The AI spend bubble has a receipt

Enterprise AI spending hit $407 billion in 2026, up nearly 35% from the year before. Worldwide, Gartner pegs total AI spending at $2.52 trillion, a 44% year-over-year jump. OpenAI and Anthropic are dominating enterprise budgets, and TechRadar recently declared that AI is becoming "THE line item." Legacy SaaS providers are scrambling to survive. But here's the uncomfortable question: is the ROI actually there, or are companies spending because they're terrified of being left behind? When fear drives the budget, the receipt always comes due.

The doubling isn't organic adoption

U.S. companies spent $37 billion on generative AI alone in 2025, according to Menlo Ventures. That's a 20x increase from $1.7 billion in 2023. Hyperscaler capital expenditure is expected to exceed $600 billion in 2026, with roughly 75% of it tied directly to AI infrastructure. These aren't numbers that reflect careful, ROI-driven procurement. This is panic buying at scale. The pattern is familiar to anyone who's watched enterprise tech cycles. A new category emerges, the board starts asking questions, and suddenly every CIO needs an "AI strategy" to present at the next quarterly review. The strategy doesn't need to work yet. It just needs to exist.

Fear-driven procurement

A recent CIO Dive survey found that 71% of global CIOs said their AI budgets would be frozen or cut if value couldn't be demonstrated within two years. That's a telling number, because it reveals the implicit truth: most of them can't demonstrate value now. McKinsey's 2025 State of AI survey found that while 88% of organizations use AI in at least one business function, only 39% report any impact on EBIT at the enterprise level. Forrester is even more blunt, reporting that just 15% of AI decision-makers saw a positive impact on profitability in the past 12 months. So why does the spending keep climbing? Because the cost of not spending feels worse. No CIO wants to be the one who told the board, "We decided to wait." The fear of missing out is a more powerful budget driver than any business case.

The shelfware problem

Enterprise tech has a long history of panic-bought tools gathering dust. Cloud, big data, blockchain, each wave produced its own generation of shelfware. Software that was purchased to check a strategic box, not to solve a specific problem. AI is following the same script. IBM's 2025 CEO study found that only 25% of AI initiatives deliver expected ROI, and just 16% have scaled enterprise-wide. One analysis puts the number even lower: only 1 in 50 AI investments delivers transformational value. The most striking stat comes from Deloitte's 2026 State of AI in the Enterprise report: 75% of companies plan to invest in agentic AI, but only 11% have agents running in production. A separate March 2026 survey of 650 technology leaders found that 78% of enterprises have active AI agent pilots, but under 15% have reached production scale. Gartner predicts that over 40% of agentic AI projects will be canceled by the end of 2027. So where is the rest of the spend going? Pilots, POCs, and PowerPoints.

The Jevons paradox in action

When DeepSeek emerged in early 2025 with dramatically cheaper AI models, Microsoft CEO Satya Nadella posted on LinkedIn: "Jevons paradox strikes again! As AI gets more efficient and accessible, we will see its use skyrocket." The Jevons paradox, named after 19th-century economist William Stanley Jevons, observes that when a resource becomes cheaper to use, total consumption often increases rather than decreases. Coal-powered steam engines got more efficient, and Britain burned more coal, not less. The same dynamic is playing out with AI. The cost per token has plummeted, but enterprise GenAI spending went from $1.7 billion to $37 billion in two years. Cheaper AI doesn't mean lower AI budgets. It means more AI everywhere, whether or not every deployment is justified. But here's what gets lost in the Jevons framing: more spending doesn't automatically mean more value. The paradox describes increased consumption, not increased utility. Companies are consuming more AI. The question is whether they're getting more out of it.

Winners and losers

Model providers win regardless. OpenAI, Anthropic, and the hyperscalers are collecting revenue whether their customers achieve ROI or not. The infrastructure buildout is real, Morgan Stanley estimates nearly $3 trillion in AI-related infrastructure investment flowing through the global economy by 2028, with more than 80% of that spending still ahead. Enterprises, on the other hand, only win if they have real use cases and the integration capacity to execute on them. The highest-ROI deployments in 2025 weren't the glamorous customer-facing chatbots. They were document processing, data reconciliation, compliance checks, and invoice handling. The boring work that nobody wants to do but everybody needs done. The companies seeing genuine returns are the ones treating AI like an operational tool, not a strategic talking point. McKinsey found that high performers often set growth or innovation as objectives alongside efficiency, rather than just cutting costs. They're using AI to do new things, not just to do old things with a chatbot layer on top.

We've seen this movie before

The cloud spending wave of 2018 to 2020 followed a remarkably similar arc. Companies rushed to migrate to the cloud because the board said so. Budgets ballooned. Then the bills arrived, and nobody could explain where the money went. What followed was the rise of FinOps, a discipline built around cloud financial management. Organizations that had spent freely during the migration rush suddenly needed frameworks, tools, and teams dedicated to rationalizing their cloud spend. The FinOps Foundation grew into a major industry force, and "cloud cost optimization" became one of the hottest categories in enterprise software. AI is on the same trajectory. Forrester predicts that enterprises will defer 25% of planned 2026 AI spend into 2027 as the pressure to demonstrate value intensifies. The optimization wave is coming.

The next hot category: AI FinOps

If the pattern holds, "AI FinOps" or something like it will be the breakout enterprise category of 2027. Companies that are spending aggressively today will need tools and practices to audit, rationalize, and optimize their AI investments tomorrow. This isn't a prediction about AI failing. It's a prediction about AI spending maturing. The technology is real and the spending is sticky. What's unsustainable is the current mode of allocation, where budgets are driven by fear and competitive anxiety rather than measured impact. IDC projects AI spending will surpass $632 billion by 2028. That money isn't going away. But the companies that survive the optimization wave will be the ones that can point to their AI spend and say, "Here's what we got for it."

The receipt

None of this means AI is a bad investment. It means that undirected AI investment is a bad investment, just like undirected cloud migration was a bad investment, and undirected big data initiatives were bad investments before that. The companies navigating this well deserve genuine empathy. The pressure is real, the technology is evolving fast, and the strategic consequences of getting it wrong feel existential. But the answer to "what's our AI strategy?" can't just be "spend more." At some point, the receipt comes due, and the board will want to see what all that money actually bought. The smartest move right now isn't to stop spending. It's to start measuring. Because the companies that can connect their AI investments to real outcomes will be the ones that thrive when the optimization wave arrives, and the ones that can't will be writing off a lot of expensive shelfware.

References