From experiment to execution: why FS&I AI workloads are coming back from the cloud

Across financial services and insurance (FS&I), AI adoption is accelerating at pace. Organisations are experimenting with new models, data pipelines and use cases faster than ever before. For most, that innovation starts in the public cloud, where teams can move quickly, prove value and build momentum.

But as Dell highlights, a clear pattern is emerging. Once AI delivers real business outcomes many FS&I firms are reaching the same crossroads, often finding that production AI workloads need a different platform.

More and more, public cloud is being used to test and validate AI use cases, then proven workloads are being qualified and, increasingly, brought back on‑premises. This is not about stepping back from cloud innovation. It’s about scaling AI in a way that is secure, compliant and operationally resilient.

Why public cloud is a good starting point

Public cloud environments are ideal for early‑stage AI work. They allow organisations to spin up environments rapidly, access specialist services and experiment without heavy upfront investment. For proof‑of‑concept activity, whether that’s intelligent document processing, advanced analytics or model training, this flexibility is invaluable. FS&I teams can move quickly, validate outcomes and build confidence with stakeholders before making longer‑term architectural decisions.

When AI moves into production, the rules change

The challenge tends to appear when AI use cases move from pilot into production. Performance becomes critical, and subsequently the requirements look very different. Costs for example need to be predictable, and data residency, sovereignty and regulatory scrutiny all move to the forefront.

In regulated industries, production AI must be auditable, resilient and tightly governed. Long‑running workloads, rapidly growing data volumes and recovery expectations expose the limitations of public cloud for certain use cases. It becomes harder to demonstrate control, manage cost at scale and meet regulatory expectations consistently.

This is where many organisations pause and reassess where business‑critical AI should run.

You don’t need to revert to legacy infrastructure

It’s important to note that bringing AI workloads back on‑premises does not mean reverting to legacy infrastructure. FS&I organisations expect their on‑prem platforms to deliver automation, self‑service, scalability and consumption‑based models; the same experiences they’ve come to value in the cloud.

This is where cloud‑like on‑prem platforms are gaining traction. Designed specifically to take AI from experiment to execution, these environments combine high‑performance compute with modern data architectures that AI depends on, while giving organisations greater control over cost, data location and resilience.

Dell APEX

Dell APEX gives FS&I organisations a way to run production AI workloads on‑premises, using a cloud operating model. Capacity can be scaled up and down as demand changes, infrastructure is delivered as a service and teams retain the flexibility they are used to from the public cloud, without losing control of where data lives or how it is governed.

This model is particularly well suited to AI, where demand can fluctuate significantly between experimentation and steady‑state production. With APEX, organisations can avoid over‑provisioning, improve cost predictability and simplify lifecycle management, while still supporting high‑performance, long‑running AI workloads.

From a data perspective, this approach also aligns well with how modern AI actually operates. Production AI increasingly relies on tiered data architectures, combining ultra‑fast storage to continuously feed GPUs with large‑scale, cost‑efficient platforms for historical and unstructured data. Intelligent metadata management becomes a differentiator, enabling real‑time insight across extremely large data estates without sacrificing performance, security or compliance.

Security, resilience and trust built in

For FS&I, security and operational resilience are not optional extras. They are foundational requirements, therefore AI platforms must be recoverable, monitored and demonstrably secure by design.

On‑prem AI platforms allow organisations to embed these controls directly into the architecture. This simplifies compliance, strengthens cyber resilience and builds trust with regulators, customers and internal stakeholders, all while supporting AI at production scale.

A pragmatic path forward

The message for FS&I leaders is clear:

• Experiment in the cloud but execute with control.
• Use public cloud services to innovate quickly and prove value.
• Scale successful AI use cases onto platforms designed for performance, resilience and regulatory confidence.

This hybrid journey reflects how AI is actually being deployed today. It allows organisations to move fast without compromising trust, turning AI ambition into sustainable, production‑ready outcomes.

If you’d like to find out more about how Softcat can support your organisation, please contact your Account Manager or our Sales team.