AI in Security: Infrastructure, Not Hype, Will Determine ROI

Artificial intelligence (AI) is no longer experimental—it is mainstream. The 2025 AI Index Report found that 78% of organizations used AI in 2024, up from 55% the year before. But adoption does not equal desired outcomes. Too many organizations still treat AI as a collection of use cases instead of a core operational capability—that gap is where value is lost.

For security leaders, the takeaway is simple: AI success depends as much on infrastructure and operations as it does on compute.

That reality is reshaping the security industry and system architectures. AI and security are converging into what many now call “physical AI”—systems that combine video, sensors, access control and operational technology to create smarter environments.

Across organizations, operations, safety and facilities teams are expecting more from existing camera systems. They want tools that do more than record incidents. They want continuous monitoring, workflow improvements and faster informed decision-making.

Security professionals are often first in line to answer. Their experience with cameras, analytics and system design puts them in a strong position to lead.

Many organizations have responded by launching AI councils and collecting ideas around automation, agentic AI and video analytics. That is a solid first step. The breakdown typically comes at execution.

A common misstep is to start in the cloud and figure out the cost later. For physical AI—especially video-based deployments—this approach can backfire quickly. Compute is only part of the bill. Data transfer, storage growth and network egress charges can drive long-term costs far higher than expected. Video intensifies the problem. It produces large, continuous data streams that must be processed, moved and stored.

What works in a pilot may fail at scale.

The pattern is familiar. The industry saw it during the shift to video storage, where short-term convenience sometimes came at the expense of long-term efficiency. Physical AI raises the stakes—it adds high-performance compute requirements along with power, cooling and low-latency demands. Infrastructure is no longer a backend concern—it is central to the return on investment.

Organizations that want results must approach physical AI as a complete system, built on four fundamentals:

• Use cases: AI is not a cure-all. Large language models, machine learning and video analytics solve different problems. Focus on use cases with clear operational value, measurable outcomes and realistic data needs.
• Integration: Systems must work together. Cameras, access control, sensors and operational platforms cannot remain siloed. Fragmented data leads to unreliable insights. Integration turns tools into an ecosystem.
• Data: This remains the biggest barrier. An estimated 80–90% of enterprise data is unstructured, making it difficult to organize and use. Much of it is also outdated or incomplete. Before scaling AI, organizations need to clean data, map processes and establish governance.
• Infrastructure: Performance and cost hinge on architecture. Servers, storage, clustering, power, cooling and placement all matter. The goal is not just to run AI—it is to run AI efficiently at scale. For physical AI, that means low latency, controlled operating costs and the ability to grow. And not all AI demands the same level of compute. Installing an overpowered system may provide certainty and speed, but it can lead to excessive upfront costs and increased, ongoing cooling requirements.

The organizations that get this right will treat infrastructure as part of the solution, not an afterthought. This will enable them to move beyond experimentation and turn AI into measurable business value.

The views and opinions expressed in guest posts and/or profiles are those of the authors or sources and do not necessarily reflect the official policy or position of the Security Industry Association.

This article originally appeared in All Things AI, a newsletter presented by the SIA AI Advisory Board.