In January 2026, PwC published a number that should've shaken every boardroom on the planet. Their 29th Global CEO Survey, based on responses from 4,454 CEOs across 95 countries, found that 56% of companies have seen zero measurable financial benefit from AI. Not reduced costs. Not increased revenue. Nothing.
Meanwhile, the world's biggest companies are pouring hundreds of billions into AI infrastructure. Microsoft alone committed $80 billion in capital expenditure for 2025. Google, Amazon, and Meta aren't far behind. Combined, Big Tech CapEx for AI data centers is approaching $300 billion annually. Enterprise AI spending globally is estimated to exceed $700 billion by end of 2026.
So where's the money going? Why is more than half of it apparently evaporating?
The Numbers Are Worse Than They Look
Let's break down what PwC actually found. Of the 4,454 CEOs surveyed between September and November 2025:
- 56% reported no significant financial benefit from AI whatsoever
- 33% reported gains in either cost reduction or revenue (but not both)
- Only 12% said AI delivered benefits on both the cost and revenue side
That 12% number should terrify people. After years of hype, after Davos panels, after McKinsey reports, after thousands of "AI transformation" initiatives, only one in eight CEOs can point to AI helping their company on both sides of the ledger.
CEO confidence is tanking alongside these numbers. Only 30% of CEOs expressed confidence in their company's revenue growth over the next 12 months. Down from 38% in 2025 and 56% in 2022. A five-year low. Mohamed Kande, PwC's Global Chairman, told Fortune that companies have "forgotten the basics." Clean data, solid processes, proper governance. He described the current moment as one of the most testing for leaders in his 25 years of consulting.
The $700 Billion Black Hole
What I find genuinely baffling: we're not talking about experimental budgets or R&D moonshots. The scale of AI spending in 2025 and 2026 is staggering.
- Microsoft: ~$80 billion in CapEx, much of it on AI data centers
- Google/Alphabet: ~$75 billion committed for 2025
- Amazon/AWS: ~$100 billion in planned AI infrastructure spending
- Meta: ~$60-65 billion in CapEx earmarked for AI
And that's just the hyperscalers. Enterprise companies on top of that are spending billions collectively on AI platforms, consulting fees, talent acquisition, and "transformation" programs.
So what did all that money actually buy? In most cases, one of three things.
Fancy demos that never shipped. Companies ran hundreds of AI pilots, chatbots, predictive models, automation tools, that looked impressive in boardroom presentations but never made it into production. An MIT study referenced alongside the PwC findings estimated that 95% of generative AI pilots were failing across the corporate sector.
Tools nobody used. Plenty of companies bought AI platforms and licenses, deployed them to employees, and watched adoption rates hover around 10-15%. The tools were either too complicated, solved problems people didn't have, or created more work than they eliminated.
Infrastructure without strategy. A lot of enterprise AI spending went to compute, storage, and cloud resources. Companies built the plumbing but forgot to figure out what they were actually building with it.
The Failures
We don't have to guess what failure looks like. We have examples.
McDonald's spent years developing an AI-powered drive-through ordering system in partnership with IBM. The project ran from 2021 through mid-2024, tested at over 100 locations. The AI couldn't handle real-world complexity. Customers with accents, background noise, unusual orders. McDonald's ended the partnership and pulled the plug. A simple ordering task that humans do effortlessly defeated one of the most well-funded AI deployments in fast food.
IBM's Watson Health was supposed to revolutionize cancer treatment. IBM invested billions and partnered with top hospitals. The system made unsafe treatment recommendations. By 2022, IBM sold off Watson Health for parts, reportedly at a fraction of what they'd invested. The technology couldn't handle the messy, incomplete, inconsistent data that characterizes real healthcare systems.
Klarna's CEO bragged that their AI assistant was doing the work of 700 customer service agents. Then quality complaints surged, customers got stuck in loops, and Klarna had to start rehiring humans. The AI could handle simple queries but crumbled when situations required judgment, empathy, or creativity. Which turns out to be most real customer service interactions.
These aren't edge cases. They're the norm. Most enterprise AI deployments follow the same arc: big announcement, pilot phase that looks promising, production deployment that reveals all the problems nobody anticipated, quiet scaling back, and eventual pivot to the next shiny thing.
What the 12% Do Differently
Now here's where it gets interesting. 12% of CEOs are seeing real returns. PwC's data shows that companies reporting both cost and revenue benefits are two to three times more likely to have embedded AI extensively across products, services, and strategic decision-making. They're not running isolated pilots. They're integrating AI into the core of how they operate.
Companies that applied AI broadly to products, services, and customer experiences achieved nearly four percentage points higher profit margins than those that didn't. That's a massive gap at scale.
PwC found that companies with strong AI foundations, responsible AI frameworks, enterprise-wide integration capabilities, clean data infrastructure, were three times more likely to report meaningful financial returns. Mohamed Kande's point about "the basics" isn't just consultant-speak. It's the actual dividing line.
So what separates the winners?
They started with the problem, not the technology. JPMorgan built their COiN platform to handle a specific, painful task: reviewing commercial loan agreements. It processes in seconds what took lawyers 360,000 hours annually. The AI didn't exist for its own sake. It solved a real, expensive, measurable problem.
They invested in data before algorithms. Netflix reportedly saves $1 billion per year through its recommendation engine. But that engine works because Netflix spent years building one of the most sophisticated data collection and processing pipelines in entertainment. The AI sits on top of enormous data infrastructure, not the other way around.
They changed workflows, not just added tools. Google DeepMind cut data center cooling costs by 40% using AI. But that required redesigning how cooling systems operated. Not just bolting an AI tool onto existing infrastructure.
They measured relentlessly. The 12% who succeed track specific, measurable outcomes: revenue per customer, cost per transaction, time to resolution. The 56% who fail tend to measure adoption metrics, how many people logged in, how many queries were processed, without connecting those to business outcomes.
The Real Problem
I think there's a deeper issue here that most AI commentary dances around. A lot of enterprise AI spending isn't really about business outcomes. It's about fear.
CEOs see competitors announcing AI initiatives. Boards ask what the AI strategy is. Analysts penalize companies that seem behind the curve. So companies spend money on AI because not spending feels riskier than spending. The result is what we're seeing now: hundreds of billions deployed with no clear connection to profit.
PwC's survey found that 42% of CEOs say their top concern is "transforming the business fast enough to keep up with technology." That's not a strategy. That's anxiety dressed up as a priority.
And what worries me most about the data: only one in four CEOs say their organization has disciplined processes to stop underperforming initiatives. Three out of four companies are running AI projects with no real mechanism to kill the ones that aren't working. Money keeps flowing because nobody wants to be the executive who admitted the AI project was a dud.
What Happens Next
The PwC data points to a widening gap. The 12% who've figured out how to extract real value from AI will pull further ahead. The 56% who haven't will either course-correct or keep burning cash on initiatives that go nowhere.
Kande described the situation as a "decade of innovation and industry reconfiguration." Probably right. But I'd add a qualifier: it's going to be a decade where most of the value flows to a small number of companies that actually know what they're doing.
For everyone else, the path forward is embarrassingly simple. Clean your data. Pick specific problems with measurable outcomes. Build proper governance. Kill projects that aren't working. Stop treating AI like a magic wand and start treating it like any other capital allocation decision. With discipline, accountability, and clear expectations about returns.
The technology works. The question has always been whether the organizations deploying it are willing to do the boring, foundational work that makes it valuable. As of January 2026, the answer for 56% of CEOs is: not yet.
