As artificial intelligence (AI) workloads accelerate across the Middle East, datacentre operators are discovering that scale alone is no longer the differentiator. Density, thermal volatility and sovereign requirements are redefining what operational excellence looks like.
In 2024, Khazna Data Centers launched NexOps, an operating model designed to support AI-scale infrastructure across its growing portfolio. The shift marks a move away from a largely supplier-led model towards a hybrid insourced structure intended to deliver greater consistency, competence assurance and governance control.
“As compute densities rise and AI workloads introduce far tighter operational tolerances, the weaknesses of a vendor-led model become more exposed,” says Bart Holsters, managing director of Khazna NexOps. “The core challenge is variability. Different teams apply procedures differently. Documentation depth varies. Change control discipline can be inconsistent. Incident handling differs from site to site. That fragmentation becomes a real risk when a portfolio must operate like a tightly coordinated system.”
The hybrid insourced model was designed to eliminate what he describes as operational drift. Every facility now runs on a single, unified operating system with identical procedures, key performance indicators, governance structures, escalation paths and training standards. “For AI workloads, where deviations as small as a delayed response or a misinterpreted procedure can have cascading effects, consistency is essential,” he says.
Competence assurance was another driver. Holsters argues that in mission-critical environments, attendance-based training is insufficient. “We needed a model where training, certification and skills validation directly govern who is authorised to perform which tasks. Only approved and competent personnel execute high-risk work. That reduces the likelihood of unintentional operational impact,” he adds.
What AI changes operationally
Holsters is clear that AI workloads are not simply an extension of the cloud. “AI fundamentally changes the operational environment. Workloads are denser. Thermal behaviour is more dynamic. The operational margin for error is significantly narrower,” he says.
In conventional multi-supplier models, variability in habits and interpretations of process can be tolerated. In AI environments, he believes that variability becomes risk. “NexOps treats the entire Khazna portfolio as one integrated system. Regardless of location, customers experience the same operating model, governance and expectations. That level of predictability is what hyperscale customers value most, especially as AI clusters push infrastructure closer to its engineered limits,” says Holsters.
NexOps treats the entire Khazna portfolio as one integrated system. Regardless of location, customers experience the same operating model, governance and expectations Bart Holsters, Khazna NexOps
A distinguishing feature is the link between competence and task authorisation. Operators must be validated and revalidated before performing defined categories of work. “In higher-density environments where the consequences of a single incorrect intervention can be significant, tightly managed competence becomes a core reliability control,” he explains.
As governments across the region invest in sovereign AI initiatives, operational sovereignty is becoming as important as policy frameworks. “A sovereign-ready datacentre provides enforceable proof of how access is governed, how work is executed and how every action is logged and auditable,” he says. “Public sector and regulated customers want confidence that workloads are handled consistently and in compliance with jurisdictional requirements.”
Holsters points to a strong boundary definition as foundational. That includes physical zoning, identity-based access control, disciplined change management and incident procedures that preserve traceability. “Sovereign customers expect continuous assurance. Structured documentation, consistent logging and governance routines ensure compliance is ongoing rather than retrospective,” he says.
The real bottleneck in AI-scale operations
While power constraints continue to shape datacentre expansion in many markets, Holsters believes the most pressing operational constraint is talent. “The most consistent bottleneck, especially for AI-scale operations, is specialised talent paired with mature operating discipline,” he says. “As densities rise and timelines compress, operators need deeper technical expertise, stronger change control and higher incident readiness. That skillset is in short supply globally.”
Operators that treat mission-critical operations as a core capability rather than an administrative necessity will be the ones able to scale reliably under AI demand Bart Holsters, Khazna NexOps
Simply adding headcount or layering additional suppliers can increase variance rather than resilience. Instead, Holsters sees standardisation and real-time, data-driven visibility as the primary levers. “Operators that treat mission-critical operations as a core capability rather than an administrative necessity will be the ones able to scale reliably under AI demand,” he says.
Automation is already embedded in modern facilities, but Holsters draws a clear line between decision support and operational authority. “AI is exceptionally strong at pattern recognition and high-frequency monitoring. It can surface likely root causes and recommend next best actions by correlating millions of datapoints,” he says. “Final approval for change management, incident management and all physical interventions sits with trained operators. AI agents and digital twins can analyse and inform around operational risk levels, but they cannot yet replace accountable human judgement in this environment.”
Looking ahead, Holsters identifies AI-readiness and adoption consistency as the defining metrics for the next 12 to 24 months. “Our priority is ensuring that every operational unit integrates intelligent capabilities in a way that is scalable, secure and aligned with our long-term strategy,” he says. “Customers are no longer consuming capacity site by site. They are depending on a unified AI-ready operating ecosystem. Portfolio-wide consistency in reliability, readiness and efficiency is what protects long-term performance.”
