Core42’s Sherif Tawfik on why governed AI beats big AI budgets
Core42 CBO Sherif Tawfik on maximising AI value and why 2026 shifts the focus from capability to control
10 August, 2026
TT
16
Enterprises have spent the past two years racing to adopt AI. The harder question, according to Sherif Tawfik, is what all that spending is actually returning. As chief business officer at Core42, Tawfik argues that the winners in this next phase won’t be the organisations that spend the most on AI, but those that treat it as a governed operating capability rather than a one-off technology purchase, matching workloads to the right infrastructure, measuring output against business outcomes, and building cost, governance and sovereignty into a single discipline from day one.
In this interview with Gulf Business, tied to Core42’s new Compass whitepaper, he explains how to maximise value from AI investments, the budgeting mistakes that quietly inflate costs, and why 2026 marks a decisive shift from training to inference and from capability to control.
How can organisations maximise value from AI investments?
The organisations that gain the most from AI are those that stop treating it as a technology purchase and start managing it as an operating capability, governed, measured and optimised like any other part of the business. Adoption alone is not the goal. The objective is to deliver useful business outcomes repeatedly, economically and at the speed each use case requires.
Every AI use case should have an accountable business owner, a measurable objective and a clear baseline, whether the goal is to resolve customer inquiries faster, improve decision-making, reduce document-review time or create a new digital service.
In practice, this requires three things. First, visibility: organisations cannot optimise what they cannot see, so leaders need a clear view of where AI is being used, what it costs and whether it is producing meaningful results. Second, each workload must be aligned with the right model, infrastructure and deployment path to improve the economics of AI at scale. Not every task requires the largest model, the most expensive accelerator or real-time processing. Matching workloads to their actual requirements can improve performance and utilisation while avoiding unnecessary expenditure.
Finally, organisations must measure AI against outcomes rather than activity. We describe the metric that matters as tokens per second per dollar, which captures how much useful output an enterprise receives, what it costs and how quickly it is delivered. Value is maximised when investment is linked directly to business results, and inference is run as an engineered, governed system.
Read: UAE’s Core42 secures $550m from HSBC to expand AI infrastructure in US, Europe
What mistakes do businesses make when budgeting for AI?
One of the most common mistakes is treating AI like traditional enterprise software, where a license is purchased once and used freely. In production, AI behaves differently: every interaction carries a marginal compute cost, and consumption scales with adoption, context size, latency expectations and workflow complexity rather than with headcount. A pilot that appears inexpensive with a handful of users can become costly once thousands rely on it daily, or when agents begin running multi-step tasks that turn a single request into many inference events behind the scenes. Cheaper and faster inference tends to encourage greater usage, so lower unit prices can still result in higher overall spending.
A second mistake is anchoring the budget to the quoted price of a token while overlooking the wider cost of delivering a business outcome. The right question for a leadership team is not what a token costs, but what it costs to resolve a customer case, process a document or complete a workflow, at the speed and reliability the business requires. Judging that way, the cheapest token does not always produce the lowest cost of a meaningful result.
The third, and often most consequential, mistake is treating cost control and governance as something to address after deployment rather than a leadership decision made before scale. When AI is opened up broadly without the accountability to see where it is being used and to what end, spend compounds quietly and becomes difficult to attribute to any department, project or business outcome. The lesson is not that AI is inherently too expensive; it is that its economics are a strategic responsibility to be managed from day one.
What should boards measure to assess AI ROI?
Boards should assess AI against business outcomes rather than raw token counts or model usage. The right question is not how much AI a company is consuming, but what that consumption is producing: measurable gains in productivity, revenue contribution, customer and employee experience, and the cost and speed of delivering the outcomes that matter to the business. Framed this way, AI is judged like any other major investment, on the value it returns rather than the volume it uses.
That value has to be weighed against the operating economics of AI. Because consumption scales with adoption rather than headcount, a board’s central concern should be whether the return per dollar is improving as usage grows, or whether spend is simply compounding. We look at the economics of AI through the relationship between cost, useful output and delivery speed. Organisations need a simple way to understand whether business value is improving as AI usage grows.
The third dimension is accountability. Boards do not need to monitor AI at a technical level, but they should be confident the organisation can answer a few fundamental questions. Is spend visible and attributable across the business? Is AI operating within approved cost and governance limits? Is sensitive data being handled appropriately for the markets we operate in? An organisation that can answer those questions confidently is far better positioned to scale AI without losing control of cost, performance or compliance, and that level of control itself is one of the clearest indicators of sustainable return.
How are AI spending priorities changing in 2026?
AI spending is moving beyond experimentation and model access towards the infrastructure and operating capabilities required to run AI reliably at scale. IDC expects global AI infrastructure spending to exceed $1tn by 2029, representing an estimated compound annual growth rate of around 31 per cent from 2025. Sovereign investment and national AI strategies across the region are accelerating this trajectory. However, the more important change in 2026 is qualitative, not simply quantitative.
Priorities are shifting from training towards inference, and from capability towards control. Organisations increasingly recognise that access to a powerful model is only the beginning. The higher and more sustained cost emerges during inference, when AI becomes embedded in everyday operations, and each interaction consumes compute. The decisive question is whether intelligence can be delivered repeatedly, economically and under control at production scale.
As a result, budgets are moving towards inference efficiency, observability, workload-aware routing across diverse silicon and governance, with sovereignty built in where required. Cost and governance are also converging into a single discipline rather than being treated as separate considerations.
Agentic AI will accelerate this shift because a single user request may trigger multiple inference events as systems reason, call tools and execute across workflows. Spending will therefore increasingly focus on platforms that make this consumption visible, allocate costs accurately and keep usage within defined limits. The organisations that gain the most in this phase will not necessarily be those that spend the most, but those that make inference efficient, observable, resilient and governed.
Why should cost, governance, and sovereignty be considered together?
In the inference era, a single AI request is simultaneously an economic, a governance, and, increasingly, a sovereignty event, and treating them as three separate problems is what leaves organisations exposed. The commercial question of what AI costs cannot be answered honestly without also knowing how it is being used, by whom, and within which boundaries. Cost, control and trust are now one conversation.
Managing these considerations through separate systems creates gaps in visibility and weakens accountability. It can also force organisations to retrofit controls after deployment, when usage has already expanded, and architectures have become more difficult to change. In regulated and high-trust sectors, these gaps can create significant financial, operational and compliance risks.
The stakes are particularly high for public-sector bodies that require sovereign control and accountability, financial institutions that depend on auditability, and healthcare organisations that must carefully manage sensitive data. Across jurisdictions, regulation is also moving towards more transparent, accountable and risk-managed AI operations.
A unified approach connects expenditure with the policies governing it. Platforms such as Core42 Compass are emerging to address this challenge by bringing together cost visibility, governance controls and sovereign deployment requirements within a single operating framework



















