AI gold rush meets ROI reality: hard truths for business leaders.

While ROI remains elusive, the costs of AI are very real – and rising. The computing power and data needed to run advanced AI models don’t come cheap. At Nvidia, a senior AI leader notes that his team’s cloud-compute costs now exceed their entire payroll expense ([1]). In other words, it costs more to run the algorithms than to pay the people who develop them – a startling shift in cost structure that seemed unthinkable just a few years ago.

AI services purchased from third parties are also straining budgets in unexpected ways. Usage-based pricing means every query or API call to a large model racks up fees, making costs grow faster than anticipated and harder to forecast ([2]). Uber’s chief technology officer reportedly blew through the ride-hailing company’s 2026 AI budget before the first quarter even ended, thanks to a surge in spending on AI coding tools ([3]). Such surprises have CFOs and boards on high alert.

The era of unchecked AI spending is ending as finance leaders impose a new discipline. According to a Deloitte 2026 CFO survey, 68% of CFOs now refuse to approve additional AI funding without hard ROI evidence, and 42% have already cut pilot projects that couldn’t justify their costs ([4]). After years of exuberant investment, we’re seeing a fiscal backlash – call it an AI austerity movement – aimed at separating genuine value from expensive experimentation.

Even AI vendors themselves are facing economic reality. OpenAI, the industry’s poster child, reportedly shut down a costly new product in March after it was found to be burning $1 million a day with no viable business model ([5]). Rival firms are slashing prices in response to customer concerns about cost: Anthropic, for example, recently cut its AI model pricing to improve cost-efficiency as usage soared ([6]). With investors scrutinizing how much useful output each model provides per dollar spent ([7]), the battle among AI providers is shifting from just model power to one of economic efficiency. For enterprise buyers, the message is clear – the "cheap to experiment, expensive to scale" nature of AI means careful cost management must be part of any AI strategy.

If AI is so promising, why do so many initiatives under-deliver? One reason is a lack of clear purpose. Executives often felt pressured to pursue AI for its own sake, leading to projects that weren’t solving a defined business problem. Nearly two-thirds of CEOs concede that FOMO – the fear of missing out – drove them to invest in AI without a strong use case or strategy in place ([1]). Leaders eager to proclaim themselves "AI-first" sometimes applied the technology to peripheral or low-impact problems, like trivial chatbots or content generators, instead of mission-critical needs ([2]).

A second factor is insufficient preparation. In the rush to “move fast,” organizations skipped foundational steps such as cleaning up data, training staff, and reengineering processes to integrate AI. Many dived in without the necessary internal expertise and without getting buy-in from end users, which one expert calls a “reckless approach” that leads directly to wasted resources and disappointment ([3]). In short, they underestimated the complexity – treating AI as a plug-and-play tool rather than a transformational project requiring significant change management.

Even when there is a solid plan, scaling AI from pilot to production has proven notoriously difficult. Many companies find their early experiments get stuck in a perpetual pilot phase – known as 'pilot purgatory' – and never translate to enterprise-wide impact. A recent analysis found that while 80% of organizations had experimented with popular generative AI tools like ChatGPT, only 20% progressed to an actual pilot deployment and a shockingly low 5% made it to a full production rollout ([4]). The rest fell by the wayside, often due to brittle workflows, lack of contextual learning (AI systems that don’t improve over time), and poor integration into day-to-day operations ([5]). These issues – not the algorithms themselves – are the common culprits behind AI’s failure to deliver on its promises.

Recognizing these pitfalls, some companies are changing tack. The fad of 'move fast and break things' is giving way to a more measured approach ([6]). Rather than launching dozens of disconnected AI pilots and hoping one strikes gold, leaders are starting to demand clear problem statements, business alignment, and validated outcomes from AI initiatives before scaling them up. The goal is to break out of the cycle of endless experimentation and start turning AI into a reliable driver of business performance.

Despite the overall challenges, a few enterprises are actually realizing strong AI returns – and they offer a blueprint for others. In a recent study of global AI performance, PwC found that roughly 20% of companies are achieving about 75% of all the measured financial benefits from AI, while the other 80% see only modest gains or none at all ([1]). Crucially, these leading organizations are not necessarily those spending the most on AI or using the fanciest algorithms, but those who align AI efforts tightly with business strategy.

One key differentiator is an emphasis on growth. AI 'leaders' focus on using AI to drive revenue and innovation, not just to cut costs. Many companies default to pitching AI as a productivity tool to trim budgets, but savings from small efficiencies often get absorbed and never hit the bottom line. By contrast, high-performing firms use AI to create new products, reach new customer segments, and accelerate sales cycles – generating new income streams where success is easier to measure and directly contributes to growth ([2]).

Another success factor is rethinking roles and processes to fully leverage AI. Rather than simply bolting AI onto existing workflows, these organizations redesign jobs and operations around what AI does best. For example, some companies have quintupled the productivity of customer support by turning agents into AI supervisors who manage AI-driven responses, redeploying the freed-up human capacity to higher-value tasks. This kind of deep change can be uncomfortable and requires retraining and cultural shifts, but it’s often the only way to get transformative results from AI ([3]).

Strong governance and cross-functional collaboration also set the winners apart. Top AI performers typically have joint ownership of AI initiatives between technology leaders (like CIOs and CTOs) and business leaders (such as business unit heads, COOs or CFOs) ([4]). This ensures AI projects are chosen and executed with both technical feasibility and business impact in mind. Companies that leave AI entirely to the IT department often end up with interesting prototypes that don’t move the needle on revenue or strategy.

Finally, the leading companies insist on meaningful metrics for AI. Instead of counting the number of models built or pilot programs launched, they measure success in terms of business outcomes – additional revenue attributed to AI features, improvements in profit margins, time saved, or customer retention gains ([5]). Snowflake provides an example: by using an AI-driven coaching platform for sales teams, the cloud-data company saved over 1,200 manager hours per quarter, translating to roughly $700,000 in annual savings and a four- to five-fold return on investment for that project ([6]). By defining clear objectives and tracking tangible results, these organizations turn AI from a flashy experiment into a real value generator.