By Dr Abhinav Srivastava
For more than a decade, enterprise cloud strategy has largely revolved around applications: where to host them, how to modernise them, how to scale them and how to operate them efficiently.
AI agents are beginning to challenge those assumptions.
Unlike conventional applications that execute predefined workflows, AI agents can interpret objectives, retrieve information, reason across context, invoke APIs and tools, interact with multiple systems and take actions with varying degrees of autonomy. What begins as a single business request can become a sequence of model calls, data retrievals, API interactions and decisions.
That distinction matters because the cloud environment designed to run applications reliably is not automatically equipped to support AI systems that continuously reason and act.
The infrastructure implications are already becoming visible. Gartner forecasts worldwide spending on AI-optimised infrastructure as a service (IaaS) to reach $42 billion in 2026, representing 96% growth. More tellingly, spending on AI inference is expected to surpass training this year, as agentic AI and production deployments drive continuous compute consumption across enterprise applications and workflows.
Google Cloud's research among more than 1,400 global IT leaders points to the same readiness gap: 83% say their organisations require infrastructure upgrades to support production-grade agentic AI, while four out of five identify security, governance or MLOps among their most significant challenges.
The message for business and technology leaders is becoming clear: AI readiness is no longer only a model-selection question. It is an enterprise architecture question.
Traditional enterprise workloads are comparatively predictable. Applications respond to user requests, execute established business logic and consume infrastructure within patterns that organisations have learned to monitor, scale and optimise.
Agentic workloads behave differently.
An agent asked to resolve a supply-chain issue, for example, may retrieve inventory data, interrogate an ERP system, compare supplier information, invoke multiple models, call external services, initiate a workflow and return to evaluate the outcome. One request can therefore create multiple compute, inference, storage and network events.
This is one reason the economics of AI infrastructure are shifting from training towards inference. Gartner expects inference to account for $23.3 billion of AI-optimised IaaS spending in 2026, compared with $19 billion for training. It specifically identifies the multistep, autonomous execution associated with agentic AI as a factor increasing compute intensity.
For CIOs and CTOs, this changes the capacity-planning conversation. Infrastructure needs to accommodate workloads that may be dynamic, compute-intensive and difficult to forecast using conventional application patterns.
The question is therefore not simply whether an organisation has access to GPUs or scalable cloud capacity. It is whether workloads can be placed, scaled and governed across cloud, hybrid and multi-cloud environments without compromising performance, resilience or economics.
Compute alone will not make an enterprise AI-ready.
Agents derive their usefulness from context. In an enterprise, that context may be distributed across ERP and CRM platforms, databases, documents, operational systems, data platforms, knowledge repositories and legacy applications.
For years, enterprises have invested in moving and storing data. Agentic AI introduces a different requirement: making the right data accessible, understandable and trustworthy at the moment an intelligent system needs to act.
Google Cloud identifies access to business context and semantic meaning as a major barrier to scaling enterprise agents, arguing that data must evolve from passive systems of record towards systems capable of supporting action.
Integration becomes equally important.
An AI assistant can generate a recommendation. An AI agent may need to execute it. That requires controlled access to APIs, applications and business services.
As agents become participants in enterprise workflows, APIs are no longer simply connectors between applications. They increasingly become controlled pathways through which intelligent systems interact with the enterprise.
This creates an important architectural question: are your enterprise systems merely connected, or are they ready to be safely consumed and acted upon by AI?
Enterprise identity architecture has traditionally focused on people, applications and machine identities.
Agents introduce another actor.
Consider an AI agent authorised to retrieve customer information, interact with an ERP platform, create a purchase request or initiate an approval. The organisation now needs to determine not only whether the agent can authenticate, but also what it is authorised to do, whose authority it operates under, which data it can access, and how to trace every action.
The distinction is important.
A chatbot producing an incorrect answer creates one category of risk. An autonomous system with excessive privileges taking an incorrect action creates another.
Agentic AI therefore makes identity, least privilege, workload protection, API security and data governance integral to cloud architecture. The objective should not be to constrain autonomy entirely, but to establish boundaries within which autonomy can operate safely.
For regulated organisations in government, financial services, healthcare and critical infrastructure, this becomes particularly significant. AI adoption cannot come at the expense of existing obligations around security, privacy, data sovereignty and accountability.
Intertec's current enterprise cloud strategy reflects this convergence, combining hybrid and multi-cloud architectures with CloudOps, FinOps, resilience, governance, compliance and data-sovereignty considerations.
Cloud observability has traditionally answered questions such as: Is the application available? Where is latency increasing? Which infrastructure component failed?
Agentic systems require another layer of visibility.
Organisations increasingly need to understand which models and tools an agent invoked, what resources it consumed, which systems it interacted with, whether an action succeeded and how that activity affected performance and cost.
In other words, observability must begin to extend from system health to agent behaviour.
This does not replace established disciplines such as Site Reliability Engineering. It expands their scope. SLOs, SLIs, error budgets, infrastructure monitoring and application observability remain essential, but enterprises will increasingly need to correlate those signals with the activity of intelligent workloads. Intertec's SRE approach already brings together operating-model assessment, reliability metrics, observability, managed cloud operations and FinOps — disciplines that become increasingly relevant as AI workloads move into production.
Without that visibility, organisations may deploy autonomous systems faster than they can explain their behaviour, diagnose failures, or understand their operational impact.
The next challenge is financial.
Cloud economics were already complex before AI. Organisations have spent years improving visibility across compute, storage, networking and software consumption through FinOps.
AI adds new variables: GPU utilisation, inference, model selection, token consumption, data movement, API calls and potentially multiple execution steps for a single business outcome.
The issue is not simply that AI can be expensive. It is that agentic workloads can make consumption less predictable.
An application typically follows a defined execution path. An agent may determine that completing an objective requires several intermediate steps. Each step can consume additional infrastructure and AI services.
The issue is not simply that AI can be expensive. It is that agentic workloads can make consumption less predictable.
An application typically follows a defined execution path. An agent may determine that completing an objective requires several intermediate steps. Each step can consume additional infrastructure and AI services.
That means the financial conversation must evolve beyond overall cloud spend towards understanding the economics of individual AI workloads.
What did this intelligent workload cost, and was the business outcome worth it?
This is where FinOps will become increasingly important to enterprise AI strategy. Cost allocation, tagging, workload visibility, and continuous optimisation will need to extend into AI consumption so technology leaders and CFOs can connect infrastructure expenditure to measurable business value.
Intertec already positions FinOps around aligning finance, operations, and technology, with cost governance, reporting, accountability, and continuous optimisation across cloud environments.
For leaders, this is not simply a cloud-cost issue. It is part of establishing the economics of enterprise AI.
Governance becomes more consequential when AI moves from recommending an action to taking one.
Much of the enterprise AI governance conversation has focused on questions such as which models may be used, where data is processed and what information can be shared with them.
Agentic AI adds a more operational question:
What is the AI permitted to do?
That requires organisations to consider approval thresholds, access boundaries, human oversight, audit trails, policy enforcement, and accountability at the point of execution.
An organisation may be comfortable letting an agent analyse invoices autonomously but require human approval before releasing a payment. Another may allow an agent to identify a cybersecurity anomaly but not automatically isolate a mission-critical workload without additional validation.
There is unlikely to be one universal level of autonomy.
The important capability will be designing governance around the risk and consequence of the action.
This is why AI governance cannot remain separate from cloud governance, cybersecurity and enterprise architecture. When intelligent systems become active participants in business processes, governance must operate where those systems interact with data, applications and infrastructure.
For enterprises across the GCC, these questions are becoming particularly timely.
In Saudi Arabia, Microsoft has confirmed that its Saudi Arabia East datacenter region will become available in November 2026, providing three Azure Availability Zones and local access to supported cloud and AI services. Microsoft explicitly positions the investment around data residency, resilience and the Kingdom's transition from AI experimentation towards deployment at scale.
This reflects a broader shift across the region. The conversation is moving beyond whether cloud and AI infrastructure will be available towards how enterprises should architect, secure and govern what they build on it.
For governments and enterprises operating in highly regulated environments, local infrastructure creates opportunity, but architecture still determines the outcome. Data residency does not automatically provide data governance. Cloud capacity does not automatically provide resilience. GPU availability does not automatically create an AI-ready enterprise.
The same principle applies as cloud and AI adoption accelerates across India and other high-growth markets: infrastructure investment must be paired with modernisation, integration, operational discipline, and governance.
There is a temptation to treat AI readiness as another infrastructure upgrade: procure more compute, establish access to models and begin deploying use cases.
That is necessary, but insufficient.
An AI-ready cloud foundation needs to bring together scalable compute, modern applications, accessible data, secure integration, identity, cybersecurity, observability, financial governance and operational resilience.
Those capabilities also need to work across the environment’s enterprises operate in: public cloud, private cloud, on-premises infrastructure and increasingly hybrid and multi-cloud architectures. Google Cloud's 2026 research found that 52% of surveyed organisations already use hybrid multi-cloud architectures, reinforcing why AI readiness cannot be treated as a single-platform infrastructure decision.
For Intertec, this reflects how we see cloud transformation evolving. Modernisation is increasingly about more than moving workloads. It requires a secure, governed, and economically sustainable operating foundation that brings together modern applications, scalable cloud infrastructure, DevSecOps, cloud security, containerisation, SRE, AI-ready compute, and FinOps.
Organisations that get this foundation right will be better positioned to move AI from isolated pilots into business-critical operations without creating a parallel architecture that becomes difficult to secure, observe, or control.
Cloud transformed where enterprises run applications. AI agents are beginning to transform how work gets executed across those applications.
That distinction should change the questions being asked in the boardroom.
The next phase of cloud strategy will not be defined simply by how much infrastructure an organisation can consume or how quickly it can deploy an AI model. It will be defined by whether the enterprise can give intelligent systems the compute, context and connectivity they require while retaining control over identity, security, cost and governance.
The cloud that successfully supported the last decade of application modernisation may therefore need to evolve for the next decade of autonomous enterprise operations.
For business and technology leaders, the time to test that readiness is not when thousands of agents are already operating across the enterprise.
It is before they are given the authority to act.
