Last Updated 2 hours ago by Kenya Engineer
In an industry being reshaped by artificial intelligence, the challenge is no longer how to build bigger data centres, but rather how to rethink the relationship between power, cooling, computing and the people who design them. On the sidelines of ITW Africa 2026/Datacloud Africa in Nairobi, Kenya Engineer’s Samuel spoke with Wojtek Piorko, managing director for Africa at Vertiv about the engineering challenges behind the AI revolution—and why collaboration may be as important as technology.
Artificial intelligence is often discussed in terms of models, chips and applications. But behind every AI system is a less visible and increasingly demanding physical infrastructure: power systems capable of handling rapidly changing loads, cooling systems designed to remove extraordinary amounts of heat, and data centres engineered to accommodate computing densities that would have been difficult to imagine only a few years ago.
That is creating a new challenge for engineers.
As AI deployments accelerate, conventional approaches to data-centre design are being tested. GPU-accelerated racks can exceed 100 kW, pushing conventional air cooling towards its limits and accelerating interest in direct liquid cooling. At the same time, power demand from AI workloads can fluctuate as thousands of GPUs synchronise their processing.
For Wojtek Piorko of Vertiv, however, the transformation is about more than replacing one cooling technology with another. It represents the emergence of a new field—one in which traditional boundaries between electrical engineering, mechanical engineering, IT and facility operations are becoming increasingly difficult to maintain.
He argues that engineers and technical professionals entering this space should approach it with an open mind. The complexity of AI infrastructure makes collaboration essential, he says, and there is considerable value in engineers learning from one another and sharing experience across different parts of the industry.
Kenya, with its growing data-centre market, strong connectivity ecosystem and position as a regional technology hub, is well placed to participate in this next phase. But doing so will require infrastructure that is not only powerful, but resilient, adaptable and designed with future AI requirements in mind.
In this conversation, Piorko discusses what AI means for data-centre engineering, why liquid cooling is becoming increasingly important, the case for rigorous commissioning, and what engineers need to understand as Africa prepares for a more AI-intensive digital economy.
On the impact of rapid AI adoption on data-centre power and cooling design
Wojtek Piorko: The AI conversation is becoming increasingly focused on the infrastructure. As AI deployments accelerate and require rapid scaling, engineers and data-centre designers are facing unprecedented challenges.
For example, established air-cooling systems, which have been sufficient for standard IT loads, struggle with the extreme power densities of modern graphics processing unit (GPU)-accelerated racks, which can exceed 100 kW. This is driving increased adoption of direct liquid cooling to support performance and reliability.
Addressing these challenges requires a more integrated approach to infrastructure design, bringing together power, cooling and IT systems from the earliest planning stages. This closer alignment can help organisations deploy and scale AI infrastructure more effectively while optimising power and thermal performance.
On how AI is reshaping collaboration across the data-centre ecosystem
Wojtek Piorko: AI is changing how data-centre infrastructure is planned and delivered. What was once a siloed industry, made up of hardware manufacturers, software developers, network providers and facility operators working largely in isolation, is now evolving into tightly integrated ecosystems.
That collaboration is essential to overcome the logistical, technical and operational hurdles involved in deploying thousands of GPUs at a pace that would have seemed impossible until recently.
The constant development of new applications and use cases will add further complexity. We will need infrastructure capable of handling dynamic, multi-user and high-value workloads securely and efficiently.
On what Kenya needs as it prepares for AI-driven growth
Wojtek Piorko: Kenya’s strong connectivity ecosystem, vibrant technology sector and growing data-centre market make it well positioned to benefit from AI adoption.
As organisations deploy AI-driven services, demand will increase for resilient digital infrastructure, including power, cooling and edge-computing capabilities.
The opportunity for Kenya is to build AI-ready infrastructure that supports innovation locally while also serving the broader East African region.
On whether existing data centres can be adapted for AI
Wojtek Piorko: In many cases, existing facilities can be upgraded to accommodate AI workloads, but the options depend on available power capacity, cooling capabilities, floor space and structural considerations.
Some organisations may choose phased modernisation, while others may require purpose-built AI facilities. Careful assessment is essential to determine the most effective path forward.
Why commissioning is becoming critical in liquid-cooled AI data centres
Wojtek Piorko: In air-cooled data centres, issues are often easier to isolate, and hardware faults or temperature problems can be escalated to the operations team. High-density AI racks introduce far greater complexity.
Before IT racks arrive, comprehensive commissioning—from Levels 1 through 5—is a critical step. This sequence validates that the mechanical and electrical systems can support the specific requirements of a liquid-cooled system before any IT hardware is installed.
Levels 1 and 2 cover factory and site acceptance testing for individual components. Level 3 involves functional testing of the integrated systems.
The most critical phases are Levels 4 and 5, which use load-bank racks. These non-functional units contain heaters and flow-control valves designed to replicate the thermal and electrical signature of a live GPU cluster.
By operating these load banks, engineers can stress-test the secondary loop at its specified rated capacity. This process identifies potential leak points, validates pressure drops across the manifolds and confirms that the facility’s control logic responds correctly to rapid thermal changes.
Only after the simulated operation proves stable is the high-value IT hardware moved onto the floor.
In high-density AI data centres, commissioning is no longer limited to validating individual systems. It also confirms that power, cooling and control systems perform effectively as an integrated infrastructure solution.
On why liquid cooling is becoming essential to next-generation AI infrastructure
Wojtek Piorko: One of the biggest challenges is removing the heat generated by high-density AI GPU racks, which can exceed the capabilities of conventional air-cooling systems.
AI factory deployments, particularly those using high-density GPU infrastructure, are increasing the role of liquid cooling in system design. Globally, some projects are also examining whether waste heat from AI clusters can be redirected for district heating or other industrial uses.
This requires a high degree of mechanical integration between the data centre and external infrastructure. It adds complexity and highlights the importance of assessing these solutions as part of a complete, end-to-end system.
As AI deployments grow, liquid cooling is expected to become increasingly important, particularly in hyperscale and high-performance computing environments.
Air cooling will continue to play an important role, but many future AI deployments will likely require hybrid or liquid-cooling solutions to support higher rack densities efficiently.
Africa has the opportunity to incorporate these technologies early as new facilities are developed.
On the electrical engineering challenges posed by AI factories
Wojtek Piorko: The electrical profile of an AI factory differs significantly from a conventional workload.
Large-scale training models do not draw power in a linear fashion. Instead, they create shifting demand dynamics as thousands of GPUs synchronise processing tasks.
Managing these dynamics requires a holistic approach to the power train—from the utility connection and switchgear through to uninterruptible power supply systems, power distribution and the IT load.
Power skids and UPS systems must be designed to support rapid changes in load demand while maintaining power quality and system reliability.
In large-scale sovereign AI projects, this often necessitates direct coordination with utilities and grid providers to support sufficient capacity and future growth. Modular power skids provide a predictable electrical footprint that can be tuned to the specific requirements of the AI hardware.
On why traditional building management systems are no longer enough
Wojtek Piorko: AI data centres require a more integrated approach to monitoring and control because power, cooling and IT systems have become increasingly interdependent.
Rather than managing individual subsystems in isolation, operators need greater visibility across the entire infrastructure stack to optimise performance, efficiency and reliability.
Increasingly, organisations are using advanced analytics and AI-assisted control capabilities to monitor data from coolant distribution units, chillers and IT equipment.
These tools can identify patterns associated with different AI workloads. A large model-training run, for example, may create a different thermal profile from inference workloads, placing different demands on cooling infrastructure.
By anticipating these changes, cooling and power infrastructure parameters can be proactively adjusted, including pump speeds, flow rates and chiller set points.
This predictive approach helps maintain stable operating conditions, improve energy efficiency and support the reliable operation of high-density AI environments.
On how Vertiv is helping customers prepare for higher-density AI environments
Wojtek Piorko: Preparation starts with assessing existing infrastructure capacity and understanding future AI requirements.
Vertiv works with customers to develop a roadmap that combines proven reference architectures, standardised infrastructure building blocks and custom solutions aligned with their operational, performance and scalability objectives.
Through an integrated approach spanning power, cooling, monitoring and lifecycle services, Vertiv helps customers deploy and scale AI infrastructure more efficiently.
This system-level approach can reduce complexity and deployment risk while supporting the performance, availability and flexibility required by increasingly demanding AI workloads.
His advice to engineers preparing for the next generation of AI infrastructure
Wojtek Piorko: This is an emerging field, and I think engineers and technical personnel need to approach it with an open mind.
The first thing I would emphasise is early collaboration. Bring hardware, facility and infrastructure partners together from day one in a workshop-style orchestration rather than relying on traditional hierarchical contracting.
That breaks down silos and allows everyone to align around reference architectures, timelines and IT-stack requirements.
The second is rigorous pre-commissioning. Invest the time in load-bank testing and Levels 1 through 5 commissioning to validate power, cooling and control systems before deployment. That upfront effort can prevent expensive downtime once high-value AI hardware is live.
We also need to design for adaptability. Use modular, hybrid-ready systems that support future density increases and allow incremental adoption of liquid cooling without requiring an overhaul of existing infrastructure.
Digital tools will also become increasingly important. AI-driven analytics and digital-twin technology can help optimise performance, test scenarios and improve planning, particularly for large-scale AI deployments with complex operational and compliance requirements.
Engineers also need to look beyond the rack. The focus should be on the full power train and thermal chain—facility fluid loops, heat rejection and energy orchestration, as well as rack-level power and cooling.
Waste-heat recovery opportunities and efficient fluid management, including temperature, pressure and flow, can directly affect power usage effectiveness and operational costs.
Finally, we need to build trust and share knowledge. The industry can benefit from lessons learned across hyperscale, colocation and enterprise deployments. Open collaboration and the exchange of operational experience can accelerate adoption while reducing implementation risk.
As AI deployment timelines continue to compress, driven by the need to reduce Time to Token while managing significant capital investments, critical infrastructure is no longer simply a supporting function. It has become a core enabler of the entire AI ecosystem.
Engineers who master adaptive liquid cooling, modular deployment strategies and integrated solutions will be best positioned to deliver reliable and efficient infrastructure for the next wave of AI innovation.
-
Product on sale
AWEBook Volume 2Original price was: KShs 3,500.00.KShs 3,000.00Current price is: KShs 3,000.00.
