel's time at Bridgewater - Ray Dalio's notoriously data-obsessive hedge fund - almost certainly shaped his conviction that structured data contains more economic value than anyone in AI was mining.
The CEO of Perplexity AI personally backed Fundamental - a notable endorsement from a founder whose entire company is built on reimagining how people extract signal from information.
Fundamental's R&D team includes engineers from AI21 Labs operating out of Israel - giving the company two centers of AI research gravity simultaneously.
Oak HC/FT described Fraenkel as possessing "financial literacy that is uncommon in the world of AI research" - a rare compliment from a firm that evaluates hundreds of technical founders annually.
Find Jeremy Online