From data to intelligence: Welcoming DataPelago to NetApp

Last October, when we introduced NetApp® AFX and the NetApp AI Data Engine (AIDE), I wrote that AI must come to the data, not the other way around. To support this, we committed the platform to near-data processing and zero-copy pipelines so AI can run wherever data resides. Now we take another big step in delivering on this promise: NetApp has acquired DataPelago, and with it, Nucleus, a GPU-accelerated data processing engine purpose-built to run analytics and AI where enterprise data lives.

CEO and Founder, Rajan Goyal of DataPelago and its engineers took on a widening gap: the world creates more data than it can process. In three years, they built an engine that outperforms the leading data processing engines in the industry to eliminate AI bottlenecks.

For twenty years, enterprise systems have moved data to compute because compute was scarce and datasets were small. Neither is true anymore. Data is now the heaviest asset a company owns, and accelerated compute is efficient enough to sit inside the data path. The architecture has to follow.

As we help customers build an Intelligent Data Infrastructure, this acquisition does more than just add an engine to our platform: it completes an architecture to accelerate AI workloads while maintaining visibility, governance and control over data.

Why AI stalls: The copy tax

Most AI projects stall before they deliver value. Not because the models aren’t good enough, but because the data isn’t ready. Gartner predicts that through 2026, 60% of AI projects that are not supported by AI-ready data will be abandoned. I cited that number in October, and everything we have seen since has confirmed it.

Data sits in one set of systems and gets processed in another, so every AI initiative begins by paying what I call the copy tax: pipelines to build and maintain, duplicate estates to store and secure, egress to pay, and GPUs idling while data is still in flight. While the tax is invisible on any single invoice, it is enormous in aggregate.

That is where most AI budgets quietly go.

Value is moving to the data layer

Everyone is buying the same chips and running increasingly similar models. The value lies in a company’s own data, and this is where accelerated compute must happen. It comes down to data gravity: the biggest datasets are the hardest to move and the most valuable to use, so the engine comes to the data. The market sees it: Gartner sized data and analytics software at $191.6B in 2025, with growth significantly outpacing the broader software market.

We see it in our own results. NetApp closed fiscal year 2026 with more than 1,100 AI and data preparation wins, nearly triple the prior year. The largest share of those wins is for data preparation and data lake modernization. The ask from customers is consistent: make data AI-ready where it sits.

Training frontier models consumes surprisingly little enterprise storage; the data is public, and the copies are few. Enterprise AI is different: it runs on a company’s own data through inference, retrieval, and fine-tuning, and by our own measure about 80 percent of AI storage use comes from inferencing. That work happens against data we already hold.

Agents raise the stakes

When analytics was driven by people, people asked questions at human speed. Batch pipelines could keep pace. But agents changed the arithmetic. An agent queries constantly, unpredictably, and at machine speed, and a pipeline built to refresh overnight cannot feed a system that acts every second. NetApp built the AI Data Engine with native support for the shift from generative to agentic AI. For agents, the copy tax stops being a cost problem and becomes an architecture problem.

Better by design: Nucleus, AIDE, and AFX

This is why DataPelago matters so much. Nucleus is a data processing engine written in Rust and built for accelerated computing from the first line of code. It takes the query plans engines like Apache Spark already produces, decomposes them into operators, and executes each one on the best available silicon, CPU or GPU, with vectorized, columnar execution throughout and no changes to the applications above. It reads and writes open formats, including Apache Iceberg, so data stays in open standards and customers keep their choice of tools. In customer deployments, it has delivered processing up to 10 times faster at up to 80 percent lower cost.

Nucleus extends the GPU acceleration we brought into the storage environment with the AI Data Engine. In combination with NVIDIA AI Enterprise software and NIM microservices already embedded in AIDE, and running on data compute nodes within AFX, certified for NVIDIA DGX SuperPOD™, it provides enterprises:

• Accelerated data preparation and open lakehouse analytics on data in place
• Curation, guardrails and vectorization from the AI Data Engine, applied at processing speed
• Comprehensive governed estate for analytics, GenAI and agentic workloads, with no duplicate copies to secure

This is true zero-copy activation: analytics and AI running on data where it lives, with no copies into separate compute infrastructure. It is the zero-copy pipeline vision we published last fall, now with an execution engine. AI that runs on copies is cut off from lineage, policy, and audit; in place, it inherits all three.

AI where your data lives

Enterprise data lives on NetApp at scale, with exabytes under management, at the edge, on premises, and in every major cloud. Customers never need to sacrifice choice. Nucleus works with the engines and formats the community has standardized, and the NetApp data platform remains disaggregated by design, so enterprises modernize without lock-in, on the infrastructure and clouds they already trust. Sovereign and air-gapped environments make the point without any economics: when data cannot leave, the processing has to come to it.

No one builds AI alone. Nucleus joins an ecosystem we have built deliberately: an NVIDIA partnership in its eighth year with more than a thousand shared customers, and ISV collaborations across the AI pipeline.

Vision realized: From experimentation to execution

NetApp’s vision for enterprise AI puts data at the center. AFX and the AI Data Engine built the foundation. Nucleus closes the distance between managing storage and activating intelligence. What customers get is what has always mattered: the right data, in the right place, at the right time, ready to move from experimentation to execution with confidence.

We will begin aligning Nucleus with the AI Data Engine and AFX, and we will share the initial milestones at NetApp INSIGHT this fall. If you’re ready to accelerate your AI journey, join us in Las Vegas.