Enterprise AI Is Spending Big. The Returns Are Not Following.

The numbers tell a story that the vendor keynotes do not. According to a WRITER-commissioned survey of 2,400 executives and employees published this year, 59% of organizations now spend at least $1 million annually on artificial intelligence. Only 29% of those same organizations report seeing significant or measurable returns on that investment. The rest are spending, hoping, and waiting.

That gap between spending and returns is not a temporary hiccup on the way to AI maturity. It is a structural feature of how most enterprises are deploying AI, and understanding why it persists matters more than any individual product announcement or benchmark result.

The Spending Is Real and It Is Growing

KPMG’s Global AI Pulse survey for Q2 2026 arrived at similar conclusions using a completely different methodology. Average planned AI spending over the next 12 months among surveyed organizations was $188 million. Seventy-nine percent of organizations now call AI a strategic priority, up from 74% in the previous quarter. Twenty-two percent say AI is simply part of everyday work, up from 13%.

The spending is not concentrated in a few early adopters anymore. It has gone mainstream across industries, company sizes, and geographies. Statistics Canada reports that Canadian businesses more than tripled their AI adoption rate in two years, from 6.1% in Q2 2024 to 19.2% by Q2 2026. The pattern is consistent across markets: organizations are investing, and the pace is accelerating.

What is not accelerating is the return. The 29% ROI figure from the WRITER survey is not an outlier. KPMG’s data shows only 7% of organizations reporting significant returns, an even starker number. The disconnect is not subtle.

Where the Money Goes

Part of the problem is visibility. Only 35% of organizations report having full transparency into their AI operating costs, according to KPMG. When you cannot see where the money is going, measuring what it produces becomes nearly impossible. Agentic AI workloads make this worse. These systems run autonomously, generating their own usage spikes and consumption patterns that do not fit neatly into traditional budget forecasts.

Gartner is telling CIOs that end-user AI spending will climb steeply, and that three levers matter most for keeping budgets in check: contract management, AI architecture governance, and ongoing vendor oversight. The advice is practical but also revealing. Organizations that have been buying AI tools for two or three years are now discovering that multi-year enterprise agreements with seat-based pricing can mask the true cost of agentic workloads that generate unpredictable usage.

The enterprise AI stack is no longer a single-vendor question. It is a portfolio management problem spanning productivity platforms, CRM AI, agentic cloud tools, and the governance infrastructure needed to connect them. Most organizations are still assembling this portfolio without a clear architecture, which means costs compound and overlap in ways that are hard to trace.

The Governance Gap

The governance gap shows up in the numbers with brutal clarity. Fifty-nine percent of organizations spending over $1 million on AI, and only 24% assigning clear executive accountability for AI outcomes. Someone is writing the checks, but in most organizations, no single person or team owns the result. The money flows into a dozen different tools, platforms, and pilots, each managed by a different team with different metrics and different definitions of success. This is not a new problem. IT governance has been messy since the early days of enterprise software. But AI makes it worse in specific ways. Traditional software does what it is told. AI systems make decisions, generate outputs, and interact with users in ways that are harder to predict and audit. When an AI agent makes a mistake, tracing which model was used, what data it had access to, and who approved the deployment requires a level of infrastructure that most organizations have not built.

This matters because AI initiatives do not fail in dramatic, visible ways. They fail quietly. A pilot succeeds in a controlled environment, gets expanded to production, and then slowly degrades as usage patterns shift, data quality fluctuates, and the initial team moves on to other priorities. Without executive ownership, nobody is responsible for noticing that degradation until it shows up in financial results months later.

Fifty-four percent of C-suite executives describe AI adoption as internally disruptive. That is a polite way of saying that AI projects collide with existing workflows, data governance policies, and organizational structures that were not designed for autonomous agents making decisions at machine speed. The disruption is real, and most organizations are still figuring out how to manage it.

Why ROI Remains Elusive

The ROI problem has several layers, and untangling them explains why the gap persists even as spending climbs. The first is measurement. Most organizations have not defined what AI success looks like in quantifiable terms. They track usage metrics, adoption rates, and satisfaction scores, but these are proxies, not returns. A team that uses an AI coding assistant to write 30% more code is not 30% more productive unless that code is correct, maintainable, and aligned with business goals. A customer service chatbot that handles 40% more tickets is not saving money if the remaining 60% require more complex interventions because the easy cases were already offloaded. The metrics that organizations typically track tell you that AI is being used. They do not tell you whether it is creating value. The second layer is time. AI investments often take 12 to 18 months to produce measurable financial returns, but executive patience runs shorter than that. Forty-nine percent of organizations have delayed or scaled back an AI initiative because of cost concerns, according to KPMG. The investment gets cut before it has had time to mature. This creates a vicious cycle: the organization cuts AI spending because it cannot see returns, which prevents the AI from reaching the point where returns become visible, which reinforces the belief that AI does not produce returns. The third layer is the gap between pilot and production. AI tools tend to perform well in demos and controlled tests. Deploying them at scale, with real data, real users, and real consequences, is a different problem entirely. Data pipelines break. Edge cases multiply. User behavior diverges from training assumptions. The 79% of organizations reporting adoption challenges are mostly dealing with this gap, and the gap is wider for agentic systems that make autonomous decisions than for simple copilot tools that assist human workers. There is a fourth layer that gets less attention: the opportunity cost of choosing the wrong AI tools. Organizations that committed early to a specific platform or vendor now find themselves locked into ecosystems that may not be the best fit for their evolving needs. Switching costs are high, data portability is limited, and the contracts are multi-year. The result is that some organizations are spending significant sums on AI tools that are no longer the best option, but switching is more expensive than continuing to pay for something that does not quite work.

What Comes Next

The next phase of enterprise AI is unlikely to be defined by bigger models or flashier capabilities. It will be defined by whether organizations can close the governance gap, build cost visibility, and create the measurement frameworks that turn spending into returns. Some organizations are already moving in this direction. Snowflake launched a unified monitoring and cost management platform for agentic enterprise AI this week, aimed at giving IT teams the visibility they currently lack. The product addresses a real pain point: when AI agents run autonomously and generate their own usage spikes, traditional monitoring tools cannot keep up. You cannot govern what you cannot see, and right now, most organizations cannot see their AI agents clearly enough to manage them. Rimini Street also announced a governance platform for AI agents this week, focused on security and interoperability. The fact that multiple vendors are launching governance tools simultaneously is a signal that the market has identified this as a gap and is rushing to fill it. Whether these tools arrive fast enough to matter is a different question. The practical takeaway for technology leaders is straightforward. Stop measuring AI success by adoption rates and usage metrics. Start measuring it by business outcomes: cost savings, revenue impact, time freed, errors reduced. If you cannot draw a line from an AI tool to a specific business outcome, you do not have an ROI story. You have a spending story. For the 59% of organizations spending over $1 million on AI, the pressure to demonstrate returns will only increase. Boards are asking questions. CFOs are demanding justification. The era of “we need to be using AI” as sufficient rationale is ending. The era of “show me what AI produced” is beginning. Organizations that cannot answer that question will find their AI budgets shrinking, not because AI does not work, but because they cannot prove it does.

Twenty-two percent of organizations are already shifting toward lower-cost models, according to KPMG. That is not a retreat from AI. It is a sign that the market is maturing, that buyers are learning to match model capability to task complexity instead of defaulting to the most powerful option available.

The organizations that figure this out first will have a significant advantage. They will be able to scale AI investments with confidence, knowing that each dollar spent produces a traceable return. The rest will keep spending, hoping, and waiting, wondering why the ROI everyone promised has not shown up yet.

The data is clear. Enterprise AI spending is real, it is growing fast, and the returns are not keeping pace. The question is no longer whether to invest in AI. The question is whether you can afford to keep investing without knowing what you are getting back. For organizations still in the early stages of AI adoption, the lesson is to start with governance, not tools. Define what success looks like before you write the first check. Assign ownership before you deploy the first model. Build cost visibility before you scale the first pilot. These steps are less exciting than a new AI product launch, but they are the difference between spending that produces returns and spending that produces nothing but invoices. The enterprises that get this right will look back at 2026 as the year they built the foundation for AI that actually works. The ones that do not will look back at it as the year they spent a lot of money on something they could not explain.

Sources

• WRITER: 2026 AI adoption in the enterprise survey
• KPMG: Global AI Pulse Q2 2026