Scality wants everyone to stop drooling over GPUs and look at the boring bits that break in production.
The cyber-resilient storage outfit announced findings from independent research by Freeform Dynamics on how enterprises build AI infrastructure that runs day in, day out.
The study with the catchy title, Storage infrastructure for enterprise AI: lessons from seasoned adopters on building scalable sovereign environments, says that as AI moves from tinkering to operations, data and storage are turning into the main constraints.
As AI use firms up, more organisations are running private AI, often branded sovereign AI, to keep control of the tin and the data that feed their models.
Keeping workloads closer to enterprise data can improve performance, help meet regulatory requirements, and take the sting out of long-term costs, at least in theory.
Scality argues that as infrastructure costs drop, private AI is becoming a practical partner to cloud AI services rather than a niche hobby.
Within these private environments, object storage continues to emerge as a foundation for on-premises AI pipelines.
As deployments harden into production, infrastructure choices start to mirror the unglamorous needs for reliability, governance, cost control and lifecycle management.
The report said the real headache is the pipeline, not the accelerators. Enterprises running private AI are focusing on systems that stage, govern, protect and reuse data across inference-driven pipelines, because the models do not run on vibes.
The survey found 91 per cent of enterprises running private AI in production make meaningful use of object storage, the highest adoption across storage architectures.
Object storage use splits into 44 per cent using it extensively and 47 per cent using it quite a bit, nudging ahead of file-based storage and well above block storage.
Another 81 per cent said private AI infrastructure they control is critical to success, pushed by sovereignty, compliance and keeping data close to where it lives.
Storage performance is flagged by 57 per cent as a priority to avoid AI bottlenecks, versus 54 per cent citing compute or GPU availability and 52 per cent citing network bandwidth.
Rather than ripping everything out, 44 per cent are adapting existing compute for AI and 42 per cent are adapting existing storage.
Purpose-built setups still show up, with 40 per cent building compute for AI and 39 per cent building storage for it, suggesting tiered hybrids are more common than clean-slate builds.
Scaling metadata is a risk for 40 per cent, and 38 per cent report mixed-workload handling issues, which arise when you try to serve training and low-latency inference on the same estate.
Freeform Dynamics, director of engagement and distinguished analyst Tony Lock said: “Most industry discussion frames AI infrastructure as primarily a compute challenge.”
“This research makes clear that enterprises running private AI in production are dealing with a broader systems reality. Many see a need for simple, scalable architectures that keep data close, support multiple AI genres, and balance performance with governance and cyber resilience across the full pipeline,” he said.
The survey targets organisations running AI in live operations rather than in pilots, with respondents already facing the consequences.
As day-to-day AI becomes more inference-driven, the research says enterprises are settling on tiered architectures that mix fast tiers for active workloads with scalable capacity tiers for persistent, reusable data.
In that setup, S3-compatible object storage becomes the foundation layer, with file and other storage types still kept around for workload-specific jobs.
Scality, chief marketing officer Paul Speciale said: “The data defines the problem, and the platform determines who scales. This research validates what we see in the field: the success of production AI depends on how effectively teams manage and operationalise data throughout the AI lifecycle. Scality provides an S3-native, tiered, cyber-resilient foundation aligned with how enterprises are building sovereign AI today. It delivers the control, predictability, and operational resilience required to scale.”
