Quick Brief
- Fundamental raised $255 million Series A at $1.2 billion valuation in February 2026
- NEXUS Large Tabular Model outperforms traditional ML on structured enterprise data
- Built by DeepMind alumni, trained on billions of tables using AWS SageMaker HyperPod
- Fortune 100 clients deploy NEXUS with single line of code integration
- Strategic AWS partnership enables enterprise-scale deployment across cloud infrastructure
Fundamental Technologies just shattered the enterprise AI status quo and the numbers prove it. While the industry obsesses over LLMs, this DeepMind-founded startup emerged from stealth with a $255 million war chest and a radically different approach: Large Tabular Models that actually understand the structured data running every business decision. NEXUS, their flagship foundation model, replaces decades-old machine learning algorithms with deep learning purpose-built for the billions of tables sitting untapped in corporate databases.
What Makes Large Tabular Models Different From LLMs
LLMs dominate unstructured data text, images, video but fail spectacularly on enterprise tables. Transformer architectures optimize for sequential patterns, not the non-linear relationships across rows and columns that define business data. NEXUS abandons transformers entirely, using a novel architecture pre-trained on billions of tabular datasets to capture dependencies invisible to classic ML.
Most enterprise data lives in structured tables, not documents. Sales forecasts, inventory systems, financial records, customer databases are all tabular. Traditional ML algorithms predate deep learning breakthroughs and require extensive feature engineering for each use case. Fundamental’s CEO Jeremy Fraenkel calls NEXUS “the OS for business decisions,” enabling one model across unlimited use cases with superior performance.
What is a Large Tabular Model?
A Large Tabular Model is a foundation model pre-trained on vast tabular datasets to understand non-sequential, non-linear relationships in structured data. Unlike LLMs optimized for text or images, LTMs process enterprise-scale tables with billions of rows without dimensionality constraints. Research from leading institutions defines LTMs as transformative for few-shot synthetic data generation, imputation, and cross-dataset contextualization.
The $255 Million Validation: Why Investors Bet Big
Fundamental’s February 2026 emergence marks the largest Series A in tabular AI history. Oak HC/FT, Valor Equity Partners, Battery Ventures, and Salesforce Ventures co-led the $225 million round, with Hetz Ventures and notable angels including Perplexity CEO Aravind Srinivas, Brex co-founder Henrique Dubugras, and Datadog CEO Olivier Pomel. Combined with a $30 million seed round, Fundamental raised $255 million total at a $1.2 billion valuation just 16 months after founding in October 2024.
The investor appetite reflects enterprise urgency. Fortune 100 companies already signed seven-figure contracts before public launch, validating demand for tabular-specific AI. AWS formalized a strategic partnership enabling direct NEXUS deployment from existing instances. Fraenkel told TechCrunch that NEXUS allows enterprises to “expand massively the number of use cases” while achieving “better performance than an army of data scientists”.
How NEXUS Delivers Superior Predictions
NEXUS integrates into existing data stacks with minimal friction, often a single line of code. Once connected, the model ingests raw tabular data and autonomously learns underlying structure, patterns, and dependencies without manual training or feature engineering. This contrasts sharply with traditional ML workflows requiring data science teams to craft features for each new prediction task.
Built on AWS SageMaker HyperPod, NEXUS comprehends non-linear interactions across entire table structures. The model was trained from the ground up on billions of tabular datasets, giving it generalized knowledge of patterns that drive outcomes across industries. Fundamental’s whitepaper claims NEXUS achieves higher win rates than classic ML benchmarks, though specific metrics remain proprietary.
How does NEXUS compare to traditional machine learning?
NEXUS outperforms legacy ML algorithms through foundation model pre-training on billions of tables. Traditional methods require manual feature engineering and separate training for each use case, while NEXUS applies one model across unlimited tasks with superior accuracy. Research confirms LLMs struggle with tabular data’s non-sequential structure, validating purpose-built LTM architectures.
Enterprise Deployment: AWS Integration and Privacy
Fundamental prioritized enterprise-grade requirements from day one. The AWS strategic partnership enables customers to deploy NEXUS directly within their cloud infrastructure, maintaining data sovereignty and security controls. Companies can integrate NEXUS via deep AWS connections without moving sensitive data to external platforms.
The deployment model addresses the core enterprise AI dilemma: leveraging powerful models without compromising proprietary data. NEXUS runs within customer environments, processing tables in place rather than requiring data extraction. Fast time-to-value matters Fundamental enables forward-looking predictions on when risks emerge or where opportunities exist, moving beyond retrospective analysis.
Use Cases: From Finance to Healthcare
Tabular data defines decision-making across every industry. Financial institutions use tables for risk modeling, fraud detection, and portfolio optimization. Healthcare organizations rely on structured patient data for treatment outcomes and resource allocation. Retail and e-commerce track inventory, pricing, and customer behavior through database tables.
Fundamental’s LTM approach unlocks use cases previously limited by ML constraints. Few-shot learning capabilities matter for domains with smaller datasets, a common tabular scenario. Synthetic data generation for privacy, debiasing, and augmentation becomes feasible at scale. Cross-dataset contextualization enables scientific discovery by relating findings across studies.
Supply chain optimization benefits from NEXUS predictions on demand fluctuations, logistics bottlenecks, and supplier risks. Sales forecasting improves when models understand complex interactions between pricing, seasonality, competition, and customer segments. Predictive maintenance schedules equipment interventions before failures occur, analyzing sensor data patterns.
The DeepMind DNA: Who Built Fundamental
Fundamental’s founding team brings DeepMind pedigree and global systems experience. CEO Jeremy Fraenkel describes NEXUS as fulfilling the promise of deep learning for the world’s most valuable data format. The company’s background spans Silicon Valley academia, global financial systems, and cutting-edge AI research.
Founded in October 2024, Fundamental moved from stealth to public launch in just 16 months with Fortune 100 clients already deploying NEXUS. The team’s research foundations at world-leading universities inform the LTM architecture. Experience building mission-critical data platforms and cloud-scale systems shaped enterprise deployment priorities.
Technical Architecture: Beyond Transformers
NEXUS breaks from contemporary AI practices by rejecting transformer architecture entirely. Transformers excel at sequential data patterns but fail on tabular structures lacking inherent order. Fundamental developed a novel neural architecture specifically for non-sequential, high-dimensional tables.
The model processes enterprise-scale datasets with billions of rows dimensionality that overwhelms LLMs. Pre-training on diverse tabular datasets enables transfer learning across industries and use cases. Research into integrated Large Tabular Models suggests combining gradient-boosted decision trees with neural networks offers optimal tabular performance.
Can LLMs analyze tabular data effectively?
LLMs perform poorly on structured tabular data due to transformer optimization for sequential, unstructured inputs. Studies show traditional ML methods outperform LLMs on regression and classification tasks with structured features. LLMs only excel when tabular data contains text-heavy columns like reviews or sentiment data.
Market Opportunity: Trillions in Untapped Value
Fundamental targets trillions in enterprise value locked in underutilized tables. Most organizations analyze past events rather than predicting future outcomes due to ML limitations. NEXUS shifts enterprises from retrospective reporting to forward-looking intelligence on risks, opportunities, and optimal decisions.
The tabular AI market lags years behind language and vision domains. Foundation models revolutionized text and images, but structured data still relies on pre-deep-learning algorithms. Academic research identifies tabular foundation models as a critical priority for unlocking scientific discovery and business value.
Quantzig analysis highlights competitive advantages from tabular AI: faster decision-making, enhanced visualization, predictive precision, and near-real-time business-ready data access. Organizations implementing tabular AI report more accurate ML predictions and streamlined operations.
Competitive Landscape and Industry Response
Fundamental faces minimal direct LTM competition given the nascent category. Academic institutions like Vanderschaar Lab advocate for tabular foundation model research priorities. Recent papers including LaTable and iLTM explore LTM architectures combining decision trees, neural networks, and retrieval mechanisms.
Traditional ML platforms and data science tools represent indirect competition. Companies like CubeDrive offer enterprise tabular AI solutions, but lack foundation model approaches. Fundamental’s value proposition centers on replacing multiple specialized models with one generalized LTM across unlimited use cases.
The AWS partnership provides distribution advantages over competitors. Existing cloud customers can deploy NEXUS without infrastructure changes, lowering adoption friction. Seven-figure Fortune 100 contracts before public launch signal strong product-market fit.
Implementation Requirements and Integration
Organizations integrate NEXUS with minimal technical overhead. The single-line code deployment claim contrasts with traditional ML requiring extensive data preparation pipelines. NEXUS handles raw tabular inputs, eliminating manual feature engineering.
Cloud infrastructure flexibility matters for enterprise adoption. NEXUS deploys on any cloud platform while maintaining enterprise-grade security and privacy. The AWS integration offers streamlined onboarding for existing AWS customers.
Data scientists can expand prediction coverage without proportional resource increases. Fraenkel emphasizes that one model tackles multiple use cases with better performance than specialist approaches. This efficiency transforms AI economics for data-heavy organizations.
Future Implications: The LTM Era Begins
Fundamental’s emergence signals a paradigm shift in enterprise AI strategy. Organizations may consolidate fragmented ML initiatives under unified tabular foundation models. The LTM category could parallel LLM growth trajectories as enterprises recognize structured data value.
Academic research predicts LTMs will revolutionize science by contextualizing datasets across studies. Few-shot capabilities enable smaller organizations to leverage AI without massive data requirements. Synthetic data generation addresses privacy concerns while expanding training possibilities.
Cross-industry applications suggest LTMs become infrastructure layers rather than niche tools. Every enterprise runs on tables financial models, customer databases, operational metrics, supply chains. Foundation models that understand this universal data format unlock transformative potential.
Limitations and Considerations
NEXUS performance metrics remain largely proprietary beyond general win rate claims. Independent benchmarks comparing NEXUS to state-of-the-art tabular ML methods would strengthen validation. Enterprises should conduct proof-of-concept testing on their specific use cases before full deployment.
Foundation models require substantial computational resources during training. While inference may be efficient, organizations must consider ongoing operational costs. The single-line integration claim likely applies to standard scenarios complex data pipelines may require additional configuration.
LTM research acknowledges that tabular foundation models lag behind language and vision domains in maturity. Fundamental leads commercial deployment, but the technology category remains young with evolving best practices. Organizations adopting early should expect iterative improvements as LTM capabilities advance.
Frequently Asked Questions (FAQs)
What is Fundamental’s NEXUS?
NEXUS is Fundamental’s Large Tabular Model, a foundation AI model pre-trained on billions of tabular datasets to predict outcomes from enterprise structured data. Unlike LLMs designed for text, NEXUS processes non-sequential table relationships with single-line code integration.
How much funding did Fundamental raise?
Fundamental raised $255 million total: $30 million seed plus $225 million Series A led by Oak HC/FT, Valor Equity Partners, Battery Ventures, and Salesforce Ventures at a $1.2 billion valuation in February 2026.
What makes Large Tabular Models different from LLMs?
LTMs specialize in structured tabular data with non-sequential, non-linear relationships across rows and columns, while LLMs optimize for sequential unstructured data like text and images. LTMs handle enterprise-scale tables with billions of rows that overwhelm LLM architectures.
Who founded Fundamental Technologies?
DeepMind alumni founded Fundamental in October 2024, led by CEO Jeremy Fraenkel. The team combines Silicon Valley academia, global financial systems experience, and cutting-edge AI research backgrounds.
Does NEXUS work with AWS?
Yes, Fundamental has a strategic AWS partnership enabling direct NEXUS deployment from existing AWS instances. The model integrates with AWS SageMaker HyperPod for enterprise-grade security and scalability.
What industries can use NEXUS?
NEXUS applies across all industries using tabular data: finance for risk modeling, healthcare for patient outcomes, retail for inventory optimization, and manufacturing for predictive maintenance. The model works with any structured business data.
How does NEXUS compare to traditional ML performance?
Fundamental claims NEXUS outperforms legacy machine learning algorithms through foundation model pre-training on billions of tables, eliminating manual feature engineering while improving accuracy. Specific win rate metrics remain proprietary.
When was Fundamental publicly launched?
Fundamental emerged from stealth in February 2026 after 16 months of development since founding in October 2024. The company announced its $255 million funding and NEXUS launch simultaneously.
