Key Takeaways
- Global AI spending is projected to reach $2.59 trillion in 2026, a 47% increase over 2025, per Gartner’s May 2026 forecast, which revised upward from its January estimate of $2.52 trillion and 44% growth.
- Stanford and BetterUp researchers found that low-quality AI-generated content costs organisations roughly $186 per employee per month in lost productivity, with recipients spending nearly two hours per incident correcting or redoing outputs.
- FinOps teams’ share of responsibility for AI costs grew from 31% in 2024 to 98% in 2026, exposing how unprepared traditional financial governance is for consumption-based billing at agentic scale. Global AI spending is heading toward $2.59 trillion in 2026, yet McKinsey’s 2026 State of AI survey puts the share of organisations that can show a positive EBIT impact at just 37%, unchanged from 2025. The gap between what enterprises are committing and what they can demonstrate is widening.
Record Spend, Flat Returns
Infrastructure alone is forecast to add hundreds of billions this year. That capital is not translating cleanly into returns, and the reasons are structural.
Token-based billing, agentic workflows, tool calls, retries and multimodal reasoning mean a single user request can trigger multiple model calls. Finance teams built for traditional cloud budgeting are poorly equipped to govern consumption at that granularity. As boards demand proof of returns on record AI expenditure, the measurement gap is becoming a governance problem. A 2H 2026 CIO survey by The Futurum Group found that 46.9% of enterprises are running AI spend over budget, leaving usage charts climbing while productive output for the enterprise stagnates.
The Productivity Paradox
McKinsey’s survey, which polled over 1,700 respondents across 97 countries, found that 80% of individual users report feeling more productive with AI tools. Deloitte‘s 2026 report puts the share of organisations reporting productivity and efficiency gains at roughly two-thirds. Yet only 37% can attribute a positive EBIT impact to AI, and just 6% of companies qualify as high performers, attributing 5% or more of EBIT to AI.
The gap between individual experience and firm-level outcomes is not a measurement artefact. An NBER study of 6,000 CEOs, CFOs and senior executives found that between 89% and 95% of firms saw no measurable impact on productivity or employment over the prior three years. AI tools demonstrably speed up specific tasks; the organisational machinery to convert those micro-efficiencies into P&L impact is absent in most enterprises.
More than half of enterprises, 52%, have actively deployed AI agents, with 39% running more than 10 agents simultaneously, scale that has outpaced the financial tooling built to track what those agents cost to run.
The Cost of “Workslop”
Stanford and BetterUp researchers identified a compounding drain they call “workslop”: AI-generated content that is unhelpful, low-effort or low-quality. Recipients spend nearly two hours per incident deciphering, correcting or redoing such outputs, costing organisations roughly $186 per employee per month in lost productivity. At scale, that runs to millions in wasted time annually before any infrastructure cost is counted.
Per-token pricing has fallen across several models, but overall AI bills have not followed. Cost growth is driven by architecture rather than unit price: the complexity of agentic workflows, tool calls, retries and multimodal reasoning compounds faster than per-token savings accumulate. That same complexity makes attribution harder, which is why the pattern rhymes with warnings about AI infrastructure spending amplifying systemic financial risk when returns disappoint.
The Solow Parallel
The current picture echoes the Solow Paradox of the 1980s, when massive IT investment failed to show up in productivity statistics for years before the curve bent upward in the 1990s. Some economists anticipate AI may follow a similar, compressed timeline, with current spending representing upfront infrastructure cost and the payoff arriving in 2028 or 2029.
The token economy adds a layer the IT parallel does not cover. FinOps teams’ share of responsibility for AI costs grew from 31% in 2024 to 98% in 2026, per available survey data, a rapid shift that reflects how unprepared traditional financial governance was for consumption-based AI billing. High-performing organisations are redesigning workflows around AI rather than bolting it onto existing processes. That is where the EBIT impact tends to appear, and it tracks with the broader argument that infrastructure investment alone does not generate returns without the operational layer on top.
From Productivity Metrics to P&L
According to PwC research, leading companies are two to three times more likely to use AI to identify growth opportunities and reinvent business models than to focus solely on efficiency. Companies that separate cost-reduction ROI from revenue-generation ROI tend to achieve stronger results and close faster on investment decisions, the distinction between optimisation and redesign that separates the 6% generating measurable EBIT impact from the rest.
Originally published at https://autonainews.com/ai-spending-nears-2-5-trillion-yet-ebit-impact-stalls/
Top comments (0)