$665 Billion Spent on AI in 2026. 73% of It Delivered No ROI. The Biggest Capital Misallocation in a Generation.

The global enterprise AI spending figure for 2026 is $665 billion. The failure rate, according to McKinsey's Global AI Survey, is 73%. That means roughly $485 billion will be spent on AI deployments that never achieve their projected return on investment.

To put that number in perspective, $485 billion is more than the GDP of most countries. It is more than the entire global advertising industry generates in a year. And it is being quietly written off as the cost of "digital transformation."

The Failure Metrics Are Converging

The 73% figure from McKinsey is not an outlier. Multiple independent studies are arriving at remarkably similar conclusions about the gap between AI investment and AI returns.

MIT found that 95% of generative AI pilots never reach production. They work in controlled environments with clean data and careful prompting. They fail when exposed to real-world conditions, messy data, and users who do not follow the intended workflow.

Across all AI project types, 85% never deliver measurable business value. They are technically functional but operationally irrelevant: the system works, but nobody uses it, or the output requires so much human verification that the efficiency gains disappear.

Perhaps most tellingly, 42% of companies scrap their AI initiatives entirely. Not scale them back. Not pivot them. Abandon them. Nearly half of all AI projects end with the company concluding that the investment was a mistake.

The Measurement Problem

One of the most damaging findings comes from MIT Sloan's 2025 research: 61% of enterprise AI projects were approved based on projected value that was never formally measured after deployment.

This means the majority of AI investments are approved with detailed ROI projections, funded with significant budgets, and then... nobody checks whether the projections were accurate. The project is declared a success based on the fact that it was completed, not on whether it delivered the promised results.

This creates a dangerous feedback loop. Leadership approves AI projects based on vendor projections. The projects are deployed. Nobody measures the actual return. The vendor case studies cite the deployment as a success. Other companies see the case study and approve their own AI projects based on similar projections.

The 73% failure rate is not a mystery. It is the predictable outcome of a market where success is defined by deployment rather than results.

The "AI Without a Home" Problem

MIT Sloan identified a pattern they called "AI without a home": AI systems that are technically delivered but never operationally adopted. This pattern appears in 41% of underperforming AI projects.

The system works. The demos are impressive. The vendor declares the project complete. And then the people who were supposed to use it continue doing their work the way they always have. The AI sits unused, burning cloud computing costs, while the humans it was supposed to augment or replace carry on as before.

This happens because most AI deployments focus on the technology and ignore the workflow. Building an AI system that can generate ad copy is straightforward. Integrating it into the actual process that a creative team uses to produce, review, approve, and publish content is an entirely different challenge, and it is the challenge that 41% of projects fail to solve.

The Hidden Costs

The failure rate numbers do not capture the secondary costs that AI deployments generate even when they "succeed."

LLM hallucinations alone cost businesses $67.4 billion in 2024. These are not theoretical risks. They are real financial losses caused by AI systems generating confident, plausible, and completely wrong information that humans then act on.

In financial services, AI-generated analysis hallucinations led to $2.3 billion in avoidable trading losses in the first quarter of 2026 alone. The AI produced analysis that looked professional, cited relevant data points, and reached conclusions that were factually incorrect. The traders who relied on it did not verify the underlying data because the output looked authoritative.

Employees spend an average of 4.3 hours per week verifying AI-generated outputs. At typical professional salary levels, this represents approximately $14,200 per employee per year in verification overhead. For a company with 1,000 employees using AI tools, that is $14.2 million per year in verification costs that never appeared in the original business case.

And 47% of business executives admit to making major decisions based on unverified AI-generated content. Nearly half of senior leaders are using AI output as decision inputs without checking whether the output is accurate. The cost of this behaviour is impossible to quantify because the errors are invisible until they cause measurable damage.

Why Companies Keep Investing Despite the Failure Rate

The rational response to a 73% failure rate would be to slow down, improve measurement, and deploy more carefully. The market is doing the opposite. AI spending is accelerating.

Three dynamics explain this behaviour.

First, the narrative reward. Companies that announce AI investments see stock price increases. Block's stock jumped 24% after announcing AI-related layoffs. The market rewards the story of AI adoption regardless of whether the AI actually delivers results. For publicly traded companies, the stock price benefit of announcing AI investment exceeds the financial cost of the investment failing.

Second, competitive fear. Every company is afraid of being the one that did not invest in AI while competitors did. This fear drives investment decisions even when the business case is weak. The logic is: "If our competitor deploys AI and we don't, we'll fall behind." The counter-logic, "If our competitor deploys AI and it fails, we'll be better positioned," is rarely considered.

Third, poor data quality. Organisations lose an average of $12.9 million annually due to poor data quality. AI systems built on poor data produce poor results. But the companies investing in AI are not investing equivalently in data quality. They are building sophisticated AI systems on top of data infrastructure that cannot support them.

What This Means for Agencies

The $665 billion AI spending surge creates both a threat and an opportunity for agencies.

The threat is obvious. Clients are diverting budgets from agency retainers to AI investments. Every dollar spent on an AI platform is a dollar not spent on an agency relationship. When the CFO needs to fund the AI transformation, the agency contract is often the first line item cut.

The opportunity is less obvious but potentially more valuable. When 73% of those AI investments fail, and the data says they will, the clients who cut their agencies will need help again. But they will not come back to the same agency offering the same services. They will come back looking for an agency that can do what their AI investment could not: deliver measurable, reliable results.

The agencies that position themselves now as the answer to "what happens when the AI doesn't work" will be positioned to win those clients back at higher margins. The sales pitch writes itself: "You spent six figures on an AI platform that produces minimally sufficient output 65% of the time. We produce work that drives measurable revenue. Here are the numbers."

The greatest capital misallocation in a generation is also the greatest opportunity for agencies that can prove their value with data their clients cannot ignore.