Harness report says AI spending waste hits one in four dollars

AI spending waste is now large enough to show up as a finance problem: Harness said Wednesday that roughly one in four dollars enterprises spend on AI is not producing value. The finding comes from a Harness survey of 700 FinOps and engineering leaders, and it points to a familiar enterprise software failure mode: adoption is running ahead of ownership, controls and measurement.

Harness did not give a single aggregate dollar figure for the respondents’ AI budgets in the findings cited. It did say more than half of businesses surveyed have no dedicated owner for AI costs, leaving accountability split across engineering, platform and FinOps teams.

The cost-control gap is surfacing as companies add AI products across infrastructure, developer tooling and productivity software. Harness said only one in five organizations can identify the cause of an unexpected AI cost spike within hours, which means many teams are responding after the bill moves rather than controlling usage as it happens.

Why is AI spending waste hard to track?

AI costs do not map cleanly to the older cloud cost model. Harness said AI expenses can include infrastructure, foundation models, SaaS subscriptions and managed services at the same time, making the bill harder to assign to a product, team or business outcome.

Provider sprawl adds another layer. Most organizations surveyed use three or more major AI providers, according to Harness, and those vendors price services differently. That makes side-by-side comparisons and internal chargeback harder than tracking compute and storage consumption from a narrower cloud stack.

Harness also singled out AI productivity tools, including copilots and coding assistants, as a major cost driver. Those products can be treated internally like ordinary software seats, even when their usage patterns and budget impact look more like consumption-based infrastructure.

Ownership and forecasting are still weak

More than half of respondents said their organizations forecast AI costs through estimates rather than data, Harness found. More than 40% still use spreadsheets to manage AI spending, even as the category becomes large enough to draw CFO attention.

Harish Doddala, vice president of cloud and AI cost management at Harness, said in the report that the pattern was broad across company sizes and regions. “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,” Doddala said. “The visibility problem, the ownership problem, the forecasting problem show up whether you are spending $300K a month or $3M.”

Doddala said time alone will not close the visibility gap. Harness said organizations with stronger cost discipline tend to assign ownership earlier, centralize cost visibility and connect spending data to engineering workflows.

Vendors are selling cost controls into the gap

The same spending problem is creating a product opening for cloud and infrastructure vendors. Oracle and AWS have introduced features and billing structures aimed at AI cost visibility, according to the report, while the Linux Foundation has launched a group focused on cost management.

OpenAI has also urged companies to improve visibility into AI usage and spending, track model outcomes and build governance into daily operations, according to CIO Dive. The advice is consistent across vendors, but the Harness data suggests many enterprises are still early in turning AI experimentation into managed operating expense.

This story draws on original reporting from CIO Dive.