The great AI ROI reality check: big investments, elusive returns.

Executives are witnessing an explosion in AI spending, but measurable returns remain frustratingly scarce. Global expenditures on AI are projected to reach roughly $2.6 trillion in 2026, a 47% jump from last year ([1]). Corporate AI budgets have doubled as a share of revenue, and 94% of companies say they’ll keep investing even without immediate payback ([2]). Yet so far, the vast majority of organizations have little to show on their bottom lines.

New data highlights how rare real ROI from AI still is. MIT found that 95% of enterprise generative AI projects in recent years failed to demonstrate any measurable financial benefit within six months of launch ([3]). Similarly, only about 5% of companies are managing to achieve truly substantial returns from AI at scale ([4]). A PwC survey of global CEOs in 2026 showed 56% admitted getting “nothing” out of their AI investments to date, while a mere 12% achieved both increased revenue and reduced costs from AI – the select few they dubbed the “vanguard” ([5]). In short, tens of billions are being spent on AI, but for most enterprises the payoff is either modest or nonexistent so far.

Part of the issue is timing and expectations. Many leaders assumed AI projects would deliver fast returns, but reality is proving otherwise. In a Deloitte study, most organizations reported it takes 2–4 years for an AI use case to pay back, far longer than the typical 7–12 month payback period for traditional IT investments ([6]). More than half of investors recently said they expect positive ROI from new AI initiatives within just six months ([7]) – a timeframe 84% of CEOs say is unrealistic ([8]). This mismatch is contributing to rising skepticism and a realization that AI is not a magic quick fix for growth.

Facing this ROI gap, boards and CFOs are losing patience with AI experiments that don’t deliver. In a mid-2026 survey, 92% of CFOs said they feel personal pressure to prove their AI investments are paying off ([1]). Finance chiefs are caught in a bind: they’re expected to fund AI to stay competitive but also to prevent waste. As one tech CFO put it, “You can’t accurately calculate ROI if you don’t know your costs” ([2]).

Now the mandate from the top is clear – show concrete value or face cutbacks. Two-thirds of corporate boards say they will only greenlight new AI spending with hard evidence of returns, and 22% have halted any new funding until existing projects prove their worth ([3]). Nearly half of finance leaders are already being asked to quantify AI’s bottom-line impact – a number 43% of them admit they simply don’t have yet ([4]). The message to AI project teams is unmistakable: the days of blank checks for “digital transformation” without accountability are over.

The challenge, CFOs say, is that traditional metrics and accounting methods struggle to capture AI’s benefits (or lack thereof). In one report, 71% of finance chiefs conceded that standard financial frameworks fail to measure the value of AI and other emerging tech investments with long-term or intangible benefits ([5]). This is driving a search for new ways to evaluate AI payoffs. For example, OpenAI’s CFO recently proposed a ‘useful-intelligence-per-dollar’ scorecard that tracks how many high-quality tasks AI systems complete for each dollar spent ([6]). The intent is to shift the focus from vanity metrics (like number of models built or user adoption rates) to tangible outcomes and cost-effectiveness. C-level leaders increasingly want AI initiatives measured by real business value delivered, not just technical achievements.

Another factor driving the new realism around AI is cost. The true expense of AI projects often far exceeds initial plans, thanks to hidden costs and overwhelming usage. In a recent study, 45% of organizations said that unforeseen AI spending overruns had to be escalated to the board level for review ([1]). A full 25% of enterprises even delayed or canceled at least one AI initiative in the past year because costs blew past expectations ([2]).

Why are AI budgets spiraling out of control? Technology costs are fragmenting across cloud infrastructure, data pipelines, vendor APIs, and internal talent, making it hard to track where the money is going ([3]). As one tech CEO noted, AI has turned once predictable IT spending into something “dynamic, distributed, and increasingly difficult to attribute” ([4]). This lack of cost visibility means many companies are essentially flying blind on ROI, only realizing the full cost of an AI initiative after budget surprises force tough conversations.

Paradoxically, even as some AI tools get cheaper, overall costs can rise. Fierce competition among AI providers has driven down unit prices – for instance, enterprise AI usage costs per unit (per million words processed) dropped ~43% between May and August 2026 amid a price war and more open-source model options ([5]). OpenAI even slashed certain AI API prices by up to 80% this summer ([6]) ([7]). But new "agentic" AI systems often perform complex, multi-step tasks that consume far more computing resources. Gartner analysts observe that the cost per completed task is actually increasing as companies deploy these autonomous AI agents, potentially “eclipsing the financial savings they were designed to generate” ([8]). In short, using AI at scale isn’t cheap – and the bill can grow faster than the benefits unless organizations plan carefully.

The cracks in the business case for AI are now visible in some high-profile retreats. One striking example is Salesforce’s much-hyped “Agentforce” project: launched in late 2025 as a flagship AI-powered product, it was hailed by the CEO as transformational. Yet within months, four senior executives leading the initiative departed, and Salesforce quietly laid off nearly 1,000 staff in early 2026 – including much of the Agentforce team ([1]). The culprit wasn’t the technology itself, but a lack of clear ROI. Leadership found that bold promises to “replace” work with AI had not translated into measurable business gains, leading to a strategic pullback.

Even the world’s tech giants have hit unexpected barriers. Uber, for instance, gave 5,000 developers access to an AI coding assistant, only to see usage explode by 80% and the entire annual AI budget exhausted in just four months ([2]). Uber’s COO told investors that the surge in cost was “hard to justify” because the link between AI expenditures and business value was still unclear, despite 70% of the company’s code now being machine-generated ([3]). Uber has since imposed strict per-employee caps on AI usage to rein in spending. Microsoft – one of the biggest AI proponents – reportedly went as far as canceling some internal AI tool licenses to control costs ([4]).

Across industries, many early AI deployments have stumbled due to non-technical shortcomings. In healthcare, AI diagnostic systems that performed well in tests proved useless without high-quality patient data integration. Financial firms found AI trading models that excelled on historical data could not cope with real market volatility. Retailers tried AI-driven demand forecasting tools, only to discover they were no more accurate than the spreadsheets they replaced ([5]). In case after case, the technology’s potential was real, but organizations lacked the data infrastructure, process changes, and upskilled workforce to translate AI into profit. Gartner projects that 60% of AI initiatives will ultimately be abandoned by 2026 for precisely these reasons – not because the AI can’t work, but because companies weren’t ready for it ([6]).

If there is a silver lining, it’s that a small minority of companies are demonstrating how to unlock genuine value from AI. Research shows that roughly 5% of enterprises are “high performers” getting significant ROI from AI, while another 30–35% see only partial returns ([1]). What sets the winners apart? They treat AI as a strategic, business-wide transformation rather than a series of isolated experiments. A 2026 global survey found that 74% of AI’s total economic benefits are being captured by just the top 20% of companies, which are redesigning processes and offerings around AI for growth, not just cost savings ([2]). Among these leaders, 44% have applied AI to reimagine their products or customer experience, compared to only 17% of other firms ([3]).

Successful organizations also invest early in the foundations that make AI pay off. They ensure data is high-quality, accessible, and ready for AI use, and they integrate AI into core operations with executive championing. In the words of one insurance CIO who has delivered consistent ROI from AI, “the value question, the ROI, was very top of mind” from day one ([4]). By prioritizing projects in areas that are “AI-ready” – where data, systems, and skills are in place – and designing solutions for reuse across the business, his team was able to fund new AI initiatives from the savings of earlier wins ([5]). In another example, the CIO of a cybersecurity firm targeted AI at internal processes with clear metrics: automating 90% of IT operations boosted speed and efficiency so much that it cut IT operating costs in half ([6]).

Finally, leading companies focus on measuring what matters. They set clear KPIs for AI projects tied to business outcomes like revenue growth, cost reduction, customer satisfaction, or risk mitigation – and they halt investments that aren’t hitting those metrics. To guide this discipline, some firms are adopting new ROI frameworks. For instance, OpenAI’s CFO proposes tracking “cost per successful AI-driven task” as a core metric, shifting the conversation from “cost per model or per query” to the actual business value generated per dollar spent ([7]). By embracing a more rigorous, outcome-centric approach – and by aligning AI initiatives with strategic goals – executives can turn the current AI hype into sustainable business results.