AI investments soar, but where's the ROI?

In the past two years, surging enthusiasm for generative AI has fueled an unprecedented surge in corporate investment. Many organizations treated AI as a must-have initiative and poured money into new models and infrastructure at a record pace. In some cases, the usual cost-benefit checks on tech spending were bypassed – analysts note that exuberance around large language models created a permissive environment for AI spend, sidestepping normal ROI scrutiny ([1]).

This fervor has driven enterprise AI spending to astonishing levels. Gartner now projects global AI expenditures will reach $2.59 trillion in 2026, a 47% increase over 2025 ([2]). Internally, companies are allocating massive budgets to AI; one survey found 59% of firms spend more than $1 million per year on AI initiatives ([3]).

Yet despite this heavy investment, actual returns remain scarce. Across industries, only a tiny fraction of organizations are realizing substantial value from AI. In PwC’s latest CEO survey, just 12% of chief executives said their AI programs have delivered both higher revenues and lower costs so far ([4]) – meaning nearly 90% have not seen a tangible dual benefit. Meanwhile, an MIT study recorded a 95% failure rate for enterprise generative AI projects (i.e. 95% of pilots produced no measurable financial returns within six months of launch) ([5]).

Paradoxically, lackluster results have not fully stemmed the tide of spending. Nearly 40% of companies that targeted double-digit cost reductions from AI realized under 10% in actual savings, yet 90% plan to boost their AI budgets again ([6]). A potent mix of competitive pressure and FOMO (fear of missing out) is still driving many to invest first and hope the ROI will follow.

As the bills come due, the economics of AI are facing intense scrutiny. In 2026, providers like OpenAI and Anthropic shifted from flat-rate plans to usage-based pricing, turning every AI query (each token of text) into a metered cost ([1]). Suddenly, every prompt an employee sends to an AI system shows up itemized on the balance sheet ([2]). This abrupt transparency has shattered the illusion that AI comes free, forcing finance chiefs to confront how fast “cheap” experiments can burn cash in practice.

During the height of the AI boom, many companies rushed to put generative AI into the hands of every employee ([3]), assuming that more usage would automatically yield more value. Many discovered this was a costly mistake. In one extreme case, an enterprise without proper cost controls was hit with a $500 million cloud bill in a single month due to runaway AI usage ([4]).

Fearing similar overages, even tech giants have tapped the brakes on internal AI rollouts. Uber, for example, reportedly blew through its entire 2026 AI budget by April and has since capped employees’ monthly AI usage at $1,500 ([5]). Walmart similarly throttled its staff’s use of an internal coding AI tool after demand skyrocketed, and Amazon’s leadership warned engineers not to use AI without a clear purpose after discovering employees racking up costs on frivolous chatbot queries ([6]). A June Financial Times report noted that companies from Amazon to Meta have implemented such measures as they quietly scale back their plans to use AI everywhere after last year’s free-for-all ([7]).

All of this is driving a fundamental shift in how organizations evaluate AI projects. The key metric is no longer cost-per-model or cost-per-token, but cost-per-outcome – tying spending directly to tangible results ([8]). In fact, some firms have even postponed roughly 25% of their planned AI investments to 2027 amid this new financial discipline, reexamining business cases more carefully before allocating further funds ([9]).

External stakeholders have growing doubts about the payoff from AI. Public markets have seen the most aggressive AI-spending tech companies underperform both the broader tech sector and even the AI infrastructure providers benefiting from the boom ([1]). Investors have de-rated some of these heavy AI spenders, reflecting skepticism that today’s large investments will translate into commensurate profits any time soon.

Company boards are likewise applying new pressure for results. One recent study found that 75% of corporate boards had approved major AI projects, yet fewer than half had established metrics or governance to track AI success ([2]). That hands-off approach is disappearing fast. Directors who were eager not to miss out on the AI wave are now far less willing to authorize open-ended spending without evidence of business impact.

This means the C-suite faces heightened accountability for AI outcomes. In one survey, 71% of CIOs said they had until mid-2026 to demonstrate real business value from AI initiatives or face budget cuts (and even potential job loss) ([3]). Advisors are urging executives to manage AI like any other strategic investment: set clear objectives and KPIs from the start, measure results rigorously, and be prepared to scale back or terminate projects that are not delivering value ([4]).

Why have so many AI efforts failed to deliver? A major factor is how these projects were set up. Many early initiatives were exploratory pilots with no clear link to operational needs ([1]). As one AI leader observed, companies often approached early AI projects as 'learning opportunities with little or no relevance to the business' – so even promising proofs of concept eventually withered on the vine without ever scaling into production ([2]).

Even well-intentioned AI projects hit internal roadblocks. Some organizations found their data and processes were not ready for AI at scale – in fact, modernizing legacy systems or cleaning up data often cost more time and money than the AI’s anticipated benefits ([3]). Without the necessary data infrastructure, integration capabilities, and cross-functional ownership, many initiatives that showed potential in pilot never gained traction or delivered meaningful ROI.

Notably, these failures usually weren’t due to problems with the AI technology itself, but rather organizational shortcomings. Industry analysts have dubbed this the 'learning gap' – the chasm between deploying an AI model and actually getting value from it in everyday operations. An MIT study found roughly 95% of generative AI pilots did not produce measurable financial returns, primarily because companies failed to redesign workflows and upskill staff to integrate the AI effectively ([4]).

Ironically, even when AI does create real improvements, companies often struggle to quantify them. In an IBM survey, only 29% of executives said they could measure the ROI of their AI projects confidently – even though 79% reported seeing productivity gains from those same initiatives ([5]). This inability to connect operational benefits to concrete financial outcomes leaves business leaders skeptical and makes it hard to justify scaling up AI pilots into full production.

Amid the current AI reality check, a few organizations have managed to capture tangible returns. One example is Massachusetts Mutual Life Insurance (MassMutual), which shifted from experimentation to execution by focusing on high-impact use cases and weaving AI into core processes across the enterprise. The 175-year-old insurer reports concrete performance gains – from 30% faster software development cycles to IT helpdesk resolution times dropping from 11 minutes to about 1 minute, and customer service calls shortened from roughly 15 minutes to 2 minutes ([1]). Another frontrunner, Mass General Brigham, likewise credits a disciplined, results-first AI strategy for measurable improvements after curbing ungoverned pilot projects and scaling up those that proved their value ([2]).

These success stories offer a blueprint for bridging the gap between AI investment and outcomes. The hallmarks of AI winners include treating AI as a strategic, enterprise-wide priority championed by leadership – not as an isolated R&D experiment ([3]). They address foundational enablers early on by ensuring high-quality data, strong governance, and reengineered workflows so that AI tools fit seamlessly into daily operations, rather than simply throwing money at the latest algorithms and hoping for results ([4]).

Equally important, top-performing companies enforce rigorous measurement and accountability for AI initiatives. They define clear metrics (such as hours saved, cost reductions, or revenue generated) for each project and track them with simple AI value scorecards to verify that each deployment delivers more value than it costs ([5]). By scaling up the initiatives that meet their ROI targets – and shutting down those that fall short – these organizations ensure that AI truly boosts the bottom line.