The New Front Line: Securing Data Centers in the AI Era

For years, physical security in the data center sector lived in the background. It mattered, of course. We cared about mantraps, cameras, perimeter fencing, guards, badges, logs, cages, and chain of custody. But for many operators, physical security was treated as a mature function. Important, yes. Strategic, not always.

Market forces have pushed our industry to quickly change that mindset about physical security.

Data centers have crossed a threshold. They’re no longer just technical facilities that support business operations. They now sit in the same category as power plants, hospitals, transportation hubs, and water systems: critical infrastructure whose disruption can create immediate economic, societal, and even geopolitical consequences. The U.S. Department of Energy has repeatedly emphasized the heightened risk of physical and cyberattacks against the electric grid, while also warning that electricity reliability and security are under increasing strain. That matters to our sector because the same logic now applies to the facilities that host AI, financial systems, public-sector workloads, communications platforms, and the digital services people depend on every minute.

One big reason to think more strategically about physical security is that the value concentrated inside these facilities has changed dramatically.

Not long ago, a stolen server might have meant losing a standard Dell PowerEdge box or similar enterprise rack server worth several thousand dollars, or perhaps tens of thousands once fully configured. Today, a single AI rack can represent a concentration of value that would have been unthinkable a decade ago. NVIDIA’s DGX H100 platform is built around eight H100 GPUs, and the DGX B200 around eight Blackwell GPUs, with associated high-density power and cooling requirements that move far beyond legacy enterprise norms. A single DGX B200 system can cost roughly half a million dollars, meaning that one rack of modern AI infrastructure can represent millions of dollars in concentrated compute value.

AFCOM’s 2026 State of the Data Center report shows average rack density climbing to 27 kW, up from 16 kW last year, while 72% of respondents expect AI workloads to increase capacity requirements and 74% plan to deploy AI-capable solutions. In other words, we’re securing far more expensive systems, in denser environments, with more strategic value and less margin for error.

Data Centers Are No Longer Quiet Targets

One of the most dangerously outdated assumptions in our market is that data centers remain obscure or operationally invisible. If you turn on your favorite news channel, or read your favorite news publication, you’ll note that data centers—and their locations-- are being discussed. A lot.

AI has made the data center industry extremely visible. Major campuses are now publicly discussed in terms of location, megawatts, GPU counts, national competitiveness, cloud strategy, and even military-adjacent relevance. Facilities that support large AI environments are increasingly seen as strategic assets. That visibility attracts investment, talent, and attention. It also attracts threat actors.

It’s not like cyber risk is getting smaller as physical threats rise. I wrote in Data Center Knowledge about the ransomware attack “that costs you everything,” and the lesson still stands: ransomware, destructive malware, and data exfiltration remain absolutely central threats. In that incident, production systems and even backups were impacted, turning what should have been a recoverable event into a full-blown business catastrophe.

But the uncomfortable truth is that we now have new targets on our backs. The same attackers, insiders, criminal networks, activists, and hostile state-aligned actors that once focused primarily on data theft or extortion are increasingly looking at access paths that include people, facilities, vendors, logistics, and the physical environment itself.

The rising physical security risk is showing up clearly in the data.

In the 2026 AFCOM State of the Data Center report, for the first time in 10 years, human threats took the top risk spot, overtaking ransomware. Specifically, 61% of respondents ranked human threats, internal or external, among the top concerns, compared with 58% for ransomware. The reports also notes that 84% of respondents changed physical security requirements over the past 12 to 24 months, with the most common actions including more comprehensive surveillance and monitoring, stronger vendor and contractor vetting, and increased perimeter security and access controls. That is a major signal. Our industry is recognizing that the threat model is no longer just code. It is people, process, access, identity, and intent.

That signal tracks with what many of us are hearing in the field.

At Data Center World 2026, we’re hosting conversations that take a broader view of the risks to data centers. One panel discussion, “Risk and resilience: Overcoming operational and technology challenges,” centers around uptime, reliable energy, physical and cyber threats, changing customer needs, and technology obsolescence. Resilience is no longer just about N+1, 2N, and service-level agreements. The market is addressing a much broader risk envelope.

The Data Center Is Both Robust and Fragile

This is where the conversation gets serious.

On one hand, modern data centers are some of the most well-engineered environments in the world. Redundant power paths. Layered cooling. Sophisticated controls. Physical barriers. Strict procedures. Compartmentalized access. Response playbooks. Monitoring everywhere.

On the other hand, these facilities can still be astonishingly fragile in the face of the wrong event, at the wrong time, in the wrong place.

I explored this years ago in my article on how the internet could be taken down. My point was never that the internet is flimsy. It was that highly interconnected systems can fail in nonlinear ways when critical nodes, routes, or facilities are disrupted. Interdependence creates efficiency and scale, but it can also magnify consequences. A seemingly localized incident can cascade across customers, services, regions, and supply chains. That reality has only become more relevant as AI increases both concentration and dependency. Critical digital infrastructure is now more essential, more interconnected, and in some cases more targetable than ever.

We have real-world reminders of this fragility. The OVH Strasbourg fire remains one of the clearest examples. An official investigation that documented the March 2021 fire that destroyed the SBG2 building underscored how physical incidents can take out major infrastructure and customer environments. Fire doesn’t care how advanced your compute platform is. It doesn’t care how impressive your AI roadmap sounds. A sufficiently severe physical event can still destroy a facility.

And now the geopolitical risk is escalating.

When Drones Strike Data Centers, the Conversation Changes

I understand I’m broaching a potentially sensitive and unpleasant topic. But we as an industry must discuss the risk to data centers during war and conflict with greater rigor and candor.

The Guardian reported that drone strikes this month against AWS data centers in the UAE and Bahrain raised fresh doubts about the Gulf’s ambitions as an AI superpower. One quoted assessment bluntly suggested that this “means missile defense on data centers.”

Read that statement again … so, missile defense systems on the rooftops of our data centers?

The article describes these incidents as among the first deliberate military strikes on commercial data center infrastructure. Separate Guardian reporting stated that data centers are becoming a target in warfare for the first time.

Even if most facilities around the world won’t need literal missile defense protection, that phrase should stop every operator in their tracks.

Why? Because it forces a new perspective.

It means we can no longer define physical security narrowly as gates, guards, and cameras. We have to think in concentric layers: airspace, perimeter, logistics, adjacent infrastructure, vendor ecosystems, wireless exposure, public visibility, local unrest, regional instability, and supply-chain compromise. The data center is now part of a much wider risk surface. The building isn’t the boundary anymore.

That also brings supply chains into the security conversation. The AFCOM report notes that 66% of respondents continue to experience ongoing supply-chain challenges, and 15% reported outages tied directly to supply-chain delays. Security and resilience are now linked to whether you can source the right power gear, cooling components, networking equipment, and replacement parts quickly enough when something goes wrong. A facility can be physically hardened and still operationally exposed if the recovery chain is weak.

Physical Security Must Expand Beyond the Traditional Checklist

So what should the industry do?

First, we need to stop thinking of physical security as a compliance box and start treating it as a core design discipline. That means involving physical security earlier in site selection, facility design, operational planning, and customer engagement. Not after the shell is built. Not after the first tenant move-in. At the beginning.

Second, we need to recognize that the old threat hierarchy is shifting. The AFCOM report is very clear here: human threats are now the top concern for the first time in a decade, ahead of ransomware. The rising risks should prompt a reassessment of budget, governance, and executive attention. Security leaders need authority that matches the stakes.

Third, we need to accept that cyber and physical security are now one conversation. The access badge, the vendor credential, the cloned voice, the spoofed identity, the maintenance visit, the wireless device, the drone, and the exfiltration campaign may all be part of the same attack chain.

That is one reason I recently spent time with Bastille discussing how hackers attack AI data centers using wireless devices. The point is simple and urgent: the attack surface around modern AI infrastructure is broader than many teams realize. Wireless threats can include rogue devices, unauthorized radios, interception, spoofing, and nontraditional access methods that fall between classic cyber and classic physical security. When you combine high-value AI infrastructure with dense operational environments and new tools for attackers to use, the result is a much more dynamic risk environment than the industry has historically planned for. Bastille’s guidance emphasizes visibility into wireless activity, layered detection, and understanding that RF is now part of the security perimeter whether teams like it or not.

Practical Advice: How to Secure Facilities for the Threats We Actually Face

If we agree the threat model has changed, then operating practices have to change too.

Start with the perimeter, since fencing, vehicle controls, bollards, cameras, and access gates still matter a lot. But they’re only the outer shell, so don’t stop there. Operators should build multiple internal control layers so that a perimeter breach doesn’t become a facility breach, and a facility breach doesn’t become a critical-space breach.

At a practical level, security leaders should focus on several strategic priorities:

•       Layer defenses beyond the perimeter. Ensure that each successive layer of the facility introduces new controls and monitoring capabilities. A perimeter breach should never translate into direct access to critical infrastructure.

•       Treat identity and access management as a physical security control. Contractor vetting, credential discipline, visitor management, and escort enforcement are just as important as cameras and gates.

•       Expand monitoring to include emerging threat vectors. This includes drones, rogue wireless devices, and other technologies that can operate outside traditional security assumptions.

•       Practice incident response regularly. Security isn’t only about prevention. Teams must rehearse coordinated responses across operations, facilities, IT, and security leadership.

In terms of real-world implementation, here are some items that should be on your must-do checklist, not a “nice to have.”

•       Harden identity and access management across physical operations. This includes rigorous contractor vetting, tight visitor workflows, temporary credential discipline, escort enforcement, and better reconciliation between physical access logs and digital authentication systems. If someone badges into a critical space, their digital access patterns should make sense. If they do not, that should trigger response. The AFCOM report’s emphasis on human threats, vendor vetting, and access controls supports exactly this direction.

•       Expand surveillance from passive recording to active detection. Many facilities have cameras. Fewer have sophisticated analytics, behavior baselining, and meaningful response integration. The goal is not just to capture footage after an incident. The goal is to identify anomalies while there is still time to intervene.

•       Take drones seriously now, not later. Facilities should work with counsel and security specialists to assess local legal options, detection technologies, response protocols, nearby approach paths, rooftop vulnerabilities, and airspace exposure. Not every site needs the same controls. But every site should have a drone risk assessment.

•       Treat wireless as part of physical security. If you are operating AI-rich environments, you should know what is transmitting in and around your facility. Unknown RF activity near critical zones should not be invisible. This is particularly relevant for facilities hosting high-value training clusters, sensitive inference environments, or government-adjacent workloads.

•       Reassess fire risk, suppression capabilities, and compartmentalization in light of higher-density AI deployments. As infrastructure value rises and power density increases, the consequence of a localized thermal event or electrical fault grows. The OVH fire is a stark reminder that entire buildings can be lost to fire.

•       Build security into supply-chain resilience planning. Validate suppliers, diversify critical vendors, identify long-lead spares, and understand which dependencies would delay recovery after a physical incident. A resilient facility without a resilient replenishment path is only half secured. The AFCOM report’s finding that supply-chain issues are still affecting two-thirds of respondents should be a wake-up call.

•       Run integrated exercises. Not just cyber tabletop exercises. Not just evacuation drills. Run cross-functional scenarios that include insider access misuse, contractor impersonation, drone overflight, wireless anomaly detection, utility disruption, media escalation, and physical damage to critical systems. Resilience lives in rehearsal.

And finally, it’s absolutely critical to elevate the physical security conversation to the executive and board level. If data centers are now strategically important critical infrastructure, then physical security belongs in board-level risk conversations alongside cybersecurity. power strategy, insurance, compliance, and customer trust.

Final Thoughts: Secure What Matters Before the Market Forces You To

This is an uncomfortable topic that forces us to confront an industry reality that is getting harder to ignore.

We’re building data centers that are more valuable, more visible, more concentrated, and more strategically important than ever before. We’re putting extraordinary compute capacities inside them. We’re tying data centers to AI innovation, national competitiveness, public services, financial systems, and global supply chains. And in doing so, we’re making them more attractive targets.

The answer isn’t fear.

The answer is thoughtful action.

The hopeful part of this story is that our industry knows how to adapt. We have done it with power. We are doing it with cooling. We are rethinking how data centers themselves are designed and operated to support AI-scale infrastructure. Now we need to apply that same level of innovation to physical security and resilience. Not with panic. Not with theatrics. With clear-eyed realism, better design assumptions, stronger operational discipline, and broader collaboration across security, facilities, cyber, and executive teams.

Let’s be honest--Missile defense systems won’t be the answer for most data centers. But a new mindset absolutely will be.

The data center operators that succeed in the next era will be the ones that understand a simple truth: in a world where data centers are critical infrastructure, physical security is no longer something you bolt on. It is something you build in.