AI is everywhere now. That is the easy part. The harder part, and the one many companies are still quietly struggling with, is execution.
At Global AI Show Riyadh 2026, the panel discussion “AI Is Accessible vs Execution Isn’t: The Infrastructure, Security & Systems Challenge of Intelligence at Scale” looked at the uncomfortable gap between having access to AI and actually making AI work inside large organizations. The discussion was not about hype, demos, or another polished promise that AI will transform everything overnight. It was about what breaks when intelligence has to operate at scale.
The session was moderated by Dr. Mohammed Abdur Rahman, Professor and Chairman, Department of Cybersecurity and Forensic Computing, College of Computer and Cyber Sciences, University of Prince Mugrin, Madinah, KSA. The panel featured Shadi Alghanim, Chief Technology Officer at AL-AYUNI Investment & Contracting Company, Ghada Alshammari, Chief Information Security Officer at Alraedah, Himanshu Shrivastava, Group Chief Technology Officer at Al-Futtaim, and Mohamed Elsaadany, Chief Governance Officer at Mowah Co.
What Fails First When AI Scales?
One of the sharpest questions from the panel was simple: at scale, what fails first? Data pipelines, compute economics, security, or organizational design?
The answer is rarely one thing. That is what makes enterprise AI difficult. A company may have the model, but the data may not be clean enough. The cloud bill may grow too fast. Security may arrive late. Teams may not know who owns the system once it moves beyond pilot stage. The technology looks impressive in isolation, then starts to wobble when placed inside real business operations.
That is the part many AI conversations skip. Scaling intelligence is not just about plugging in a model. It needs reliable pipelines, governance, infrastructure, security, cost discipline, and people who know how to make decisions around the system. Without that, AI remains available but not truly executable.
The Execution Layer Will Decide the Winners
The panel also explored which execution layer will define enterprise AI winners over the next five years: cloud, edge, silicon, or platforms. Each one matters, but not in the same way.
Cloud gives enterprises scale and flexibility. Edge brings intelligence closer to where decisions happen. Silicon shapes performance, efficiency, and cost. Platforms decide how easily organizations can deploy, monitor, govern, and improve AI systems across departments.
The winners may not be the companies with the most exciting AI experiments. They may be the ones that quietly build the strongest execution layer underneath. Less glamorous, maybe. More important, definitely.
This is where infrastructure becomes strategy. Companies that treat AI infrastructure as a backend IT concern may find themselves stuck in pilot mode. Companies that treat it as a business capability will move differently. Faster in some places. More carefully in others. But with a clearer path from prototype to production.
Data Pipelines Are Still a Weak Link
AI systems depend on data, but enterprise data is rarely neat. It is scattered across departments, formats, legacy systems, regional operations, and compliance boundaries. When AI systems scale, weak data pipelines often become visible very quickly.
A pilot can survive messy data because the scope is small. A company-wide AI system cannot. It needs consistent access, quality control, governance, monitoring, and ownership. Otherwise, intelligence starts making decisions from unreliable inputs, and that creates risk.
This is why the conversation around AI infrastructure at scale has to include data pipelines. Compute may get more attention because it is expensive and visible, but pipelines decide whether AI systems are actually fed with usable information. Without that, even powerful models become unreliable.
Compute Economics Can Break the Business Case
Compute is another pressure point. AI may be accessible, but running it at scale is not cheap. Training, inference, storage, monitoring, and system integration all carry cost. The more AI becomes embedded into daily operations, the more compute economics starts to matter.
For enterprises, this creates a different kind of challenge. It is not enough to ask whether AI works. They also need to ask whether it can keep working at a cost that makes business sense.
That means choosing the right mix of cloud, edge, hardware, and platforms. It means understanding when to use large models, when smaller models are enough, when local processing makes sense, and when centralized infrastructure is more efficient. The companies that get this wrong may still deploy AI, but they may struggle to sustain it.
Secure-by-Design AI Cannot Be Optional
Security was a major part of the discussion, and for good reason. AI systems are increasingly connected to sensitive data, business decisions, customer workflows, financial processes, infrastructure operations, and government services. Securing them after deployment is not enough.
The panel raised the question of whether AI systems can be built secure-by-design rather than secured later. That shift matters. Secure-by-design AI means security is part of the stack from the beginning, not added as a patch after the system is already exposed.
That stack should include data protection, identity controls, access governance, model monitoring, audit trails, threat detection, secure APIs, compliance checks, and clear accountability. It also requires cybersecurity teams to be involved early, not invited after the product team has already shipped the system.
AI security cannot be treated as a final review step. By then, the risk may already be inside the architecture.
Organizational Design Is Part of the Infrastructure
There is another layer that often fails quietly: organizational design.
AI at scale needs more than technical teams. It needs business leaders, cybersecurity teams, governance officers, data owners, legal teams, compliance units, and operations leaders working from the same map. That sounds obvious. In practice, it is usually where things get complicated.
Who approves an AI system? Data ownership also needs a clear answer. When errors happen, someone must investigate them. Performance needs continuous monitoring. Security incidents require a defined response team. Finally, leaders must decide when the system should be paused, retrained, or retired.
These questions are not side issues. They are part of the infrastructure. A company can have strong cloud systems and still fail because decision rights are unclear. It can buy advanced platforms and still struggle because departments are not aligned. Execution depends on systems, but also on the organization around those systems.
Sovereign GPU Clusters Enter the National AI Debate
The panel also touched on a bigger infrastructure question: how sponsors and governments can jointly build sovereign GPU clusters to enable localized AI processing at scale.
This is becoming more important as countries think about AI sovereignty. Localized processing can reduce dependency on foreign infrastructure, improve data control, support national security requirements, and help industries build AI systems closer to their own regulatory and operational environment.
But sovereign GPU clusters are not simple projects. They require investment, power, cooling, technical talent, procurement discipline, and long-term public-private coordination. Governments may provide strategic direction and national priorities. Sponsors and enterprises may bring funding, demand, and use cases. Technology partners may bring systems expertise.
The challenge is making all of that work together without creating expensive infrastructure that is underused, poorly governed, or disconnected from real market needs.
AI Scale Is a Systems Problem
The strongest message from the panel was that AI scale is not only an AI problem. It is a systems problem.
The companies and governments that succeed will not be the ones that simply gain access to the newest models. Access is already becoming normal. The advantage will come from execution: infrastructure that holds, security that is built in, pipelines that deliver trusted data, compute that makes financial sense, and organizations that know how to govern intelligence in motion.
That is why the title of the session felt accurate. AI is accessible. Execution is not.
And as enterprises move from experiments to real deployment, that gap will become impossible to ignore.
