Fundamental launches with $255M and an AI model optimized for tabular data – News

Artificial intelligence startup Fundamental Technologies Inc. launched today with $255 million in initial funding from a group of prominent investors.

The company raised the bulk of the capital through a $225 million Series A round led by Oak HC/FT. The firm was joined by Salesforce Ventures, Valor Equity Partners, Battery Ventures and Hetz Ventures. Fundamental says that its investor roster also includes the chief executives of Wiz Inc., Perplexity AI Inc., Datadog Inc. and Brex Inc.

The company’s flagship offering is an AI model called Nexus. It’s specifically optimized to process tabular data, or information organized into rows and columns.

According to News, Nexus is not based on the transformer architecture that underpins most large language models. Fundamental opted against the technology partly because it uses a mechanism called a tokenizer to process prompts. A tokenizer breaks down input data into smaller chunks to ease processing. That approach is effective for text input, but can reportedly create quality issues during tabular data analyses.

Fundamental trained Nexus on billions of tabular datasets. It carried out the training using Amazon Web Services Inc.’s Amazon SageMaker HyperPod service. The service enables developers to spin up cloud-based AI clusters with thousands of chips. It also speeds up administrative tasks such as defining cluster configuration settings and recovering from training errors.

Large tabular datasets often have to be organized into a form that lends itself better to analysis before they can be analyzed by an AI model. Nexus removes that requirement. The model thereby reduces the need for manual data preparation, which lowers the costs associated with AI projects.

One way Nexus speeds up the data preparation workflow is by automatically interpreting ambiguous records. For example, the model could determine whether spreadsheet fields that contain the values “Yellowstone” and “Yosemite” are describing national parks or conference room names. That automation removes the need for developers to manually find and remove potentially ambiguous table elements.

Companies can use Nexus to generate forecasts based on tabular data. For example, a manufacturer could upload a spreadsheet that contains equipment telemetry and ask the model to estimate when the next malfunction will occur. Nexus also lends itself to tasks such as predicting floods and store traffic.

“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,” said co-founder and Chief Executive Officer Jeremy Fraenkel (pictured, right, with co-founders Annie Lamont and Gabriel Suissa).

The AI developer says that it has already nabbed multiple seven-figure contracts with Fortune 100 companies. Those customers are using Nexus for tasks such as forecasting customer churn. Additionally, Fundamental has partnered with AWS to make it simple for enterprises deploy to its AI model in their cloud environments.

The company will use its new funding to procure more computing infrastructure and grow its workforce.

Photo: Fundamental