For Every $1 Spent on AI, Companies Pay $0.44 Fixing Bugs, $0.27 Rewriting Code, and $0.11 on Review Delays, Study Finds

Weeks after TechStartups revealed how one company accidentally spent $500 million on Claude AI in a single month, new research suggests runaway token bills are only part of the enterprise AI cost equation.

In late May, TechStartups reported on an anonymous company that accidentally spent $500 million on Anthropic’s Claude AI in a single month after failing to set usage limits for employees. The story highlighted how quickly enterprise AI spending can spiral out of control once organizations encourage widespread adoption without guardrails.

A new study suggests runaway token bills may be only part of the problem.

For Every $1 Spent on AI Tokens, Only 18 Cents Ultimately Reaches Production

Researchers at Entelligence AI Research analyzed data from 2,444 companies and found that for every $1 organizations spend on AI tokens, only 18 cents ultimately reaches production. The remaining 82 cents is consumed fixing AI-generated bugs, rewriting code, and dealing with review and merge delays before software ever reaches customers.

Credit: Entelligence Research

Up to 82% of AI Engineering Spend Lost to Bugs, Rewrites, and Delays: Study Finds

The findings paint a sobering picture of enterprise AI adoption. Companies spend $0.44 fixing AI-generated bugs, $0.27 rewriting or reworking AI-generated code, and $0.11 navigating review and merge friction. The study concludes that 82% of every AI dollar never reaches the final product.

“An analysis of 1M+ pull requests across 2,444 engineering organizations. More AI spend. More code volume. More production failures. We measured where AI engineering effort actually goes, how code review has responded to 2.6× volume growth, and why the reactive work treadmill keeps accelerating. The findings: $0.82 of every AI dollar is consumed before a single feature reaches users,” Entelligence AI wrote.

The numbers arrive at a time when businesses across Silicon Valley are reassessing the true cost of generative AI. Earlier this year, the focus centered on getting employees to use AI as much as possible. Today, executive conversations are shifting to a different question: whether those growing AI bills translate into measurable business value.

“Entelligence AI surveyed 2,444 companies and found that for every $1 spent on AI tokens, $0.44 covers bug fixes, $0.27 rewrites AI-generated code, and $0.11 vanishes into review and merge delays,” Yahoo Finance noted.

The change in sentiment is already showing up inside some of the world’s largest technology companies.

The Hidden Cost of AI-Generated Code

Source: Entelligence Research

Uber reportedly exhausted its entire 2026 AI coding budget by April. Microsoft quietly suspended access to Anthropic’s Claude Code for most employees after previously making the coding assistant widely available internally. Salesforce is said to spend roughly $300 million annually on Anthropic’s technology, prompting CEO Marc Benioff to advocate for routing requests to the most cost-effective AI models based on the task.

The shift marks a sharp departure from the industry’s recent obsession with “tokenmaxxing,” a term that emerged in 2025 to describe maximizing AI token consumption across organizations. Many companies encouraged employees to use AI for nearly every task, viewing higher usage as proof that digital transformation efforts were succeeding.

That strategy produced unintended consequences.

Employees increasingly relied on enterprise AI tools for routine requests ranging from writing emails and meeting notes to generating software code. Token consumption soared, yet many organizations struggled to determine whether the additional spending translated into better products or higher revenue.

Investor Shruti Gandhi captured the problem with a simple comparison.

“A tokenmaxxing company is like a business that measures productivity by leaving all the lights on—spending more money doesn’t mean producing more.”

The Entelligence findings align with a broader body of research suggesting that measuring AI adoption by usage alone misses much of the economic picture.

Engineering operations platform Faros AI previously studied 20,000 developers over two years and found that software output increased alongside AI usage, but bug rates and code rewrites climbed too. Engineering management platform Jellyfish reached a similar conclusion. Engineers who consumed the most AI tokens were roughly twice as productive as lighter users, yet they spent nearly 10 times more tokens to achieve those gains.

Nicholas Arcolano, Head of Research at Jellyfish, told TechCrunch that per-developer AI consumption increased 18.6 times over a nine-month period, driven largely by increasingly autonomous AI coding tools.

“Whether extreme spend pays off comes down to the ultimate business value of shipped code (e.g. revenue), which most companies still can’t measure,” Arcolano said.

That measurement gap is becoming one of the biggest challenges facing enterprise AI.

Tokenmaxxing: Your AI Bill Is About to Exceed Your Developer’s Salary

A JPMorgan report recently warned that AI token costs are consuming internet profits, reflecting growing concerns that spending is outpacing measurable returns. At the same time, only 14% of CFOs report seeing clear, measurable returns from their AI investments, according to data cited in the research.

Uber Chief Operating Officer Andrew Macdonald echoed those concerns during a podcast discussion, saying it remains difficult to connect improvements in individual employee productivity with overall business performance. Engineers may write code faster or employees may complete reports more quickly, yet those gains do not automatically translate into higher revenue or better customer experiences.

Former Microsoft Chief AI Officer Sophia Velastegui pointed to another issue that many companies overlook.

“Most people default to automating tasks they dislike, rather than those that add the most value to the company.”

That observation helps explain why AI can generate impressive productivity statistics without materially improving business outcomes.

The growing scrutiny around AI spending is creating demand for an entirely new category of enterprise software focused on cost visibility and financial governance.

Meanwhile, OpenAI’s Head of Enterprise Alexander Embiricos said that customer conversations have changed dramatically over the past six months.

“Six months ago, I would have a conversation with a customer and it would be all about ‘What can it do? Is it good enough?'” Embiricos told TechCrunch. “Our conversations are never about that now. Now the conversations are about, ‘hey, we’re spending so much. What visibility do you have? What auditability do you have? What token controls do you have? What is the efficiency of your models?'”

That demand has prompted companies including Ramp, Datadog, New Relic, CloudZero, Harness, Pay-i, Paid, Jellyfish, Waydev, and Faros AI to introduce products that monitor AI spending, measure token efficiency, and connect AI costs to business outcomes.

The Linux Foundation has gone a step further by launching the Tokenomics Foundation, an industry initiative aimed at creating common standards for AI token accounting, cost measurement, and financial reporting.

J.R. Storment, Executive Director of the FinOps Foundation, said the conversation shifted dramatically this spring.

“In April and May, I started hearing from companies: ‘Oh my god, we are 3x over our entire 2026 token budget and it’s only April,'” Storment told TechCrunch. “We started hearing existential crises, and the whole conversation shifted from tokenmaxxing and ‘go fast’ to ‘we need guardrails, how do we control this?'”

Goldman Sachs projects global AI token usage will increase 24-fold by 2030, suggesting spending pressures are unlikely to ease anytime soon.

The industry’s priorities, though, are already changing.

Earlier this year, success was often measured by how many AI tokens employees consumed. New evidence suggests that metric tells only part of the story. Companies are beginning to judge AI by a different standard: how much business value reaches customers after accounting for bugs, rewrites, reviews, and the hidden work that follows every generated line of code.

For enterprises racing to adopt AI, that distinction could prove far more valuable than simply buying more tokens.