A featured contribution from Leadership Perspectives: a curated forum reserved for leaders nominated by our subscribers and vetted by our editorial board.
Mohan Veloo is a technology leader with deep expertise in Applications, Networking, and Security. He drives digital transformation, presales excellence, and innovative solutions across large enterprises, bridging technical leadership with strategic business outcomes. A change agent and visionary, Mohan has led global teams, architected AI- and cybersecurity-driven initiatives, and consistently champions resilience, innovation, and scalable technology adoption across the APAC region.
In this authored article, Mohan Veloo shares insights on shifting enterprise AI in India from efficiency to trust, reliability, and operational availability.
For the past two years, enterprise AI has been framed as a race.
The race to deploy faster, to reduce inference costs, and to maximise GPU utilisation; efficiency became the headline metric. It was measurable, reportable, and easy to celebrate. But across boardrooms in India, a different question is now being asked. Not ‘How fast can we deploy AI?’ but ‘Can we trust it when it matters most?’
As we move into 2026, enterprise AI will no longer be judged purely by how efficiently it runs. It will be judged by how reliably it remains available, delivering correct and secure outcomes under real-world conditions. And that is a far more demanding standard.
The Efficiency Illusion
At a recent roundtable in India, senior technology leaders shared a common reality. Many had successfully moved beyond pilots. AI was already embedded into customer journeys, internal copilots, fraud detection systems, and decision engines. ‘Deployment was no longer the challenge. Operational confidence was!’.
Efficiency is tangible. Latency improves. Cost per interaction falls. Infrastructure metrics look healthy. These signals create the impression of progress.
Yet, F5’s Architecting the AI-Enabled Infrastructure. Research shows that only a small fraction of organizations are truly prepared to scale AI securely and reliably across their enterprise environments. Most are accelerating model deployment faster than they are building the governance, observability, and resilience required to sustain it. This is the efficiency illusion.
AI systems rarely fail in dramatic ways. They do not simply crash. They degrade. Outputs become inconsistent. Models drift. Responses remain fast but subtly incorrect. Hallucinations slip into workflows unnoticed.
At the India AI Impact Summit, leaders were candid about this gap. Monitoring infrastructure is a solved problem. Monitoring AI behaviour is not. Without effective oversight and guardrails, availability becomes probabilistic rather than engineered.
In a lab environment, that may be tolerable. In a live enterprise setting, it is not.
Enterprise AI will no longer be judged purely by how efficiently it runs. It will be judged by how reliably it remains available, delivering correct and secure outcomes under real-world conditions.
India’s Acceleration and the Complexity Curve
India has emerged as one of the fastest-moving AI markets globally. Enterprises across banking, telecom, manufacturing, and digital services are running multiple generative AI use cases in production. AI is no longer experimental. It is operational. But rapid adoption has introduced complexity.
Hybrid environments span data centres and multiple clouds. Different models are integrated across business functions. GPU investments have surged, often ahead of mature operating frameworks.
The first wave focused on acquiring compute. The next wave must focus on controlling and operationalising it. One executive at an India CXO roundtable summarised it succinctly: ‘We are not struggling to build models. We are struggling to industrialise them’.
Industrialisation demands more than efficiency. It demands availability.
Also Read: How Data, AI, and Safety Engineering Are Shaping the Future of Driving
Rethinking Availability in the AI Era
In traditional IT, availability meant uptime. A system was either accessible, or it was not.
In AI systems, availability is multidimensional. The model must be reachable. The inference pipeline must perform under load. Outputs must be correct and consistent. Security controls must prevent prompt injection and data leakage. Behaviour must remain stable under unexpected inputs.
An AI system that responds instantly but produces unreliable results is not truly available. It is operating as a risk multiplier. As AI becomes embedded in mission-critical workflows, the tolerance for silent failure diminishes. A flawed recommendation in a chatbot is an inconvenience. A flawed output in a financial decision engine, healthcare workflow, or network automation system is systemic risk.
By 2026, AI will sit at the heart of customer experience, revenue generation, and operational control. The enterprises that succeed will not necessarily be those with the lowest inference cost. They will be those whose AI systems remain stable, secure, and trustworthy under pressure.
Architecting for Availability
Mohan’s point of view is simple. We must architect for availability first and optimise for efficiency second. AI should be treated as an application tier, not as an experimental add-on. That requires deliberate design across traffic management, security enforcement, observability and portability.
Organisations must ensure that AI traffic is intelligently routed and controlled. Security policies must extend to AI ingress and egress points. Observability must cover not only performance metrics but also behavioural consistency. Guardrails must be embedded as a design principle rather than bolted on reactively.
At the India AI Impact Summit, one message resonated clearly: AI value compounds only when trust compounds.
Trust is built when systems behave predictably under stress. It is reinforced when enterprises can explain, monitor, and control AI decisions. It is sustained when customers experience reliability rather than volatility.
Availability, therefore, is not just a technical metric. It is a business strategy. It protects brand equity. It reduces regulatory exposure. It enables AI initiatives to move from experimentation to monetisation.
Enterprises will not pay for AI capacity alone. They will pay for dependable AI services that integrate seamlessly into core operations without increasing risk.
The Reckoning Ahead
We are entering a new phase of enterprise AI. The first phase was experimentation, second was acceleration and third will be operational accountability. In this phase, efficiency still matters. But resilience, correctness and control will matter more.
The conversation is shifting from ‘How fast can we deploy?’ to ‘Can we depend on it when it counts?’
Efficiency drives headlines. Availability builds enterprises. In 2026, that distinction will define who truly succeeds in AI.