Box integrates Nvidia’s Nemotron for faster content retrieval system

Box wants AI agents to stop rummaging through your company’s files like a raccoon in a dumpster. At its developer conference on Thursday, the cloud content company unveiled a feature that builds Nvidia’s Nemotron embedding system directly into the heart of its product.

The feature is called Box Content Retrieval. Its pitch is simple: AI queries that are faster, more accurate and less expensive.

What Box actually announced

The launch took place at the BoxWorks 2026 Developer Summit on November 5, 2026. Box CEO Aaron Levie and Nvidia CEO Jensen Huang presented the feature during the opening keynote.

The engine underneath is Nvidia’s Nemotron family. Specifically, Box is tapping the Nemotron 3.5 Retriever models, which Nvidia released in October 2026.

Those models come in three flavors: Embed, Rerank and Parse. Each handles a different step in the job of finding the right information inside a pile of documents.

Box Content Retrieval uses Nemotron to generate metadata that gets embedded alongside the content stored in Box. AI agents then lean on that metadata to answer questions, instead of brute-forcing their way through raw files every time someone asks something.

Why embeddings matter for enterprise AI

The three Nemotron components split up the labor. Parse helps pull usable content out of documents, Embed turns that content into searchable representations, and Rerank sorts the results so the most relevant ones rise to the top.

The target here is unstructured data: the contracts, slide decks, PDFs and spreadsheets that make up the bulk of what companies store.

The security angle Box keeps emphasizing

Box is not just selling speed. The company is leaning hard on governance, positioning itself as a secure foundation for AI applications.

The announcement stresses that compliance, permissions and audit controls stay intact even as AI workflows touch the content. In practice, that means an agent should not be able to show an intern the CFO’s compensation files just because the question was phrased cleverly.

By baking retrieval into its own platform, Box can apply its existing permission rules to what agents see. The data never has to leave the governed environment to become searchable.

What this means for Box, Nvidia and enterprise buyers

For Box, the strategy is clear. It wants to be the layer where enterprise content and AI agents meet, rather than a storage bucket that other AI tools raid for data.

Box has not, at least in the announcement, put a hard number on those savings. Buyers will want to see benchmarks from real deployments before rewriting their AI budgets.

For Nvidia, the deal is another example of its push beyond selling chips. The company has been building out software and model families like Nemotron, and getting them embedded in a widely used enterprise platform gives those models a direct line to corporate data.