Stephen Draper-Garcia, data centre expert for the UK and Ireland at Desoutter Tools, explains why manufacturing quality and traceability are becoming critical to supplier assurance in the AI era.
Data centre operators understandably design redundancy around what happens after equipment enters the facility. However, integrated AI infrastructure arrives with significant manufacturing and assembly work already completed.
Complete rack and server assemblies are increasingly delivered ready to connect and operate, reducing the need to validate every upstream manufacturing operation again on site. With more assembly and integration moving into the supplier’s factory, some of a data centre’s eventual reliability has already been determined before the rack arrives. This raises an important question: where does uptime actually begin?
More integration, more risk
Several deployment models are used across the data centre ecosystem, including full factory integration, integrator-led deployment and modular pod deployment. Across these approaches, a greater proportion of assembly, validation and integration is completed before equipment reaches site. This approach offers clear advantages. Factory integration can reduce deployment times, improve consistency and support the scale required by AI-driven infrastructure growth. However, it also changes the nature of reliability.
When a highly integrated rack arrives at a facility, operators inherit more than the hardware itself. They also inherit the outcome of hundreds, or potentially thousands, of upstream assembly operations completed elsewhere in the supply chain. Historically, some of these activities might have been validated during installation and commissioning. With more work moving upstream, that opportunity becomes more limited.
Confidence in reliability therefore becomes linked to confidence in how equipment was manufactured.
AI is accelerating supply chains
The industry is scaling at extraordinary speed, putting more pressure on manufacturers and suppliers alike to expand capacity without compromising consistency. AI demand is driving major investment in servers, networking equipment, power systems and supporting infrastructure, while manufacturers throughout the supply chain are being asked to expand capacity rapidly.
Unlike mature sectors such as automotive, where high-volume manufacturing processes have evolved over decades, parts of the data centre supply chain are still adapting to this pace of expansion. For a manufacturer, a production stoppage may be an efficiency problem. For a data centre operator, however, such disruption can raise questions about variability in the manufacturing process behind the equipment being purchased.
Parts shortages, component substitution, interrupted production schedules, changing operators and rapid expansion are not necessarily quality problems in themselves. The key issue is whether suppliers have sufficiently controlled processes to ensure these changes do not alter the quality of the finished product. For operators, growth is not necessarily the risk. Uncontrolled variability is.
A passed product is not the same as a traceable product
Traditionally, quality assurance has focused heavily on final testing. An end-of-line test can demonstrate that a product worked when it left the factory. This remains important, but it answers only one question. Another question for operators is: how was this particular unit built?
Traceability addresses that question by creating visibility into the manufacturing process itself. A sufficiently detailed manufacturing history could associate a serial number with information such as the production sequence, relevant assembly results, operator or workstation records, timestamps, positional confirmations and any non-conformance or rework events that occurred during production.
These are examples of the information a digitally controlled assembly process can make available. Not every supplier will capture every data point, nor should buyers assume that every manufacturer does so. The important point is that end customers can benefit from manufacturing information they can interrogate and understand, rather than simply accepting that records exist somewhere within a supplier’s organisation.
In this context, passing a final test and demonstrating a controlled build process become two different things.
Traceability enters procurement
As AI infrastructure investments increase, manufacturing assurance is moving beyond the factory floor and into procurement conversations. Traditionally, supplier selection has focused heavily on delivery capability, quantity, lead times and commercial terms. Those considerations remain important, but increasing infrastructure complexity is encouraging buyers to look deeper into how products are manufactured.
For buyers, relevant questions include whether manufacturing records can be associated with individual units, whether critical operations can be shown to have been completed within specification and how process changes are controlled as production volumes increase. There is also the question of how non-conformance events are managed and whether suppliers can distinguish the manufacturing history of one rack or assembly from another.
The key consideration is whether quality is consistently controlled throughout the production process and whether suppliers can demonstrate that control. The ability to provide that evidence is becoming an important part of supplier evaluation.
When something goes wrong
When equipment develops an issue in service, operators need enough information to investigate what happened and identify the factors that may have contributed to it. Manufacturing records can provide a route back through the production process, helping investigators establish whether affected equipment was produced during a particular period, associated with a specific component batch or linked to a particular assembly process.
They can also show whether other units share a similar manufacturing history and therefore warrant further investigation. This information does not automatically identify the cause of a failure. Instead, it can provide a structured evidence base for fault isolation, supplier management and corrective action.
That gives manufacturing data a purpose beyond internal factory quality control. It becomes information that can support decisions throughout the equipment’s operational life.
Data centre operators do not need to specify the manufacturing technology their suppliers use. What they need is confidence that critical processes are controlled, results are attributable to individual products and exceptions are visible rather than disappearing into the production line.
For suppliers, demonstrating this level of control is becoming more important as infrastructure complexity and production volumes rise. This means being able to demonstrate that the products they deliver were built through controlled, repeatable and transparent processes, with sufficient evidence to support those claims.
The boundaries around data centre reliability are widening. Redundancy, commissioning and operational excellence will always remain fundamental to uptime. However, with racks becoming more integrated and deployment becoming faster, operators also need greater confidence in the processes that take place before equipment reaches the facility.
This means looking beyond simply asking whether a rack works when it arrives. The way it was manufactured, the consistency of the processes used to build it and the evidence available to verify its production history can all influence the confidence an operator has in the equipment entering the data centre.
For operators, knowing that a product has passed its final test is valuable. Knowing that its critical manufacturing processes were controlled, and that sufficient evidence exists to investigate issues if they arise, provides a stronger foundation for reliable deployment at scale.
As AI infrastructure grows, supplier assurance therefore has an increasingly important role to play in reliability. Uptime does not begin when a rack is switched on. It starts much earlier, with the processes, controls and evidence behind the equipment that eventually enters the data centre.
