Big AI investments, elusive returns: a reality check.
Big spending, low returns.
AI budgets are soaring, but returns are lagging. A recent survey indicates companies plan to double their AI spending to roughly 1.7% of revenues in 2026, and nearly all say they will keep investing even without immediate returns ([1]). The generative AI craze has also spread across industries: more than 80% of enterprises have piloted tools like ChatGPT and Copilot, with almost 40% already deploying them in some form ([2]).
Yet for all this activity, tangible ROI remains elusive for most. An MIT study found 95% of organizations have seen no measurable financial return from their generative AI projects ([3]). Likewise, Gartner reports only 28% of enterprise AI initiatives fully meet their ROI goals, while roughly 20% fail outright ([4]) – the rest produce only limited or unclear benefits.
In fact, at the median, AI projects are losing money. S&P Global research shows a typical AI initiative costs $6.8 million but delivers just $1.9 million in value ([5]). That represents a negative 72% return on investment for the average project – a stark indicator of how far expectations are from reality for most companies.
Facing these sobering numbers, some firms are starting to pull back from hyped AI bets. Salesforce, for example, recently scaled down its 'Agentforce' AI team after internal projects failed to show clear value and several senior AI executives departed ([6]). Across industries, there are signs of an end to "AI for AI’s sake" – organizations are becoming more willing to shut down experiments that are not delivering results, refocusing their resources on use cases with clearer paths to ROI.
The hidden costs of AI.
Even as organizations chase AI’s promise, many are discovering unplanned expenses that eat into returns. Spending on AI infrastructure and compute has skyrocketed – Goldman Sachs projects global AI-related capital expenditure will reach $527 billion in 2026 ([1]). These up-front investments in data centers, specialized chips, and cloud capacity often come long before any payoff, weighing down near-term ROI.
Operating costs are rising too. In regions with heavy data center growth, wholesale electricity prices have jumped 267% over five years ([2]), sparking community backlash and halting nearly $100 billion in planned data center projects in 2025 ([3]). Those surging power and cooling costs flow straight to companies’ bottom lines, eroding the savings that AI initiatives are supposed to deliver.
Meanwhile, the price of advanced AI software and services is climbing. Vendors have imposed steep increases – for example, OpenAI’s latest GPT-5.5 model, while more efficient, came with higher usage fees that caused enterprise customers’ cloud bills to spike by ~40% ([4]). In effect, providers are capturing efficiency gains as higher revenue, turning “cheaper” AI into a pricier undertaking for clients.
These economics are forcing tough "build vs. buy" choices. Many companies initially tried developing AI models in-house, but the tide is turning: 76% of enterprise AI use cases are now handled by external solutions, a complete reversal from 2024 when 47% were built internally ([5]). Creating cutting-edge AI from scratch requires hefty investment and talent, while buying capabilities means relying on third-party pricing and infrastructure. Either way, leaders must weigh total cost of ownership against realistic ROI projections more rigorously than ever.
Boardroom pressure vs. reality.
While returns remain scarce, the C-suite faces intense pressure to justify AI spending amid rampant hype. Board members, eager to seize AI’s potential, are prodding executives to accelerate deployments. Fully 61% of CEOs say their boards are pushing AI projects too fast out of fear of missing out on the technology’s promise ([1]). Many chief executives worry this enthusiasm is outpacing the organization’s readiness and clouding boardroom judgment.
Investors are similarly impatient for quick wins. In one global survey, 53% of investors expected to see positive ROI from AI within six months, whereas only 16% of large-company CEOs believed such rapid payback was realistic ([2]). In truth, meaningful business benefits from AI often take longer: in Deloitte’s 2025 study, just 6% of organizations achieved payback on an AI use case in under a year, with most finding it takes two to four years to realize significant value ([3]).
Financial gatekeepers are responding by tightening oversight. Nearly 70% of CFOs in a 2026 Deloitte survey said they will not approve new AI investments without proven returns, and 42% have already cut funding for initiatives that weren’t delivering results ([4]). Finance chiefs are increasingly involved in vetting AI business cases, setting ROI targets, and even co-leading AI strategy. The message from the board and C-level is clear: the era of open-ended AI experimentation is over – from now on, AI projects must earn their keep.
Closing the AI value gap.
What are successful adopters doing differently to capture real ROI from AI? A new global study by PwC found that 20% of companies are reaping 74% of all AI-derived economic value today ([1]). These frontrunners treat AI as a strategic transformation enabler, not just a tech experiment. They are roughly three times more likely than others to use AI to generate new revenue and to reinvent core business processes around AI, rather than simply adding tools on top of old workflows ([2]).
Crucially, top performers double down on people, not just technology. Gartner observed that companies which eliminated jobs to “make room” for AI saw no improvement in outcomes – their results mirrored those of peers that kept headcount stable ([3]). By contrast, organizations that invested in upskilling employees and integrating AI into roles saw far greater benefits. As one Gartner analyst put it, layoffs might free up budget but do not create returns; real gains come when firms 'aggressively invest in skills, new roles and operating models' that allow humans to work effectively with AI ([4]). In fact, companies that empower employees with AI are achieving about double the profit margin growth of those fixated on labor cost reduction ([5]).
Finally, capturing value from AI demands patience and fundamental change. Smart leaders are fortifying data quality, modernizing IT infrastructure, and fostering cross-functional collaboration to support AI at scale. Adopting AI is more like the shift from steam power to electricity than a typical software upgrade ([6]) – it requires rethinking processes and retraining people before the full benefits appear. By setting clear metrics and focusing on high-impact use cases, executives can translate AI’s potential into sustainable, measurable returns.