Key Takeaways
- San Francisco–based AI lab Fundamental has emerged from stealth with $255 million in funding (seed plus Series A), valuing the company at about $1.2 billion and giving it instant unicorn status.
- The startup is launching Nexus, a Large Tabular Model (LTM) designed to analyze massive structured datasets like enterprise tables and databases, rather than text and images.
- Fundamental has signed seven‑figure deals with Fortune 100 customers and sealed a strategic partnership with AWS, enabling Nexus to be deployed directly inside customers’ existing AWS environments.
- The team pitches Nexus as delivering inference that is efficient, affordable, and “makes GPUs optional,” by betting on purpose‑built infrastructure and deep integration with AWS’s AI stack.
What Happened?
An AI lab called Fundamental has come out of stealth with $255 million in total funding and a reported unicorn‑level valuation, unveiling its flagship model Nexus, a Large Tabular Model built for enterprise structured data. In an announcement on X via the company’s @LTMpredict account, alongside press releases and a TechCrunch feature, Fundamental also disclosed a strategic partnership with Amazon Web Services (AWS) to bring Nexus natively to AWS customers.
Emerging today from stealth; $255M in funding; and a partnership with @awscloud 🚀https://t.co/wapNkXlpxg
— Fundamental (@LTMpredict) February 5, 2026
Nexus and the AWS Tie‑Up – A Different Kind of Foundation Model
Fundamental’s bet is that the next big frontier for AI is not more text and images, but the billions of rows of tabular data that underpin enterprise operations from transactions and risk scores to supply‑chain and telemetry feeds. Nexus is positioned as a Large Tabular Model (LTM) rather than a large language model, using a non‑transformer architecture that is deterministic (the same input always yields the same output) and designed to reason over entire datasets rather than a limited context window.
The company’s $255 million haul combines a $30 million seed with a $225 million Series A led by Oak HC/FT, Valor Equity Partners, Battery Ventures and Salesforce Ventures, with participation from Hetz Ventures and prominent angels including Perplexity’s Aravind Srinivas, Brex’s Henrique Dubugras and Datadog’s Olivier Pomel. That capital will mostly fund compute, go‑to‑market and an expanded research and engineering team to scale Nexus into production environments across industries.
On the distribution side, the AWS partnership is critical. Nexus is trained and deployed on AWS infrastructure and can be provisioned directly via the AWS console, similar to buying compute or storage. For AWS, the deal fills a well‑known gap: transformer‑based models on GPUs struggle with massive tables, while AWS’s own Trainium and Inferentia accelerators are explicitly optimized to deliver better price‑performance for inference at scale. Fundamental’s public messaging “inference that’s efficient, affordable, and finally makes GPUs optional” leans into that trend of purpose‑built accelerators displacing general‑purpose GPUs for certain workloads.
Early traction appears real rather than theoretical. Fundamental reports seven‑figure contracts with Fortune 100 enterprises, where Nexus replaces brittle stacks of legacy predictive models with one generalized model for forecasting, anomaly detection and optimization across business units. The model is designed to plug into existing data stacks with minimal code, ingest raw tabular data and automatically learn patterns and dependencies without extensive feature engineering.
Why This Matters in the AI Market Now
The launch lands at a moment when enterprises are discovering the limits of general‑purpose LLMs for structured data. Transformers excel at unstructured content but are constrained by a finite context window and probabilistic outputs, making them ill‑suited for reasoning across entire data warehouses or billion‑row tables where determinism, auditability and full‑dataset coverage are mandatory.
A growing wave of “tabular foundation models” from academic work like TabPFN to commercial offerings aims to close this gap by treating tables as a first‑class modality. Analysts estimate structured‑data AI could represent a tens‑of‑billions‑of‑dollars market as enterprises look to automate forecasting, fraud detection, logistics optimization and more over their existing databases. Fundamental’s Nexus positions the company as an early category leader in this emerging segment, in the same way that frontier LLM labs defined the unstructured‑data era.
Competitive Landscape
- Fundamental – Nexus LTM (enterprise‑scale tabular foundation model).
- Prior Labs – TabPFN, a tabular foundation model spun out of academic research, focused on small‑to‑medium tables and data‑constrained settings.
- Neuralk AI – Tabular Foundation Model for Commerce, a Paris‑based startup building tabular models for retail and catalogue-centric use cases.
| Feature / Metric | Fundamental – Nexus (LTM) | Prior Labs – TabPFN | Neuralk AI – Tabular FM for Commerce |
| Context Window | Not token‑based; designed to handle enterprise‑scale tables, including billions of rows, without a fixed context window limit. | Optimized for small‑to‑medium datasets (up to ~10k samples) with fast, single‑pass inference. | Focused on commerce catalogs and product data, tuned for multi‑table, schema‑messy retail datasets rather than petabyte‑scale warehouses. |
| Pricing per 1M Tokens | Enterprise contracts; usage‑based / custom pricing, not exposed as per‑token because workloads are table‑ and use‑case‑driven. | Early‑stage API and licensing; research‑driven, with no public per‑token pricing and more experimentation‑oriented deployments. | SaaS/API for retailers with contract‑based pricing tied to catalog size and traffic; no standard per‑token metric. |
| Multimodal Support | Focuses on structured/tabular data; can ingest numeric and categorical columns, with limited emphasis on images or rich media. | Primarily tabular‑only, but recent work adds support for text features and contextual metadata in tables. | Built for tabular + text‑heavy product fields (titles, descriptions, attributes) within commerce datasets. |
| Agentic Capabilities | Exposed as a model primitive; no native “agent” layer but integrates into orchestration stacks (e.g., AWS workflows, BI tools) for decision automation. | Focused on core predictions; agents and workflows are typically built externally around the API or library. | Positioned as a vertical solution for retail operations, often embedded into recommendation, search and catalog‑cleaning pipelines managed by higher‑level agents. |
Nexus clearly wins on raw scale and deterministic enterprise‑grade behavior, targeting Fortune‑level datasets and regulated industries where full‑table reasoning matters most. Prior Labs and Neuralk AI, by contrast, are stronger fits for narrower or data‑constrained domains, such as startups, commerce platforms and teams that need powerful tabular modelling on smaller datasets or vertical‑specific workflows without building a massive data‑infrastructure stack.
Bayelsa Watch’s Takeaway
In my experience, this kind of fundraise and distribution story is a strong bullish signal for a new AI category, not just a single company. I think this is a big deal because it shifts attention and serious capital toward the structured data that actually drives P&Ls, not just the text and images that dominate AI demos. By pairing a purpose‑built tabular model with AWS’s maturing accelerator stack, Fundamental is effectively betting that enterprises want predictable, explainable, GPU‑light inference wired directly into their existing data lakes.
If the team can keep converting Fortune‑100 pilots into long‑term contracts, Nexus could quietly become as central to enterprise decision‑making as today’s headline LLMs are to chatbots, only this time, the value will show up directly in forecasts, risk models and revenue optimisation rather than in viral screenshots.
Sources
