Western Digital’s Alex Segeda explains why AI surveillance systems must be built around data, not just compute.
As AI-enabled systems are increasingly adopted in UK’s public safety programmes, a fundamental shift in supporting IT architecture is underway.
With data storage volumes expanding at unprecedented rates, the real challenge is no longer how much AI compute power is deployed, but how this growing flood of information is captured, stored and managed over time.
From intelligent video analytics in city centres to facial-recognition trials and automated threat detection, AI is transforming how security teams operate.
The UK has a high density of CCTV systems, with millions of cameras generating vast streams of data every day.
As AI capabilities are layered onto this infrastructure, the data footprint compounds rapidly. It is this data that ultimately determines the long-term performance, value and success of surveillance systems.
For years, AI infrastructure has been defined primarily by compute, with a strong focus on GPUs, CPUs and performance benchmarks.
That approach made sense during early experimentation, when the goal was simply to run models at scale.
However, as AI surveillance systems move into continuous, real-world operation, that perspective no longer holds.
Once AI surveillance systems move into production, a clear divergence emerges: compute is episodic, data grows continuously.
This is especially true with dense high-resolution video, where storage demands can grow, especially as businesses move toward 4K, 8K and beyond.
Processing workloads, such as analysing live video feeds or running detection models, scale up and down depending on demand.
Hardware can be reused and optimised. Data, however, behaves differently. It does not reset. It accumulates.
Every AI-enabled camera generates new data, including metadata, logs, behavioural patterns and derived insights.
A single incident detection can produce layers of contextual information comparable in size to the original footage.
Across large-scale deployments, such as smart cities, this accumulation becomes a defining characteristic of the system.
As a result, AI infrastructure is not just a pure compute system. It behaves like a data system.
A critical but often underestimated factor is the rapid growth of generated AI data.
Beyond raw video footage, AI systems create embeddings, annotations, alerts and operational intelligence.
This derived layer can quickly outpace original inputs, becoming the primary scaling challenge.
In the UK, regulatory frameworks such as GDPR and the Surveillance Camera Code of Practice add further complexity.
Organisations must balance the need to retain valuable data with strict requirements around privacy, access and storage duration.
This is where workload-specific, tiered storage architecture becomes foundational.
Modern AI surveillance environments must be inherently multi-tiered. High-performance flash storage supports real-time analytics, while capacity-optimised HDD-based tiers manage warm and long-term retention of video and metadata.
Single-tier approaches using flash only, for example, quickly become inefficient and very costly at scale.
Designing across tiers is therefore essential to balance performance, cost and compliance over time.
A common assumption is that storage should scale in line with compute.
While this may apply in early deployments, it becomes unreliable in production environments.
Compute investment typically follows cycles, driven by hardware refreshes and efficiency gains.
Storage demand, by contrast, grows continuously with data accumulation, retention policies and governance requirements.
Data is persistent and compounds. Over time, storage becomes not just a supporting element, but the critical foundation in the design factor.
When storage is treated as an afterthought, two gaps emerge.
The first is architectural. Storage is positioned downstream, despite being responsible for long-term durability and accessibility.
The second is economic. Costs expand unpredictably as data volumes increase, making total cost of ownership (TCO) more difficult to control.
These issues tend to develop gradually. Systems may perform well initially, but as data scales across nationwide deployments or integrated platforms, strain begins to appear.
Importantly, this challenge is rarely due to compute limitations, but to an under-architected data layer.
Speed remains important in AI surveillance, but performance is no longer defined purely by processing throughput.
It is about ensuring that data is captured, consistently available, durable and resilient.
Reliable access to critical data empowers AI systems to function at their full potential, making robust architecture and scalable infrastructure the foundation of effective AI performance.
This is particularly significant in security environments, where uptime, reliability and evidential integrity are non-negotiable.
From policing to infrastructure protection, systems must operate continuously and withstand disruption without data loss.
The security sector is moving rapidly from experimental AI deployments to persistent, operational systems.
Decisions made now will shape infrastructure resilience, costs and effectiveness for years to come.
Organisations that succeed will recognise that AI surveillance systems scale on data, not just compute. They will design infrastructure around the full data lifecycle.
This requires a forward-looking approach. Data volumes are expected to grow significantly over the next three to five years, driven by higher-resolution cameras, wider deployment and more advanced analytics.
Revisiting infrastructure decisions after deployment is complex and costly. Building with data at the centre from the outset is therefore essential.
