Big AI Bets, Small Returns: A Reality Check for ROI

Generative AI has dominated boardroom conversations and tech budgets over the past year. Companies across finance, healthcare, retail, and other sectors rushed to invest in AI-powered solutions, keen to streamline operations and outpace competitors. But halfway through 2026, many of these organizations are confronting a hard truth: the financial returns from their ambitious AI bets remain scarce, even as spending reaches new heights.

Multiple new surveys underscore this widening ROI gap. A global poll of 4,500 CEOs found that 56% of organizations saw no significant financial benefit from their AI investments in the last 12 months ([1]). Similarly, analysis by MIT indicates 95% of corporate AI projects have yet to deliver any tangible returns ([2]). In short, while nearly every large firm is experimenting with AI, the majority are still waiting for meaningful payoffs in revenue or efficiency.

This disconnect between investment and outcome is forcing a rethink. AI’s value appears to take much longer to materialize than typical tech projects: one study found most AI use cases require two to four years to achieve satisfactory ROI, compared to under 12 months for average IT investments ([3]). For impatient stakeholders expecting quick wins, these extended timelines have come as an unwelcome surprise, fueling growing skepticism about the true business case for AI.

One reason returns are lagging: the cost of deploying AI systems is much higher than many executives anticipated. New data shows that 84% of enterprises say surging AI infrastructure expenses – from cloud computing and GPUs to software licenses – are already hurting their profit margins by more than 6% ([1]). Even more worrying, 80% admit they can’t accurately forecast these costs, with nearly a quarter underestimating AI spend by over 50% ([2]). In many cases, AI was billed as a great cost-cutter; instead, unplanned bills for computing power and data are eating into savings.

These hidden costs have led to some high-profile wake-up calls. At Uber, for example, giving 5,000 developers access to generative AI coding assistants drove a massive spike in usage – and expenses. Uber exhausted its entire 2026 AI budget by April, just four months into the year ([3]), after engineers enthusiastically embraced tools like Anthropic’s Claude. By March, 95% of Uber’s engineers were using AI to write code, and roughly 10% of all new code was machine-generated ([4]). The result? The company’s AI-related R&D costs ballooned 17% year-over-year to $951 million in Q1 ([5]), overshooting all projections.

As Uber’s President and COO, Andrew Macdonald, put it, the connection between these runaway AI costs and real business outcomes "is not there yet" ([6]). Uber’s leadership is now openly questioning whether the skyrocketing "token" bills for AI tools are delivering enough value to justify themselves ([7]). Even tech giants are pulling back to control spending: in May, Microsoft instructed thousands of its engineers to stop using a costly third-party AI coding system and switch to the company’s own GitHub Copilot tool – a move seen as reining in an expensive new AI habit ([8]). Across industries, finance chiefs are coming to realize that without proper guardrails, supposed “efficiency” tools can quickly rack up unwieldy expenses.

The ballooning costs and unclear returns of AI have not gone unnoticed in executive suites. In fact, 61% of senior business leaders now feel significantly more pressure to prove returns on AI investments than they did a year ago ([1]). After years of enthusiasm about AI’s transformative potential, CFOs and boards are increasingly impatient for evidence of business impact – or at least a credible plan to achieve it.

The investor community’s patience is wearing thin as well. Meta’s stock, for example, suffered its worst drop in three years after the company announced a major increase in AI spending without a clear road to profitability, sparking fears of an "AI bubble" among shareholders ([2]). This turn in market sentiment sends a clear message to other firms: simply pouring money into AI – without demonstrated ROI – is no longer being rewarded. As one industry insider put it, companies feel compelled to invest in AI ("if we do not do it, someone else will – and we will be behind"), but that logic is wearing thin ([3]) as stakeholders demand to see real results.

In response, business leaders are shifting from a fear-of-missing-out mindset to a more hard-nosed approach to AI expenditures. Analysts note that many CFOs are reevaluating AI projects with a more critical eye and more nuanced metrics ([4]). A key challenge is measurement itself – more than a third of global executives (39%) say that quantifying AI’s business impact is one of their primary hurdles ([5]), since many of AI’s benefits so far are indirect (for example, better decision-making or faster processes that don’t immediately show up on the balance sheet). To address this, forward-looking CFOs are expanding their definition of “return” to include intermediate benefits like improved customer satisfaction or innovation capabilities. Still, they are also making it clear that projects which fail to demonstrate tangible progress toward those goals will face tough scrutiny or funding cuts.

Amid the hand-wringing, a small group of companies is proving that AI can deliver significant value – and their approach offers a blueprint for others. Only about 5% of enterprises today achieve substantial ROI from AI at scale ([1]), but these “AI vanguard” organizations share common traits. They forged strong data and technology foundations and deploy AI broadly in ways that align with their core business strategy ([2]), rather than siloed experiments. In fact, 44% of these high-performers have successfully embedded AI into their products, services, and customer experiences, versus just 17% of other firms ([3]).

Crucially, the leaders design their AI initiatives with clear value metrics from day one. Unlike peers who toss around vanity metrics (one company touted “19 trillion tokens processed” with "zero GAAP impact" to show for it ([4]) ([5])), the top performers establish concrete baselines and key performance indicators (KPIs) before deploying AI. This practice ensures that every algorithm and automation is tracked against business outcomes – whether it’s boosting revenue, cutting cycle time, or improving service quality – instead of mere activity counts.

Finally, winning organizations manage AI as a portfolio of investments rather than betting everything on a single moonshot. As Gartner analysts advise, success comes from balancing quick, tangible wins with longer-term strategic bets ([6]). High-ROI enterprises iterate quickly: they scale up pilots that show measurable results and aren’t afraid to shut down projects that don’t move the needle. By coupling ambitious vision with disciplined execution and measurement ([7]) ([8]), these companies are starting to turn AI’s promise into real performance improvements – and offering a path forward for others to follow.