THE AI RECKONING: Profit, Promise and the Cost of Capital

AI Business Briefing | April 2026

The global AI investment narrative is approaching a critical inflection point. After a decade of exponential capital deployment, with the four largest AI hyper-scalers alone projected to have spent over $325 billion in 2025, rising toward a potential $1 trillion annually by 2027, the question that boards, CFOs, and institutional investors are beginning to ask with growing urgency is a deceptively simple one: Where is the profit?

Our AI briefing below, examines the evidence for and against meaningful AI bottom-line impact, and interrogates the opportunity cost of the capital deployed, as well as assessing whether a financial and political reckoning is forming on the horizon.

THE CASE FOR: WHERE AI IS MOVING THE NEEDLE

The optimistic case for AI ROI is not entirely without substance, although it is significantly more concentrated by sector, company size, and geography, than its proponents suggest.

In financial services, the evidence for AI adoption is the most compelling. In this sector, AI-powered loan processing has delivered up to an 80% reduction in approval times, and compliance monitoring tools have demonstrably cut false positives. Zest AI has reportedly generated between $1 million and $12 million in annual profit growth per lending institution. While in the healthcare sector, automation of document processing has freed tens of thousands of staff hours monthly at scale, with measurable turnaround improvements (this has been mainly in developed economies with advanced healthcare systems). In marketing and sales, companies like US Bank have reported 260% increases in conversion rates following AI-assisted lead scoring implementations.

At the macro level, early adopters across industries report an average of $3.70 returned for every dollar invested in AI, within the top segment of performers are achieving returns of approximately 18% — comfortably above a typical 10% cost-of-capital hurdle rate.

PwC’s 2025 Global AI Jobs Barometer found that leading AI adopters reported three times higher revenue per worker growth than peers. And intra-African trade data aside, the most verifiable productivity gains from AI have emerged in narrowly defined, high-repetition, data-rich environments, which is precisely where the technology was always likely to perform best.

These are not trivial numbers. However, they represent a thin slice of the enterprise landscape, and the critical qualifier throughout is “early adopters” — a group that, by definition, does not represent the market majority.

THE CASE AGAINST: THE GAP BETWEEN HYPE AND HARD RETURNS

The counterargument is where the data becomes distinctly uncomfortable for AI evangelists.

A 2025 IBM survey of 2,000 CEOs globally found that only 25% of AI initiatives over the past three years had delivered expected ROI.

An MIT study titled The GenAI Divide: State of AI in Business 2025 found that a staggering 95% of enterprise AI pilot programmes failed to deliver measurable financial returns. The S&P Global dataset adds further weight: the share of companies abandoning most of their AI projects jumped from 17% in 2024 to 42% in 2025. The average ROI for scaled AI projects, once the honeymoon period of pilots fades, has settled at approximately 7%, well below the 10% cost-of-capital threshold that serves as a standard capital expenditure hurdle rate for most listed companies.

Critically, prices have not fallen. If AI were genuinely delivering the efficiencies its advocates promise, the logical corollary of lower costs passed to consumers, such as reduced service pricing or cheaper goods, should be visible. This is clearly not the case, at any meaningful scale.

Overbuild Risks Persist

Infrastructure costs are, in fact, rising. Data centre electricity consumption in the US alone reached 183 terawatt-hours in 2024 and is projected to reach 426 TWh by 2030. McKinsey estimates global data centre capital expenditure will reach $6.7 trillion by 2030, with $5.2 trillion of that directly tied to AI workloads. These are costs that ultimately flow through the economy in the form of higher energy prices and infrastructure, are direct burdens on ratepayers, not lower prices for end consumers.

Microsoft’s own CEO Satya Nadella has publicly warned about a potential overbuild of AI infrastructure and the need to start measuring real impact. Goldman Sachs has questioned whether the estimated $1 trillion in projected AI capital expenditure will ever deliver a meaningful return.

No major frontier AI company relying on third-party data centres is yet profitable on a standalone basis. OpenAI, Anthropic, and their peers are, at present, sustained by investment capital, not revenue surpluses.

Impact on SME’s Amplified

For SMEs, the picture is bleaker still. The cost of meaningful AI adoption that requires hardware upgrades, skilled technical staff, data infrastructure, change management, and ongoing maintenance, is formidable relative to balance sheet capacity. The gains available to large enterprises with proprietary data, existing digital infrastructure, and dedicated AI teams are structurally inaccessible to most smaller businesses. The AI consultant class that has emerged to bridge this gap has, in many cases, compounded the cost problem without proportionate returns.

THE OPPORTUNITY COST QUESTION: CAPITAL MISALLOCATED?

This is where the analysis becomes most consequential — and most contested.

The hypothesis that AI represents the greatest misallocation of capital in recent history deserves serious examination rather than dismissal.

The numbers are not abstract. Goldman Sachs estimates that AI-related capital expenditure could reach $2.9 trillion between 2025 and 2028. The venture capital community has dramatically rotated toward AI and adjacent defence technology, crowding out investment in sectors with more tangible human need.

The FAO Food Price Index rose 6.9% year-on-year in 2025. The UN projects that by 2025, half the world’s population could be living in water-stressed regions. In agri-food technology – precisely the sector capable of addressing both food security and global poverty, venture investment has been in steep decline since 2021, accounting for a tiny fraction of AI-directed capital flows.

The counterfactual argument — had even a portion of the capital deployed into AI infrastructure been directed instead toward physical infrastructure, advanced food production systems, clean water, and education — is not frivolous.

These are sectors with guaranteed demand, positive social externalities, measurable outcomes, and the capacity to expand the consumer base of the global economy by lifting populations out of poverty. Every person lifted from subsistence poverty becomes a consumer, a taxpayer, and a contributor to global growth. The economic multiplier from that trajectory is arguably more durable and compounding than the productivity gains from automating the knowledge work of already-wealthy economies.

Furthermore, the concentration of AI wealth and its benefits within the Magnificent Seven (Apple, Microsoft Meta, Tesla, Alphabet, Amazon, Nvidia), which now account for more than 30% of S&P 500 market capitalisation, represents a narrowing of the economic pyramid, not a broadening. This is the inverse of what broadly distributed infrastructure investment produces.

IS A RECKONING COMING?

The signals are accumulating. A recent survey of 500 global CFOs found that one in two would cut funding for AI initiatives that cannot demonstrate measurable ROI within twelve months. Gartner placed generative AI on the downslope of its hype cycle heading into a “trough of disillusionment” – a phase that describes where inflated expectations collide with accountants.

The AI Now Institute has flagged the systemic risk of what amounts to a publicly subsidised infrastructure buildout for the benefit of a handful of private firms.

China’s AI development experience is instructive: reports as recently as March 2025 indicated that up to 80% of its newly built AI computing resources were idle, a result of misallocated capital chasing prestige over demand.

For listed companies, the reckoning may be closer than many boards currently appreciate. Analysts are beginning to forecast that CFOs will face direct questions on earnings calls about their AI strategy, and then, within a few quarters, will be required to report measurable outcomes. The patience window for returns on AI investments is narrowing. Investors want scale, productivity gains, and revenue growth within timelines, a requirement that AI, as currently deployed across most enterprises, is not meeting.

CONCLUSION: MEASURED OPTIMISM REQUIRES HONEST ACCOUNTING

The intellectually honest position on AI’s economic impact in 2026 is neither the utopian narrative of transformative productivity nor the dismissive claim that it is entirely without merit. AI is delivering verifiable returns in specific, well-defined applications such as financial services automation, healthcare document processing, high-volume data analysis, and particularly for organisations with the data maturity and technical infrastructure to deploy it effectively. That group is however quite small.

For the broader enterprise economy, the evidence suggests that AI is, at present, a significant cost centre with uncertain and often sub-threshold returns. The absence of any visible price deflation in goods and services, which is the most direct market signal of genuine efficiency gains, is perhaps the most telling indicator of where the technology actually sits in its economic maturity curve.

The opportunity cost argument, meanwhile, cannot be dismissed as romantic anti-technology sentiment. It is a legitimate macroeconomic concern backed by data on capital concentration, declining investment in foundational human needs, and the historical evidence that broad-based infrastructure investment of roads, power grids, clean water, nutrition, education, provides clear certainty these generate more durable and equitably distributed economic growth than technology revolutions that concentrate value at the top.

For African states regularly espousing the “Leap-Frog” potential of AI developments in Africa, this is more a matter of hopeful thinking than a strategy based in reality. Large infrastructure and energy gaps combined with connectivity black-holes, makes this even a harder prospect to embrace logically, despite individual successes such as the many fintech startups on the continent that have succeeded.

The reckoning is not inevitable. But for companies, investors, and policymakers, the window for asking, and honestly answering the hard questions about AI’s real returns is closing faster than the capex commitments suggest anyone is prepared for.

Sources:

IBM CEO Study Q1 2025; MIT GenAI Divide Report 2025; S&P Global AI Abandonment Data 2025; McKinsey Global AI Survey 2024; PwC Global AI Jobs Barometer 2025; Goldman Sachs AI Capex Analysis; AI Now Institute 2025; Global Ag Tech Initiative Capital Allocation Report 2025; FAO Food Price Index 2025; Gartner Hype Cycle 2024; Federal Reserve AI Competition Report 2025.