Vultr and NetApp: What If AI Compute Could Use Your Existing Datasets With No Migration?

Our view, shaped by nearly every major briefing this year, is that the data layer determines the success of AI projects, not GPUs or models. Compute availability will continue to expand across regions and providers, and models will increasingly be consumed rather than built. What does not change is the requirement for clean, consistent, policy-aligned data delivered to the right place at the right time. This is where most organizations struggle today, and this is where the architecture announced here offers meaningful relief.

The most important outcome of this announcement is how directly it addresses the enterprise data bottleneck. Organizations want to take advantage of cloud GPUs, yet their most valuable datasets remain governed, curated, and protected on-prem. Moving that data is slow, costly, and often unrealistic. By extending the ONTAP environment into Vultr using SnapMirror and ONTAP Select, enterprises gain a straightforward way to run cloud AI on trusted datasets without replatforming or changing familiar governance practices.

The real customer benefit is operational continuity. Teams do not need to learn new data models or convert workloads to object storage. Instead, the cloud-side presentation of the data behaves just like ONTAP on-prem, preserving access controls, retention rules, and semantics. This reduces friction for both IT and AI practitioners, giving organizations a predictable and low-risk path to begin accelerating AI initiatives.

In the broader market, we see multiple prevailing approaches to solving the AI data challenge. For example, a new global data platform designed for AI performance requires customers to adopt a completely new architecture. Another relies on global metadata virtualization to span heterogeneous systems, but this adds abstraction that may complicate governance. The strategic tension for enterprise architects is whether to extend a stable, governed data foundation into cloud GPUs or to migrate toward a next-generation, AI-first data platform that may require significant reinvention.

The Vultr and NetApp announcement strengthens the case for extension by offering a practical implementation that moves compute to the existing data estate with minimal disruption, preserving the maturity, lineage, and control enterprises already rely on. For organizations that prioritize continuity and compliance, this is a highly differentiated and pragmatic model.

There is a competitive perspective here; the Vultr/NetApp architecture highlights a critical bifurcation in the enterprise AI market: the choice of GPU ecosystem. NetApp and Vultr maintain a parallel partnership with NVIDIA, which also leverages ONTAP and the SnapMirror mechanism to power AI workloads. NVIDIA’s competing solution, anchored by the NVIDIA AI Data Platform and the proprietary AI Enterprise (AIE) software suite, is deeply embedded in the enterprise space and offers its own validated workflow patterns (NVIDIA Blueprints).

This creates a direct competitive dynamic for enterprise architects. While the NetApp/Vultr architectural model provides a consistent data foundation, the decision becomes a trade-off between:

• NVIDIA, which could be seen as the software-priority path, leveraging the mature and dominant CUDA software stack, extensive ecosystem support, and enterprise-grade AIE software, this path yields high performance and development speed but potentially at a higher cost and with vendor lock-in risk.
• AMD, a more hardware-focused path, embracing cost-competitive hardware with substantial future headroom (MI355X/MI450), and open-source ROCm platform freedom to reduce long-term TCO and avoid proprietary constraints - but with less developer adoption and platform maturity.

Of note, Vultr and AMD’s commitment to scaling GPU infrastructure continues in parallel: the two recently announced a 50-megawatt AI supercluster being built at a new data center in Springfield, Ohio, with ~24,000 AMD Instinct MI355X GPUs scheduled to come online in early 2026. In addition to expanding raw GPU capacity, Vultr has rolled out its MI355X-based GPU cloud offerings more broadly and reaffirmed its long-term backing from AMD through prior financing rounds and infrastructure investments. This growing hardware and financial commitment strengthens the value proposition of the architecture described above: customers will not only be able to access their existing data via ONTAP + SnapMirror, but also tap into a rapidly expanding, globally distributed pool of high-performance GPU compute that supports both training and inference at scale.