How a 17-year-old file storage company is reinventing itself for the GenAI era-and why 95% of enterprise AI pilots are failing
If you’ve been around the data center industry long enough, you’ve seen plenty of storage vendors come and go. You’ve also seen the smart ones adapt. CTERA Networks, a company that’s been quietly building hybrid cloud file storage solutions for 17 years, showed up at the 64th edition of The IT Press Tour with something that caught my attention, not because they’re chasing the AI hype train, but because they’re actually solving real problems that are killing enterprise AI projects.
Let me back up. CTERA isn’t a household name like NetApp or Pure Storage, but they’ve carved out a solid position in the distributed enterprise file storage market. They’re the company with 185 employees, growing at 35% year-over-year, with a 125% net retention rate that would make any SaaS CFO jealous. They’ve raised $180 million, most recently an $80 million round led by Red Dot Capital Partners. And here’s the kicker: 66% of their revenue comes from North America, which is why CTO Aron Brand recently relocated to New York.
But what brought me to attention during this briefing wasn’t the metrics. It was the framing of their journey as “three waves of innovation” – and more importantly, what they’re saying about why AI projects are crashing and burning across the enterprise.
The Brutal Reality of Enterprise AI
Before we get into CTERA’s strategy, let’s talk about the elephant in the room. Oded Nagel, CTERA CEO, shared a survey his marketing team conducted with 300 IT and security professionals at companies with 2,000+ employees. The top priorities for 2025? Cybersecurity (80%), Strategic AI (57%), and Data Growth Management (54%).
Nothing shocking there. But then Aron Brand dropped this stat: 95% of GenAI pilots fail.
Let that sink in. Despite Morgan Stanley projecting that GenAI enterprise software spend will hit $401 billion by 2028 – representing 22% of total global software spend – almost every proof of concept is flopping.
Why? As Aron put it during the briefing, there’s a “naive approach” happening across the industry: “In their imagination, they’re saying, we’ll just point our AI tools at all our data. We’ll vectorize everything. We’ll add a RAG layer. We’ll plug in GPT-5, the smartest model that we can find. And then we sit back and watch the magic happen.”
The problem, he explained, is garbage in, garbage out. “When you feed this low quality data into GenAI, you don’t get better data. You get very confident errors, right? GenAI is very, very confident that what it’s saying is correct, and it’s very easy to believe it.”
That’s not a storage problem. That’s a data management problem. And CTERA believes they’ve been building toward the solution for nearly two decades.
Wave 1: Location Intelligence-The Foundation Nobody Sees
CTERA’s first wave of innovation, which they’ve been delivering for years, is what they call “location intelligence.” In practical terms, it’s a global namespace that unifies file silos across edge, core, and cloud environments.
Here’s why this matters: by 2028, Gartner predicts large enterprises will triple their unstructured data capacity across on-premises, edge, and public cloud locations. That’s a management nightmare. You’ve got NFS exports, SMB shares, S3 buckets, OneDrive, SharePoint-data everywhere, in different formats, with different access protocols.
CTERA’s Global Namespace creates a single, unified view across all of it. As Saimon Michelson, VP of Alliances, explained: “We see a connection point between what we have been addressing all those years and where this market is going.”
The architecture is straightforward: CTERA deploys edge filers at remote sites that cache frequently accessed data locally for performance, while everything else lives in object storage – either on-premises S3-compatible targets or public cloud (AWS S3, Azure Blob). Multi-site sync, built-in data protection, rapid DR. It’s the hybrid cloud file storage playbook.
Real-World Example: Naval Fleet Data Management
One of the more interesting use cases Saimon shared involved a naval customer managing data across a fleet of ships. Previously, when an analyst in California needed payload data from a vessel for weapon system tuning, they’d wait for the ship to dock, then send the hard drives via helicopter.
“Not reliable, and cannot really fit in today’s world,” Saimon noted. “In today’s world, we have to respond much quicker to these threats.”
CTERA deployed edge systems on each ship that automatically sync data over satellite links – handling high latency and low bandwidth through deduplication and optimization. The result? Analysts get data in near real-time instead of waiting 48 hours. The business outcome isn’t just efficiency; it’s faster recommendations for tuning weapon systems. When you’re talking about military applications, minutes matter.
This use case also illustrates CTERA’s target market. As Saimon put it: “We find the best product-market fit” in industries that share three characteristics – lots of data, highly distributed operations, and intense security requirements. Think healthcare, defense, government, retail, manufacturing, financial services.
Wave 2: Metadata Intelligence-When Storage Becomes Security
The second wave addresses a problem that’s kept storage admins up at night for the past few years: ransomware.
CTERA launched Ransom Protect in April 2024, which uses AI-powered real-time anomaly detection to spot ransomware attacks as they happen. But here’s what differentiates it from the backup-and-pray approach: they’re running the detection engine on the file system itself, in real time.
Oded explained the impact: “We tested this in a third-party lab with various ransomware families. Our AI engine identified the attack and encryption within 10 to 50 seconds, which limits the attack to just a few files instead of the entire file system.”
Compare that to traditional recovery methods. “You know how long it can take you to recover from an attack of 1 million files, right?” Oded said. “It can take weeks, even months, just to restore the data.”
But fast detection is only half the equation. Aron Brand highlighted the exfiltration problem: “That’s currently the most concerning thing in ransomware. Once your data is leaked, there’s no recovery. The data is out there. And even if you pay, you have no assurance that they won’t leak the data anyway.”
The fines and compliance nightmares from data exfiltration are brutal. Having detailed forensics about what was accessed, by whom, and when becomes critical for incident response and regulatory disclosure.
CTERA Insight: The Air-Gapped Audit Trail
That’s where CTERA Insight comes in. Launched in November 2024, Insight is a SaaS-based (or privately deployable) analytics platform that captures every file read, write, delete, and rename across the entire global namespace-and stores it for at least a year.
The architecture is clever. Insight runs as an air-gapped system, separate from the production file storage. “We created an air gap system where all this metadata is stored securely within a Big Data database with cloud-based analytics,” Aron explained. “We keep the data for at least a year, so you have every file read, every file write, every delete, every rename, everything that happened.”
No production data gets sent to Insight-only metadata. That’s important for customers concerned about data sovereignty and compliance. And because it’s built on elastic cloud infrastructure (AWS), it can scale to handle customers with 100 billion files under management.
During the demo, Saimon showed how you can drill down into specific user activity, filter by time and location, and see exactly what happened during a ransomware attack simulation. For companies dealing with auditors or insurance companies after an incident, having this level of forensic detail changes the conversation.
“Being able to be very accurate and specific about what was impacted is super important,” Saimon noted. “It’s loss of confidence, consumer confidence, and at the end of the day, loss of business.”
Wave 3: Enterprise Intelligence-Where It Gets Interesting
Here’s where CTERA is making a big bet. Wave 3 is CTERA Data Intelligence, which Aron Brand described as “turning your data into an asset by actually peering into the content.”
This isn’t just indexing metadata. This is full content analysis, curation, and making enterprise data AI-ready. And they’re going all-in on Model Context Protocol (MCP), the open standard that anthropic and others are pushing as “the new thumb drive for the Internet.”
The Three Quality Killers
Aron laid out what he calls “the three quality killers” that doom enterprise AI projects:
1. Messy Data
“Organizations have messy data that is mixed, data that is not organized, not classified,” Aron said. “When you feed this data to GenAI, it tends to be very confused.”
The solution? AI-assisted data classification and metadata enrichment. Not just tagging files by type, but extracting contextual information – who signed this contract, what’s the contract value, what are the key terms.
2. Data Silos
“The data is not only dispersed between different locations, it’s also between different storage systems,” Aron explained. “You may have some data in OneDrive, something in SharePoint, some data in NetApp. It’s also dispersed in different file formats-PDF files, Office files, recordings, transcripts.”
The answer is an unstructured data lake that ingests from all these sources and converts everything into a unified format (typically Markdown) that GenAI can process. This includes OCR for images, transcription for audio/video, and text extraction from complex documents.
3. Compliance & Security
“These lawyers are rightly concerned about risky data lakes feeding or training AI models on things that might contain PII, health information, or sensitive financial reports,” Aron said.
CTERA’s approach combines data guardrails (redacting or dropping sensitive content) with permission-aware access. “If the files have permissions or ACLs, and the AI system is fully aware of user credentials and group membership, and it’s enforcing the same permissions, then we’re relatively safe,” Aron explained. “Because you’re not exposing information that users don’t already have access to.”
The Data Curation Pipeline
The CTERA Data Intelligence pipeline works like this:
1. Timely Ingestion – Pull data from NFS, SMB, S3, OneDrive, SharePoint, wherever it lives
2. Format Unification – Extract textual content, transcribe audio/video, perform OCR, convert everything to a unified format
3. Metadata Enrichment – Use AI to extract contextual information and create semantic tags
4. Data Filtering – Apply guardrails to drop or redact sensitive information
5. Vectorization & Indexing – Create embeddings and index for retrieval
The result is what they call ‘curated, high-quality data sets’ that are ready for AI consumption – not the entire file system dumped into a vector database.
Virtual Employees and MCP Integration
The end vision is what CTERA calls “virtual employees” – AI agents that act as subject matter experts, built by business users (not IT) for specific workflows.
During the demo, they showed several examples:
– Paralegal Agent: Connected to legal documents, it could search petroleum contracts, compare three contracts in a detailed table, and cite exact sources
– Lab Assistant: Used multimodal AI to extract structured data from medical reports (doctor name, exam date, exam type, findings summary)
– News Analyst: Used MCP to fetch RSS feeds and generate custom reports based on the user’s role and interests
– Maintenance Officer: Analyzed a 1962 helicopter maintenance manual to provide corrective procedures when engine RPM exceeded limits
The MCP integration is bidirectional. CTERA acts as both an MCP server (so external tools like Claude, ChatGPT, Copilot can access CTERA data) and an MCP client (so CTERA agents can invoke external tools like email, databases, web search, image generation).
“Any Enterprise storage solution that won’t have MCP in the next year will essentially make your data inaccessible,” Aron predicted. Bold statement, but given how fast MCP adoption is growing, maybe not wrong.
Hybrid Deployment for Sensitive Data
One of the smartest architectural decisions CTERA made is supporting fully private deployments. You can run the entire Data Intelligence platform on-premises with private LLMs (say, on an Nvidia DGX cluster) and never send data outside your network.
Alternatively, you can do hybrid: keep all your data and vector embeddings on-premises, but call out to GPT-4 or Claude for inference. Your sensitive content never leaves your environment; you’re just using the public LLM as a compute engine.
“This is what we call corporate AI,” Aron said. “For the Enterprise, where you have top secret and confidential information, ChatGPT will not know the answer. And if your employees copy your sensitive data into ChatGPT, they violate all your regulations.”
Real-World Impact: Medical Law Firm Case Study
The company shared a customer example that illustrates the ROI. A medical law firm uses doctors to analyze insurance claims-reviewing hundreds of documents that can cost thousands of dollars per case.
Using CTERA Data Intelligence with metadata extraction, they automatically pull key fields and generate medical summaries from documents. The result? They’re cutting roughly 80% of the analysis effort.
“The business value here is much faster recommendations,” Aron explained. “These are not just productivity gains-in healthcare and legal contexts, faster analysis can mean better outcomes for patients or clients.”
The Go-to-Market Challenge
During the briefing, a fellow journalist asked a great question about how CTERA’s evolution from infrastructure to AI platform changes their sales motion.
Saimon Michelson was candid: “It adds another way of prospecting. It supplements because we will still be campaigning on the day-to-day basis on efficiency, security, cost.”
The challenge is different personas. Wave 1 (storage modernization) sells to IT infrastructure teams focused on cost, performance, and DR. Wave 2 (cyberstorage) sells to security teams worried about ransomware and compliance. Wave 3 (AI) sells to CTOs and line-of-business leaders thinking about GenAI strategy.
“Selling to this also helps from an alliances standpoint,” Saimon added. “To make sure that you’re maintaining alignment with all the other infrastructure players.” CTERA partners with IBM, Hitachi, HPE, Dell – companies that are also building vector databases, S3 tables, and AI infrastructure.
The product isn’t in partners’ catalogs yet for resale. “Our intent and strategy is yes, to be able to get to that point,” Saimon said. “But the proof point is on us to say we’re not there yet. This product is one that we’re taking to market initially ourselves. Once we’ve proven how customers are seeing it and the value, the next step would be to go through these business relationships and have them offer that as well.”
What This Means for the Industry
Here’s what I find compelling about CTERA’s approach: they’re not pretending storage is sexy. They’re not rebranding as an AI company. They’re saying, “Look, we’ve been managing distributed unstructured data for 17 years. We know the pain points. Now GenAI has created a new requirement, and we’re extending our platform to meet it.”
The three-wave framework actually makes sense. Wave 1 is table stakes – you need unified access to distributed data. Wave 2 is where storage becomes security, which is a hard requirement in 2025. Wave 3 is where it gets strategic, because if you can’t feed quality data to your AI initiatives, you’re part of that 95% failure rate.
The MCP bet is interesting. If MCP becomes the standard way AI agents interact with enterprise data sources (and that’s looking likely), then CTERA’s early adoption puts them in a good position. The hybrid deployment model addresses real concerns about data sovereignty that are killing AI projects in regulated industries.
But there are questions. Building an AI platform is very different from building storage software. The skills required, the go-to-market motion, the competitive landscape – it’s all different. CTERA is competing with companies building dedicated AI data platforms, with hyperscalers offering integrated AI services, and with enterprises just building their own.
Aron acknowledged this: “The Holy Grail of the industry is having an intelligent data fabric that consolidates all the data from different parts of the organization into something that allows GenAI to have access to high-quality, curated, and safe information.”
He’s right. The question is whether a 185-person company, even one growing at 35%, can move fast enough in a market where the giants are also racing. After this briefing, I wouldn’t bet against them.
Final Thoughts
CTERA’s presentation at the IT Press Tour wasn’t about making big acquisition announcements, or signaling splashy new partnerships. Just a methodical explanation of how they’re building on 17 years of distributed file storage expertise to solve the data quality problem that’s killing enterprise AI.
Time will tell if the three-wave strategy resonates with customers. But after watching 95% of GenAI pilots fail because companies can’t get their data house in order, CTERA’s focus on data curation, security, and quality feels more grounded than most of the AI hype I’ve seen.
As Aron put it: “An airline shouldn’t build their own airplanes, right? It doesn’t make sense. We want to build the airplanes to allow these customers to fly.”
If enterprise AI is ever going to get off the ground, somebody needs to build those airplanes. CTERA is betting they’re the ones to do it.
##
