Ask most boardrooms why their AI program is moving slowly and the common answers include unproven models, expensive infrastructure, or scarce ML talent.
But the real answer is usually disorganized data, sitting in storage.
In its August 2025 research note, Governing Unstructured Data for AI Readiness: A Strategic Roadmap, Gartner® analyst Melody Chien puts a number on a problem many data and security leaders have felt for a while. Gartner has seen approximately a 150% increase in client inquiries about unstructured data management over the past twelve months.
Indeed, the absence of GenAI-ready data is now the top reason GenAI deployments fail. Up to 90% of enterprise data is unstructured and much of it is not yet AI-ready. So getting that data organized and ready for AI use should be a top priority for any organization looking to successfully launch any enterprise-wide AI initiative.
The Inert Data Problem
Most enterprises are living with an unstructured data paradox. Their unstructured data (including PDFs, contracts, support tickets, slide decks, voice recordings, images, and videos) holds the richest context about customers, products, and operations.
But it is also the least visible data asset a company owns.
Gartner estimates this data is growing 40% to 60% a year. Additionally, more than half of this data sits in what Chien describes as an “inert state,” present in storage but not providing value to AI initiatives. It is the dark matter of the modern enterprise: heavy, pervasive, and almost entirely unobserved.
You cannot classify what you cannot see, you cannot protect what you have not classified, and you cannot safely feed it to a large language model without risking a compliance event you never saw coming. So how do you activate this massive store of valuable data?
What Gartner Recommends
Chien lays out a practical five-step roadmap that can activate inert unstructured data to support enterprise AI initiatives. These include:
Discovering and cataloging unstructured data across the estate
Preprocessing and analyzing it to make it usable
Tagging and classifying by sensitivity, topic, and domain
Integrating unstructured data into structured data and AI pipelines
Enforcing data policy along the way
For a full account of this five-step process, you should read the report. It’s really worth your time.
What stayed with me are the assumptions Gartner attaches to these five steps. By 2027, 60% of data governance teams will be mandated to prioritize governance of semistructured and unstructured data for GenAI use cases. And through 2028, teams that try to build their own unstructured metadata solutions are projected to spend more than 300% as much as teams that use existing document and records solutions.
What does this mean? Notably, good governance of unstructured data is now a mandate, not a side project, and a do-it-yourself unstructured data strategy is the expensive way to get up to speed.
The Rubrik Point of View
Fortunately, the roadmap to good unstructured data governance is not the hard part. But operationalizing it across petabytes, without copying the entire estate or standing up a parallel pipeline, is.
You cannot make unstructured data AI-ready unless you can first see, protect, and recover it. That is what Gartner calls as risk mitigation and it is where Rubrik starts. Rubrik for Unstructured Data protects petabyte-scale unstructured data across NAS and object storage with immutable, air-gapped backups, and the same scanning and indexing engine that discovers files at scale also surfaces sensitive data exposure. The data feeding your AI becomes data you can see, secure, and restore.
It also speaks to the value-generation pillar Gartner calls cost optimization: by centralizing management and applying retention and lifecycle policies, Rubrik for Unstructured Data keeps 40% to 60% annual data growth from turning into runaway storage cost, the retention-and-storage lever Chien ties directly to controlling spend.
Making that data usable for AI without copying it twice is the next step. Rubrik Annapurna, the AI-ready unstructured data layer we unveiled at Rubrik Forward, scans and catalogs unstructured data in place and publishes a queryable catalog into the lakehouse. This way, data teams can put dark files to work without the duplication and ETL overhead that has historically kept unstructured data out of AI pipelines. It runs on Rubrik Security Cloud, alongside your existing storage and lakehouse, with no new infrastructure or agents to deploy. It is available today for qualified enterprise partners.
Annapurna can also help with another of Chien’s recommendations, linking unstructured data to structured context via active metadata and knowledge graphs to feed RAG pipelines. Annapurna's queryable catalog lands unstructured file metadata in the same governed lakehouse as your structured data, ready to be related and retrieved. Because it delivers only what a query pulls rather than copying the whole estate, it aligns infrastructure costs with consumption.
Read the Report
If you have any influence over data, security, or AI strategy, you should read Chien's roadmap in full. It is the most concise framing I have seen of where unstructured data governance needs to go. Then take a hard look at your own inert data: how much of it you can see, how much is sensitive or regulated, and how much is actually feeding your AI initiatives?
Unstructured data is the largest, fastest-growing, and least-governed asset most organizations own. The companies that make it visible, protected, and AI-ready first will be the ones that ship AI first. Gartner has published the roadmap. The rest is execution.
Read the complimentary Gartner report
See Rubrik for Unstructured Data