306 Companies Say They're Doing AI. About 15 Actually Are.

In Q3 2025, 306 S&P 500 companies cited "AI" on their earnings calls — the highest number in a decade, up from a five-year average of 136 and a ten-year average of 86. The mentions aren't casual. CEOs are naming initiatives, announcing partnerships, and forecasting billions in AI-driven efficiency gains. Wall Street is rewarding them for it: companies that mentioned AI on Q3 calls saw an average price increase of 13.9%, compared to 5.7% for those that didn't.

Meanwhile, MIT's State of AI in Business 2025 report — based on 52 executive interviews, surveys of 153 leaders, and analysis of 300 public AI deployments — found that 95% of generative AI pilots yield no measurable business return.

That is the gap. Not between hype and reality — that framing is too generous. This is the gap between what publicly traded companies tell shareholders and what actually ships. Between the $2.52 trillion the world will spend on AI in 2026 and the fewer-than-10% of companies that have scaled a single AI agent to production. Between the earnings call and the engineering standup.

The Numbers Don't Reconcile

Start with the spending. Gartner forecasts $2.52 trillion in worldwide AI spending for 2026, a 44% increase from $1.5 trillion in 2025. AI infrastructure software spending alone will hit $230 billion — nearly 4x from $60 billion two years ago. Compute and storage infrastructure spending for AI deployments increased 166% year-over-year in Q2 2025, reaching $82 billion in a single quarter. AI startups received 63% of all venture capital in the 12 months through Q3 2025, up from 40% the prior year.

Now look at the results. 42% of companies abandoned most of their AI initiatives in 2025, up from 17% in 2024. Over 80% of AI projects fail to reach production — twice the failure rate of non-AI technology projects. McKinsey found that while 78% of companies have "deployed AI" in some form, fewer than 10% have scaled agents to production. Nearly two-thirds of organizations remain stuck in the pilot stage.

The revenue ambition is equally disconnected. 74% of organizations want AI initiatives to grow revenue, but only 20% have seen it happen. 42% of AI projects show zero ROI). MIT estimates that enterprise GenAI spending sits at $30-40 billion with 95% yielding no measurable P&L impact.

Put differently: the enterprise world is running a $2.52 trillion experiment with a 5% success rate.

The Pilot Purgatory Problem

The pattern is remarkably consistent across industries. A company announces an AI initiative with a press release, a consulting partner, and a slide deck. Six months later, a pilot goes live — usually in a controlled environment with clean data and motivated stakeholders. And then nothing. The pilot doesn't scale. It doesn't die either. It enters what ISG calls the "pilot purgatory," where 32% of organizations stall after their initial pilot, never reaching production.

The numbers from Asia Pacific are particularly revealing. According to CIO.com's State of the CIO 2025 report, organizations in the region conducted an average of 24 GenAI pilots over 12 months, but only 3 progressed into production. That's a 12.5% conversion rate from pilot to production — and those are the companies that got to the pilot stage at all. 63.7% of enterprises report no formalized AI initiative whatsoever, despite the earnings call rhetoric.

There is a bright spot. The share of organizations with deployed agents nearly doubled from 7.2% in August 2025 to 13.2% in December 2025. 31% of use cases reached full production in 2025, double the amount from 2024. The curve is inflecting — but from a very low base.

Why the Pilots Fail

The failure isn't a mystery. It's well-documented. The problem is that almost nobody wants to hear the answer.

73% of 500 enterprise data leaders identified "data quality and completeness" as the primary barrier to AI success. Not the model. Not the vendor. Not the infrastructure. The data. The Informatica CDO Insights 2025 survey found three near-equal top obstacles: data quality and readiness (43%), lack of technical maturity (43%), and shortage of skills (35%).

The MIT 2025 report went further, arguing that the core barrier to scaling GenAI is not infrastructure, regulation, or talent — it is learning. Most GenAI systems do not retain feedback, adapt to context, or improve over time. They are static tools deployed into dynamic environments. The pilot works because the environment is controlled. It fails in production because the real world isn't.

This explains a counterintuitive finding: winning programs invert typical spending ratios, earmarking 50-70% of timeline and budget for data readiness rather than modeling. The companies that succeed at AI aren't spending more on AI. They're spending more on plumbing.

And then there's the talent gap. AI talent demand exceeds supply by 3.2 to 1 globally, with over 1.6 million open positions and only 518,000 qualified candidates. 68% of companies face moderate to extreme AI talent shortage. The average salary for AI specialists has hit $206,000 in 2026 — $50,000 more than 2024, and 67% higher than traditional software positions. Only 20% of organizations say their talent is highly prepared for AI. The companies that need AI transformation the most are the companies least equipped to execute it.

The Case Studies Nobody Wants to Talk About

The high-profile failures tell the story more vividly than any survey data.

McDonald's ended its Automated Order Taking partnership with IBM in July 2024. The system, deployed across test locations, failed to meet accuracy levels when confronted with different accents and dialects. A drive-thru AI that can't understand a meaningful percentage of its customers isn't a pilot that needs refinement. It's a product that doesn't work.

Volkswagen's Cariad unit launched in 2020 with a sweeping mandate: build one unified AI-driven operating system for all 12 VW brands. By 2025, it had become automotive's most expensive software failure. The ambition was enterprise transformation. The result was billions burned and software that couldn't ship on time for a single brand, let alone twelve.

Air Canada was taken to court after its chatbot gave misleading information on bereavement fares — a case that established a legal precedent: companies are liable for what their AI tells customers, regardless of whether a human would have said the same thing.

Taco Bell expanded AI voice-ordering to over 100 locations, but the system misinterpreted orders in noisy environments. A viral incident of a customer being quoted 18,000 cups of water was funny on social media and catastrophic for the business case.

These aren't edge cases. They are representative. The failure mode is consistent: AI that performs well in a demo environment — with clean data, predictable inputs, and controlled conditions — collapses when confronted with the entropy of the real world.

The Consulting Gold Rush

The companies failing at AI are, however, generating extraordinary returns for someone: their consultants.

Accenture has booked $3.6 billion in generative AI consulting, with Q1 FY2026 Advanced AI revenues hitting $1.1 billion — up 120% year-over-year. The firm plans to have 80,000 data and AI professionals by 2026. McKinsey's QuantumBlack unit, with 1,700 dedicated AI staff, now accounts for roughly 40% of the firm's total revenue. CEO Bob Sternfels says McKinsey deploys 25,000 AI agents alongside 40,000 human consultants, targeting parity by end of 2026. EY added 61,000 technologists since 2023 and commits over $1 billion annually to AI platforms.

The AI consulting services market will grow from $11.07 billion in 2026 to $90.99 billion by 2035 at a 26.2% CAGR. That's the projected revenue for advising companies on AI — a number that grows regardless of whether the advised companies succeed.

This is the structural misalignment at the heart of the enterprise AI boom. Consulting firms are incentivized to sell AI transformation programs. Their revenue comes from the engagement, not from the outcome. A $20 million pilot that fails to reach production and gets replaced by a $30 million "Phase 2" program is, from the consultant's perspective, a success.

Shadow AI: The Transformation That Actually Happened

While the official AI programs stall, something else has been happening quietly.

81% of employees and 88% of security leaders use unapproved AI tools. Shadow AI tool usage increased 156% from 2023 to 2025. MIT found that while only 40% of companies say they purchased an official LLM subscription, workers from over 90% of companies surveyed report regular use of personal AI tools for work.

The irony is severe. Companies spend billions on top-down AI transformation programs that don't ship. Meanwhile, their employees spend $20/month on ChatGPT Plus and quietly transform their own workflows without permission, training, or governance. The AI transformation that executives talk about on earnings calls isn't happening. The AI transformation they don't know about is.

The risks are real. Shadow AI costs companies an average of $412K per year. Security breaches linked to unauthorized AI tools cost $670,000 per incident. Shadow AI increases attack surface by 340%. 20% of organizations experienced security incidents linked to Shadow AI in 2025. And only 37% of organizations have governance policies for AI tools — meaning 63% are flying blind.

The governance gap is staggering. Only 43% of organizations have an AI governance policy. Only one in five companies has a mature governance model for autonomous AI agents. Info-Tech Research Group identified a 2.8-point gap between the importance and effectiveness of data governance — the single largest capability gap in its survey. AI is the top strategic priority for CIOs. Governing it properly is an afterthought.

The SEC Steps In: AI Washing Meets Enforcement

The gap between AI announcements and AI reality has caught the attention of regulators. The SEC created the Cyber and Emerging Technologies Unit (CETU) in February 2025, tasked with combating "AI washing" as an immediate priority.

The first enforcement action landed quickly. Presto Automation claimed its Presto Voice AI eliminated the need for human drive-thru order-taking. The SEC found that "the vast majority of drive-thru orders required human intervention." The company said AI. The reality was humans with headsets. In April 2025, the SEC filed a civil complaint against the former CEO of Nate Inc. for similar misrepresentations.

The trend is accelerating. Securities class actions targeting alleged AI misrepresentations increased by 100% between 2023 and 2024 with no signs of slowing. In the SEC's 2026 examination priorities, AI concerns have displaced cryptocurrency as the industry's dominant risk topic. That's a regulatory regime change.

The incentive structure explains why AI washing is so tempting. S&P 500 companies that cited AI on earnings calls saw an average price increase of 13.9% versus 5.7% for those that didn't. When mentioning "AI" on a quarterly call is worth an 8-percentage-point stock bump, the temptation to exaggerate capabilities becomes a governance problem, not just a marketing one.

What the 5% Club Does Differently

Not everyone is failing. And the gap between the companies that ship and the companies that don't is instructive.

Companies that reach production share several patterns. They invest 50-70% of their timeline and budget in data readiness before touching model development. They scope narrowly — solving one specific problem rather than pursuing "enterprise AI transformation." They set quantitative success criteria before the pilot begins, so there's a clear line between "this works" and "this doesn't."

The WEF's MINDS programme recognized 33 companies across two cohorts that report double-digit gains in productivity and revenue from scaled AI. What separates them isn't budget or talent. It's that they treated AI as an engineering problem rather than a transformation narrative. They didn't announce. They built.

The ROI for companies that do reach production is compelling. Enterprises that ship report an average $3.70 return per dollar invested. Visionary AI adopters show 1.7x revenue growth, 3.6x three-year total shareholder return, 2.7x return on invested capital, and 1.6x EBIT margin compared to laggards. McKinsey found cost savings of 26-31% across supply chain, finance, and customer operations in organizations that scale successfully.

The prize is real. Getting there is the problem.

What Comes Next

The enterprise AI story in 2026 is a market that is correcting in slow motion. IDC predicts over one-third of organizations will remain stuck in the experimental phase through the end of the year. 54% of CIO respondents cite staffing and talent shortages in AI, cybersecurity, and data science as the most significant hurdle. The compliance burden is growing: 60% of enterprises identify integrating with legacy systems and addressing risk and compliance as their primary challenges in adopting agentic AI. Compliance costs already average $2.7 million annually for large enterprises operating in Europe.

But two forces are converging that could break the pattern. First, the production deployment rate is genuinely accelerating — doubling in the second half of 2025. The companies emerging from pilot purgatory are publishing playbooks, and second-movers are learning from first-mover failures. Second, SEC enforcement against AI washing is raising the cost of empty announcements. When exaggerating your AI capabilities risks a federal lawsuit, the incentive to ship something real increases.

The $2.52 trillion question isn't whether AI works — it does, for the 5% that reach production. The question is whether the enterprise world can close the gap between the earnings call and the engineering org. Between the consulting deck and the deployed system. Between the announcement and the thing.

306 companies say they're doing AI. The market is about to find out which ones are telling the truth.