NetApp announced on July 16 that it has acquired DataPelago, a California-based AI-data-infrastructure startup, bringing DataPelago's Nucleus data-processing engine into NetApp's storage platform. The pitch is straightforward: most enterprise AI pipelines today copy data out of storage into a separate GPU compute environment before it can be processed, adding cost, latency and complexity. DataPelago's technology processes that data directly at the storage layer instead, which NetApp says can cut infrastructure costs by up to 80% and improve processing performance by as much as 10x.
The acquisition is part of a broader race among storage and infrastructure vendors to own the 'data readiness' layer of enterprise AI -- the unglamorous but increasingly valuable work of getting messy, siloed enterprise data into a form models can actually use. Dell has been pushing its own AI-optimized storage lines built on Nvidia's Vera Rubin platform, and both AWS and Google Cloud offer native data-preparation services designed to keep customers from needing a third-party layer at all. NetApp's bet is that enterprises with large on-premises and hybrid-cloud storage footprints will pay for zero-copy AI readiness rather than migrate everything to a hyperscaler-native stack.
Financial terms weren't disclosed, and DataPelago will continue operating as a wholly owned NetApp subsidiary rather than being folded directly into existing product lines immediately.
For infrastructure-focused investors, the deal is another data point in a trend that showed up repeatedly this week: capital and M&A activity concentrating on the AI supply chain's less visible layers -- storage, inference serving, brokerage rails -- rather than consumer-facing model applications. What to watch next: whether NetApp integrates Nucleus into its core ONTAP platform broadly or keeps it as a separate premium offering, and how quickly Dell and the hyperscalers respond with competing zero-copy claims.