AI Spending Soars, ROI Falls Short

Companies are pouring money into artificial intelligence at historic rates, betting on big future payoffs. Global AI spending is projected to reach a staggering $2.5 trillion in 2026 ([1]). Tech’s four biggest players alone have announced plans for nearly $700 billion in combined AI investments this year, a level that’s already pressuring cash flows as Amazon even risks running negative free cash flow to fund its AI ambitions ([2]). Virtually every industry – from finance to healthcare – now views AI as essential for competitiveness, but this gold rush comes with a crucial question: Is all that spending translating into real business value?

So far, the evidence is sobering. In one recent survey of IT leaders, only 28% of AI projects in core operations achieved their expected return on investment, while roughly 20% failed to deliver any meaningful result ([3]). Meanwhile, a global CEO study found that more than half of companies have seen no measurable lift to revenues or cost savings from their AI initiatives so far ([4]). Just one in eight firms reports attaining both higher revenue and reduced costs through AI – a sign that truly transformative ROI remains rare. In short, the money is flowing, but the results are often lagging behind.

This widening gap between investment and impact is starting to chip away at executive confidence. PwC’s latest global CEO survey recorded a five-year low (around 30%) in business leaders’ optimism about revenue growth, with uneven AI returns cited as a key factor ([5]). In other words, as early enthusiasm gives way to impatience, the era of large-scale AI experiments without clear line-of-business outcomes is drawing to a close.

One reason for this shift in sentiment is that the cost side of the AI equation has become impossible to ignore. Developing and deploying advanced AI – especially generative AI – can rack up enormous cloud and infrastructure bills. In extreme cases, expenses have spiraled out of control: one AI consultant revealed that a large enterprise client accidentally incurred a $500 million charge in a single month on Anthropic’s Claude model by allowing unlimited employee usage ([1]). A well-intentioned AI rollout turned into a budgetary catastrophe virtually overnight due to basic failures in cost oversight.

Even more deliberate AI adoption can strain budgets. Uber, for example, recently blew through its entire annual AI allocation in just four months after encouraging staff to 'use AI as much as possible' ([2]). The overspend forced Uber’s leadership to slam on the brakes, imposing a firm $1,500 per-user monthly cap on certain AI tools to contain the overruns ([3]). Tellingly, Uber’s COO remarked that it was 'very hard to draw a line' from all that AI activity to any new customer features or efficiency gains ([4]).

These incidents highlight a broader problem: AI investments often generate hefty expenses long before they deliver savings. Even technology giants with deep pockets are not immune to this imbalance. Amazon reportedly shut down an internal programme that gamified employee AI usage after workers gamed the system without improving output ([5]). And while cloud titans like Alphabet, Microsoft, Meta and Amazon are plowing unprecedented sums into AI capabilities, investors have grown more selective about footing the bill. Goldman Sachs analysts note that as AI capital expenditures for 2026 keep climbing, markets are rewarding only those companies that can tie their rising AI spend to clear, near-term results – and penalising those that can’t ([6]).

The days of open-ended AI projects are numbered. With growing evidence of underwhelming returns, corporate boards and finance chiefs are enforcing much stricter oversight of AI investments. In a recent Deloitte survey, 68% of CFOs said they will not approve further AI spending without demonstrated ROI, and 42% have already pulled funding from projects that fell short ([1]). In practice, AI teams must now show concrete business impact – whether via revenue gains, cost reductions or efficiency improvements – if they want to keep their projects alive.

Industry watchers are calling this the 'AI accountability' era. The long run of experimental AI pilots with vague promises of transformation – often referred to as 'pilot purgatory' – is coming to an end ([2]). Gartner experts advise managing AI as a portfolio of bets: some aimed at quick wins, others at longer-term breakthroughs – rather than expecting one uniform ROI formula for every project ([3]). As one Gartner analyst cautioned earlier this year, 'the improved predictability of ROI must occur before AI can truly be scaled up by the enterprise' ([4]).

Shareholders are likewise ramping up the pressure. When a major tech company recently announced an eye-popping increase in AI investment, its stock price stumbled amid questions about when those bets will translate into profits ([5]). Simply put, investors and boards are no longer willing to subsidise AI for its own sake – they expect to see a credible business case. This shift is forcing tough decisions about which AI programs get the go-ahead, which are paused or cut, and how to rigorously measure success.

Amid the current reckoning, a few frontrunners are proving that meaningful AI ROI is attainable – but only with the right approach. Roughly 5% of enterprises are already realizing outsized value from AI initiatives ([1]). These leaders pair technology deployment with organisational change from day one. By contrast, the average AI program directs 93% of its budget to technology and only 7% to training people and redesigning processes ([2]). Top performers flip that script, investing in solid data infrastructure while heavily upskilling staff and reworking workflows. This ensures AI outputs translate into business outcomes, rather than languishing as idle 'science projects'.

Another key to success is redefining how results are measured. Leading adopters are abandoning raw usage stats – the kind of 'vanity metrics' one banking executive derided as counting billions of tokens per day – and focusing on core operational KPIs like productivity, quality and customer experience ([3]). At BNP Paribas, for example, Chief AI Officer Charles Holive asks teams, 'What can we do now that we couldn’t do before? How much faster can we do it?' ([4]) This shift from counting AI activity to tracking real performance ensures every AI experiment aligns with strategic business goals.

Finally, successful companies stay disciplined about where they apply AI. Instead of trying to automate everything at once, they concentrate on a few high-potential use cases where AI clearly boosts revenue or cuts costs. As G-P COO Nat Natarajan advises, leaders must move past the hype and zero in on areas where AI truly 'moves the needle' ([5]). He further notes that a smart AI strategy isn’t about doing everything simultaneously – it’s about starting with achievable projects and preparing the organisation to leverage them ([6]). Yet many firms still rush ahead without the proper guardrails: only 21% of companies using advanced AI agents have a mature governance model in place ([7]). By building strong governance and tying every AI initiative to specific business outcomes, executives can convert today’s AI excitement into sustainable, measurable performance gains.