One in four AI dollars are going to waste

A significant portion of enterprise AI spending is wasted

AI adoption is moving fast. Cost controls are not moving at the same speed. According to Harness’ report, one in every four dollars spent on AI goes to waste. For executives increasing AI budgets, that is a material efficiency problem.

The challenge is not simply that companies are spending more. AI is spreading across engineering, operations, productivity software, and other business functions. Without clear guardrails, teams can buy overlapping tools, run services without sufficient oversight, and continue paying for AI products that deliver limited business value. Innovation remains important, but experimentation without cost visibility becomes expensive at scale.

This changes the executive conversation. The objective should not be to minimize AI spending. It should be to understand what the organization is buying, who owns each expense, how extensively each product is being used, and what outcome the investment produces. A rapidly growing AI budget can be justified when it creates measurable value. Waste cannot.

The nuance for executives is that aggressive cost cutting can create another problem. AI remains an evolving technology, and some experimentation will inevitably fail. Organizations therefore need to distinguish productive experimentation from uncontrolled spending. Clear budgets, usage monitoring, defined owners, and measurable outcomes can provide that discipline without slowing useful innovation.

The central data point is significant: Harness reports that roughly 25% of AI spending is wasted. If AI investment continues growing, even maintaining that waste rate would turn a manageable problem into a much larger financial issue. Improving efficiency early gives companies more capacity to fund AI projects that actually perform.

AI costs are more complex than traditional IT or cloud spending

Managing AI expenditure is difficult because there is no single AI bill. Costs can appear across computing infrastructure, foundation models, software subscriptions, managed services, copilots, and other AI products. Each category can have a different pricing model and a different internal owner.

Traditional cloud spending is often easier to map to resources such as servers and storage. AI introduces additional variables. A company may pay for model usage based on consumption, purchase AI features through software subscriptions, operate its own infrastructure, and use managed services at the same time. Harness also reports that most organizations work with three or more major AI providers, each with its own pricing structure.

That fragmentation matters at the C-suite level. Finance may see subscriptions. Engineering may see model and infrastructure consumption. Procurement may see vendor contracts. Business units may buy their own AI tools. Unless those records are connected, leadership can have a technically accurate view of individual expenses while still lacking a reliable view of total AI spending.

Executives should therefore treat AI cost management as a cross-functional responsibility. Finance, procurement, engineering, IT, and business leadership need consistent definitions for what counts as AI expenditure. Centralized reporting can then show cost by provider, product, team, project, and ideally business outcome. That makes budgeting and forecasting substantially more useful.

Using several AI providers is not inherently inefficient. Different providers can offer advantages in capability, reliability, price, geographic availability, or negotiating leverage. Consolidating vendors purely to simplify reporting could remove useful flexibility. The better objective is visibility: management should know why each provider is being used and whether its contribution justifies its cost.

Harness’ finding that most organizations use at least three major AI providers shows why conventional IT budgeting processes can struggle. The companies that solve this early will have a clearer understanding of AI economics. That creates room to invest more intelligently.

AI productivity tools are becoming a major but difficult-to-track cost

AI copilots, coding assistants, and other productivity tools are becoming a meaningful part of enterprise AI spending. The problem is visibility. These products are often purchased as standard software subscriptions, so their costs may sit inside existing SaaS budgets rather than appear as dedicated AI expenditure.

That makes the real cost of AI easy to underestimate. Different departments can purchase separate tools, employees may receive licenses they rarely use, and several products may offer similar capabilities. Each expense can appear reasonable on its own. Across a large organization, however, unused licenses and overlapping products can create substantial costs.

For executives, the important question is not simply how many AI tools the company has purchased. The focus should be whether people actively use them and whether that use produces measurable results. For a coding assistant, that could include changes in development speed, software quality, or time spent on repetitive work. For a general productivity copilot, leaders could examine adoption, frequency of use, time saved, and whether the tool improves specific business processes.

Low initial adoption does not automatically mean a product has no value. Employees may need training, workflows may need to change, and some tools produce greater benefits for specific roles than for an entire workforce. Executives should therefore avoid both automatic renewals and premature cancellation. Usage and business outcomes should guide those decisions.

The Harness report identifies productivity software, including AI copilots and coding assistants, as one of the largest AI cost drivers. It does not provide a specific percentage for this category in the supplied text. That distinction matters: the evidence supports the significance of the category, but not a precise estimate of its share of total AI spending.

Weak visibility, forecasting, and ownership are limiting AI cost control

Many companies are spending significant amounts on AI without having reliable systems to forecast or manage those expenses. According to Harness, more than half of respondents forecast AI spending using guesswork rather than data, while more than 40% still rely on spreadsheets for cost management.

That approach becomes increasingly risky as AI budgets grow. AI consumption can change quickly as more employees adopt tools, applications generate more model requests, or teams expand deployments. If cost information arrives late, CFOs and technology leaders can identify an overspend but may have limited ability to prevent it.

The problem also extends beyond forecasting. Someone needs to own the spending. Engineering teams understand technical consumption, finance controls budgets, procurement manages contracts, and business units understand the intended outcome. Without clear accountability across these functions, each team can have part of the information while no one has a complete view.

Harish Doddala, VP of Cloud and AI Cost Management at Harness, emphasized how widespread this issue is: “What surprised me most wasn’t the size of the spend or the speed of the growth, but how consistent the gaps are across every size and geography.” He added: “The visibility problem, the ownership problem, the forecasting problem show up whether you are spending $300K a month or $3M.”

For C-suite leaders, this suggests that scale alone will not solve the problem. Doddala argues that time will not be sufficient to close visibility gaps; organizations need targeted investments. That can include centralized reporting, explicit budget owners, better forecasting based on actual consumption, and cost information integrated into engineering workflows.

Spreadsheets are not inherently ineffective, especially during small pilots or early adoption. The issue emerges when AI spending becomes distributed, dynamic, and financially material. At that point, manual processes can create delays, inconsistent definitions, and limited accountability.

Harness’ figures make the management gap clear: more than 50% of respondents rely on guesswork to forecast AI spending, and more than 40% use spreadsheets to manage costs. Combined with Doddala’s observation that the same challenges appear at monthly spending levels from $300,000 to $3 million, the message for executives is straightforward: stronger AI economics requires better information and clearly assigned responsibility.

Technology providers are improving AI cost management, but enterprises still need their own controls

AI spending has become large and complex enough that technology providers are building new tools to help customers understand it. Oracle and AWS have introduced features and billing structures designed to improve visibility into AI costs. The Linux Foundation has also launched a group focused specifically on cost management.

These developments matter because AI expenses can be distributed across infrastructure, model usage, managed services, and software subscriptions. Better billing information can help finance and technology teams identify where money is going and how consumption changes over time. It can also support more accurate internal reporting as organizations expand their use of AI.

But improved vendor tooling does not automatically create complete financial visibility. Most organizations use three or more major AI providers, according to Harness. Each provider can present usage, pricing, and billing differently. A company may therefore have detailed information within individual platforms while still struggling to understand its total AI expenditure across the organization.

C-suite leaders should view vendor cost-management features as one part of a broader internal capability. Finance, procurement, engineering, and IT need a consistent method for combining provider data with software subscriptions and other AI-related expenses. Leadership can then evaluate spending by business unit, application, vendor, and business objective rather than reviewing separate bills in isolation.

There is also a strategic issue around dependence on vendor reporting. Providers can give customers detailed information about consumption within their own platforms, but enterprises remain responsible for determining whether that consumption creates sufficient business value. Cost visibility and value measurement are related but different management tasks.

The direction of travel is positive. Oracle and AWS are increasing AI cost visibility, while the Linux Foundation’s initiative indicates broader industry attention to cost management. Executives should therefore evaluate them based on measurable improvements in forecasting, allocation, and decision-making rather than assuming new tooling will solve the problem by itself.

Effective AI cost control requires ownership, centralized visibility, and governance inside daily operations

The strongest approach to AI cost management starts with accountability. Harness says organizations with better cost discipline assign ownership early, centralize cost visibility, and integrate spending information into engineering workflows. This allows financial considerations to become part of operational decisions rather than appearing only after invoices arrive.

Clear ownership is especially important. Every significant AI deployment should have someone accountable for its spending and expected results. That does not mean finance should control every technical decision. It means executives should be able to identify who owns the budget, what the investment is intended to achieve, how consumption is changing, and whether the results justify continued expenditure.

Centralized visibility provides the next requirement. Companies need to combine spending information from model providers, cloud platforms, AI software subscriptions, managed services, and relevant infrastructure. This gives CFOs and technology leaders a more complete basis for forecasting and resource allocation. It can also reveal duplicate products, unused capacity, unexpected consumption, and costs that individual departments may not see.

Harness also recommends bringing cost information into engineering workflows. This is important because engineers often make decisions that directly affect AI consumption. Giving technical teams timely information about cost allows them to consider price alongside performance, reliability, and product requirements while systems are being developed and operated.

OpenAI offered similar guidance. Earlier in the month, OpenAI urged businesses to improve visibility into AI usage and spending, track model outcomes, and embed governance into everyday operations. Tracking outcomes is particularly important: lower spending is not necessarily better if a more expensive model or service produces substantially greater business value.

For C-suite executives, governance should therefore support both control and growth. Excessive approval processes can slow useful experimentation, while insufficient controls can allow fragmented spending to expand without evidence of value. Organizations can address both concerns by setting clear thresholds: smaller experiments can operate within predefined budgets, while larger deployments receive greater financial and performance oversight.

Harness does not provide quantitative evidence in the supplied text showing how much these specific practices reduce AI spending. Its report identifies early ownership, centralized visibility, and integration with engineering workflows as characteristics of organizations with stronger cost discipline. Combined with OpenAI’s recommendation to monitor usage, spending, and outcomes, the broader executive priority is clear: AI governance needs to become part of normal operations as AI investment scales.

Key executive takeaways

• Control AI waste before budgets scale: Harness reports that one in four AI dollars goes to waste. Leaders should set clear spending guardrails and connect AI investments to measurable business outcomes without restricting productive experimentation.
• Get a complete view of AI costs: AI spending spans infrastructure, models, SaaS, and managed services, while most organizations use three or more major providers. Centralize cost data so finance and technology leaders can see total expenditure and make informed allocation decisions.
• Track productivity tools as AI investments: Copilots and coding assistants are major cost drivers but can disappear inside standard software budgets. Monitor license utilization, adoption, and measurable productivity gains before expanding or renewing deployments.
• Replace guesswork with accountable forecasting: More than half of organizations forecast AI spending through guesswork, while over 40% rely on spreadsheets. Establish clear budget owners and use timely consumption data to shift from reacting to costs to actively managing them.
• Use vendor cost tools without depending on them: Oracle, AWS, and industry groups are improving AI cost visibility, but vendor-level reporting does not provide a complete enterprise view. Combine these capabilities with internal controls that measure costs across providers and against business value.
• Build AI governance into daily operations: Organizations with stronger cost discipline assign ownership early, centralize visibility, and integrate spending data into engineering workflows. Leaders should pair these controls with outcome tracking so teams optimize AI investment for value.