Is the AI Infrastructure Challenge Shifting From Capacity to Efficiency?
Sun 20 Sep 2026
Demand for practical AI knowledge is emerging as a priority among technology professionals, while discussions at Tech Show London suggest that improving existing infrastructure could be as important as investing in additional capacity.
Two recent Tech Show Insights articles approach this challenge from different perspectives. Research from Frankfurt has shown what technology professionals want to learn in 2027, while insights from around the UK have explored how organisations can improve the performance of GPU infrastructure before committing to further hardware investment.
Together, they raise a question for enterprises moving AI into production: are organisations making the most of the infrastructure they already have?
Practical AI Leads Audience Priorities
A survey of 166 technology professionals conducted ahead of Tech Show Frankfurt 2027 found that 65% of respondents who answered the relevant question selected Data & AI in Practice among their three preferred programme themes.
Interestingly, only around a quarter of respondents identified AI, data, and analytics as the area most relevant to their work. Although the questions had different response bases, the findings suggest practical AI has appeal beyond specialist AI roles.
AI Factory & Infrastructure Readiness was the second most selected theme, followed closely by Cyber Resilience & Human Risk and Sovereign AI & Secure Cloud.
The survey also revealed a preference for learning through practical examples. Case studies, hands-on demonstrations and technical deep dives were each selected by around half of respondents, ahead of traditional keynote presentations and panels.
These findings reflect the priorities of those surveyed rather than the wider German technology market. Nevertheless, they offer a useful indication of the questions some industry professionals want future conference programmes to address.
More GPUs May Not Solve the Problem
Insights from Before You Buy More GPUs, Fix the Ones You Have examines a practical example of why understanding infrastructure performance matters for the UK.
As organisations expand AI workloads, growing demand for accelerator capacity can make additional hardware seem necessary. Yet an apparent GPU shortage may reflect limitations elsewhere in the system.
At Tech Show London, Seif Mazigh of Crusoe explained how bottlenecks in storage or networking can leave GPUs waiting for data rather than performing useful work.
Jeff Fan, Senior Solutions Architect at DigitalOcean, also described how changes to GPU architecture improved performance in a specific inference deployment without adding hardware.
“How you split your GPU matters more than how many GPUs you have,” said Fan.
The example illustrates why organisations need to examine how their existing infrastructure is configured before assuming more accelerators will improve performance.
Understanding Performance Before Increasing Investment
Findings from the Frankfurt survey and the analysis by Tech Show London provide different forms of evidence. One identifies demand for practical learning among prospective event attendees; the other examines the operational decisions involved in running AI infrastructure efficiently.
The connection is particularly relevant as organisations move beyond experimentation.
Access to additional compute can support growth, but it cannot resolve every performance problem. If workloads are constrained by inefficient memory use or poorly configured systems, further hardware investment may increase costs without addressing the underlying limitation.
Equally, optimisation cannot eliminate genuine capacity shortages. Additional GPUs may remain necessary once organisations have established what their workloads require and measured the constraints affecting performance.
Frankfurt’s findings suggest an appetite for the practical knowledge needed to make these decisions. London’s analysis demonstrates how that knowledge can influence infrastructure investment.
For enterprises, the implication is that AI readiness depends on understanding what existing technology can deliver before deciding what more needs to be purchased.
