Dead Data Is Killing AI Projects. Only 4.4% of Enterprise Storage Is Actually in Use

New research scanned 16 petabytes of live enterprise file data. Almost none of it is doing any work — and that is why so many AI programs stall after the pilot.

September 17, 2026 · 9 min read

Most enterprises do not have a model problem. They have a graveyard problem.

On September 15, 2026, CTERA published an analysis of 16 petabytes of live production data across 856 enterprise NAS file-share scans, including regulated industries. The finding is blunt: only 4.4% of stored capacity is actively used. The other 95.6% sits on primary storage as a de facto archive — backed up, protected, permissioned, and paid for as if it were still part of the business.

That unused mass has a name operators already know: ROT data — redundant, obsolete, or trivial. It raises cost. It widens the blast radius of a breach. And it is now quietly poisoning AI readiness, because models and agents retrieve whatever they can reach.

This is not a storage-vendor story. It is a data lifecycle story. Most vendors stop at modernization. The value — and the risk — lives in what happens after go-live: optimize, secure, then retire and recover.

What the 4.4% number actually measured

CTERA did not survey opinions. It measured activity on real file estates.

Read that last row twice. In a typical quarter, nine-tenths of the estate is untouched. Yet it still lives on the most expensive tier, inside the same shares, under the same permissions, inside the same backup jobs as the 4.4% the business actually uses.

“When only 4.4% of capacity is actively used, enterprises are running their most expensive storage tier as what is effectively a cold archive. That creates a multiplier effect across storage and data protection costs, security exposure, and now the quality of information available to AI systems.”

— Aron Brand, CTO, CTERA

Brand’s next line is the one most “just delete it” posts skip: the answer is not mass deletion. It is deciding where data belongs — retain, archive, sanitize, or retire — without leaving a cold archive on hot infrastructure.

Three ways dead data taxes the business at once

1. Cost: you pay for it three times

Inactive files are not a one-line storage bill. They are copied into snapshots, replicas, disaster-recovery targets, and cloud backups. Every extra terabyte on primary storage multiplies across the protection stack.

That pattern shows up off the file share too. Flexera’s 2026 State of the Cloud Report put estimated wasted IaaS/PaaS spend back at 29% — the first rise in five years — as AI and new platform services added cost complexity. Cold file data and idle cloud capacity are the same failure wearing different labels: no one owns the lifecycle after the asset is provisioned.

This is also where EcoGreen’s own optimization work keeps landing. In one enterprise database rebuild documented in our case studies, storage dropped from 17TB to 2TB (88%), runtime fell from 95 minutes to 2 minutes, and compute and memory use were cut in half. That was not a “buy more cloud” project. It was an optimize-the-estate project. Dead weight came out. The platform got faster because there was less junk in the way.

2. Security: the blast radius is the whole share

CTERA’s research is explicit on exposure. Inactive data often remains in the same network-accessible environments, under the same permissions, as active data. A ransomware event, a misconfigured share, or an over-privileged identity does not politely skip the 88% of files nobody has opened in a year.

For boards and insurers, that is no longer an IT hygiene issue. It is a resilience issue. You cannot claim a tight security posture while the majority of your unstructured estate is ungoverned, unreviewed, and still online. Cybersecurity and resilience work — DR, failover, compliance-grade controls — only holds if the data in scope is known, classified, and reduced to what the business still needs.

3. AI readiness: garbage in, at machine speed

Gartner has warned that through 2026, organizations will abandon 60% of AI projects that are not supported by AI-ready data. Separate analyses keep repeating the same post-mortem: poor data quality, weak governance, rising cost of ownership — not “the model was bad.”

Now put CTERA next to that prediction. If an assistant or an agent can search the file estate, it will retrieve stale contracts, superseded specs, duplicate customer extracts, and abandoned project folders with the same confidence it retrieves last week’s source of truth. Relevance drops. Hallucinations rise. Compliance teams lose the ability to explain why a model cited a file that should have been retired two years ago.

AI-ready is not “we bought a lakehouse.” AI-ready is data engineering plus lifecycle governance: current data on the hot path, retained data in the right tier, retired data sanitized and off the network.

Why “lift and shift the whole share” makes this worse

The default modernization move is still to copy the mess into a new cloud. That is how a 17-terabyte problem becomes a 17-terabyte cloud bill with a better logo on the invoice.

EcoGreen’s position on this is the same one on the homepage: from legacy systems to cloud-native, to sustainable retirement. Cloud modernization is stage two of five, not the finish line.

1. Assess — audit infrastructure, the data estate, and compliance exposure. Find what is live, what is ROT, and what is legally retained.
2. Modernize — move what still earns its keep onto Azure, AWS, OCI, or Fabric. Do not migrate the graveyard.
3. Optimize — cut storage, runtime, and spend on the live estate. This is where 88% storage reductions come from.
4. Secure — shrink permissions, prove DR, keep chain-of-custody on anything that still holds regulated data.
5. Retire & Recover — certified disposal and ESG-aligned asset recovery. Value out, risk off the floor. EcoGreen has recovered $1.2M+ in asset value to date on this last mile.

Most firms pay four vendors to cover those five steps and then wonder why nobody owns the ROT. One accountable partner for the complete lifecycle is not a slogan. It is how you stop paying for the same dead file in production, in backup, and in the AI index.

A 30-day pass over your own 4.4%

You do not need a new platform to start. You need a decision framework and an owner.

Week 1 — Measure activity, not capacity. Age and last-access reports on the largest file shares and object buckets. Separate “not modified in 12 months” from “not accessed in 90 days.” Those are different piles. One may be a legal hold. The other is usually waste.

Week 2 — Classify before you touch a disk. Four buckets only: active, retain-in-place / archive, sanitize and retire, destroy. If a file cannot be placed in one of those four, it stays in assess. Do not invent a fifth bucket called “we’ll deal with it after the AI pilot.”

Week 3 — Take dead data off the hot path. Archive what must be kept. Remove orphaned shares. Right-size the databases and warehouses that grew because nobody deleted the staging copies. This is the same discipline behind EcoGreen’s 17TB → 2TB result — less data in the working set, not a faster disk under the same junk.

Week 4 — Close the last mile. Hardware and file systems that are truly done leave with chain-of-custody, certified sanitization, and an ESG record. That is asset recovery, not a loading-dock afterthought. Residual value funds the next modernization cycle instead of disappearing into a dumpster and an untracked risk.

If you run manufacturing, finance, telecom, or government estates — EcoGreen’s core industries — the same pattern holds at petabyte scale. We already operate in 5PB+ ecosystems with hundreds of pipelines. The constraint is rarely Spark or Kafka. The constraint is trust: which data is allowed to be current.

What “good” looks like when the AI program starts asking for data

When a line-of-business leader wants an agent on internal files next quarter, the honest questions are operational:

• Which shares are in the retrieval path, and who approved them?
• What percentage of that path has been touched in 90 days?
• Where does legally retained data live so it is available without sitting on primary NAS?
• What is the offboarding path when a project, an app, or an agent is retired — data and identity included?
• Can you produce certificates and inventory if an auditor or an insurer asks what left the building?

Those questions sit across data lifecycle management, cybersecurity, and asset recovery. They are not answered by a Copilot license.

Proof, not promises: EcoGreen has moved 300+ databases with 20+ zero-downtime VLDB cutovers, deployed 60+ Oracle RAC clusters and 40+ Data Guard setups, and trained 1,000+ IT professionals. The through-line is the same 5-stage loop. Assess what you have. Modernize what still matters. Optimize the working set. Secure the remainder. Retire the rest and recover the value.

The board version of this article

If you only take one paragraph upstairs, take this one.

You are financing a cold archive on hot infrastructure. Independent measurement now puts active use at 4.4% of enterprise file capacity. That idle mass inflates storage and protection cost, expands cyber exposure, and feeds AI systems stale or trivial content. Deleting everything is the wrong move. Leaving everything online is the expensive one. The competent move is a governed lifecycle — with one owner from first assessment to certified retirement.

Sources

• CTERA, “New CTERA Research Reveals Just 4.4% of Enterprise Storage Is Actively Used,” September 15, 2026; and The Cold Data Storage Report.
• Flexera, 2026 State of the Cloud Report — estimated wasted IaaS/PaaS spend at 29%.
• Gartner — organizations will abandon 60% of AI projects unsupported by AI-ready data through 2026.
• EcoGreen Solutions proof points: homepage, services, case studies.

EcoGreen Solutions is a data lifecycle strategy firm based in Wheaton, IL. Six capabilities, one accountable partner — cloud modernization, data engineering, lifecycle governance, asset recovery and ESG, cybersecurity and resilience, and capability building.

Follow and connect

• Website: theecogreensolutions.com
• Articles: theecogreensolutions.com/articles
• Insights: theecogreensolutions.com/insights
• X: @ecgsol
• LinkedIn: Muhammad Adeel
• Email: info@theecogreensolutions.com