Leaked Accenture audio: AI token costs spiraling, ROI unclear

Key Takeaways

• Leaked Accenture meeting audio shows enterprises struggling with runaway AI token costs and no reliable ROI metrics
• One unnamed company reportedly spent $500 million on AI tokens in a single month
• Amazon killed its AI leaderboard; Uber is capping AI use; companies are switching to cheaper models

Leaked audio from consulting giant Accenture reveals what many enterprise leaders suspected but few said aloud: companies have no idea whether their AI spending is paying off, and the bills are growing fast. The recording, obtained by 404 Media, captures Accenture staff wrestling with how to help clients rein in token costs that have ballooned beyond anyone's projections.

The term 'tokenmaxxing,' a somewhat ironic label for the recent corporate rush to use AI tools as aggressively as possible, appears to be hitting its limits. One unnamed company reportedly burned through $500 million in AI tokens in a single month. Amazon has quietly shut down its internal AI leaderboard. Uber is capping employee AI usage to control costs.

What did the Accenture leak reveal?

The leaked audio features Justive Kwak, Accenture's agentic AI strategy lead, describing a growing crisis in enterprise AI spending. 'What we're seeing right now is just rapid escalation in AI token spend,' Kwak said, pointing to companies scaling from simple chatbots to agentic workflows, automation, and enterprise-wide deployment of tools like Copilot, Claude Code, and Codex.

“It's really not a niche problem. It is a problem that every enterprise will face if they are bullish on AI, if they haven't already.”

— Justive Kwak, Accenture

The meeting made clear that even Accenture, a firm that previously told employees they risked missing promotions if they didn't embrace AI heavily enough, is now questioning its own usage. Kwak noted that token spend was increasing 'exponentially, as more and more people are starting to use AI.'

The core problem goes beyond sticker shock. Accenture staff acknowledged that trivial tasks being offloaded to AI are causing massive overspend, particularly when agentic AI systems are involved. These autonomous agents can call other AI services repeatedly to complete tasks, multiplying costs in ways that are difficult to predict or control.

Why is measuring AI ROI so difficult?

The shift to token-based billing changed the economics of enterprise AI overnight. Where subscriptions once offered predictable pricing, companies now pay for every token they input and every token the AI outputs. That includes verbose responses, mistakes, hallucinations, and the follow-up corrections they require.

Kwak framed the challenge bluntly: 'Leadership, especially at the CFO, COO, and CIO level, are still asking the question of whether they're getting value from what we're spending on in the context of AI.'

The math doesn't cooperate. When you can't predict how many tokens a task will consume, whether the AI will complete it correctly on the first attempt or the third, or whether the output will be useful or fabricated, traditional ROI calculations fall apart. 'We're hitting this inflection point where AI is becoming material to the cost structure; spend is becoming very unpredictable,' Kwak said.

How are companies responding?

The corporate AI gold rush is cooling. Amazon reportedly killed its AI leaderboard. It's rumored to be the mystery company that spent half a billion dollars on tokens in one month. Uber is capping AI use to cut costs. Axios reported in late May that multiple CEOs are switching to more affordable models and monitoring employee usage more closely.

Some developers have adopted what's called the 'caveman trick' to reduce token consumption, essentially writing prompts that minimize back-and-forth with AI systems. Even OpenAI CEO Sam Altman acknowledged that token costs have become a major concern for users.

Nvidia CEO Jensen Huang's statement earlier this year that he'd be 'alarmed' if engineers weren't spending at least 50% of their annual salary on AI tokens now reads like advice from a different era. The mandate to use AI constantly is giving way to questions about when it actually makes sense.

What happens next for enterprise AI spending?

The leaked Accenture audio suggests a reckoning is underway. Large language models excel at specific, well-defined tasks. Their value across broader applications remains uncertain, and the economics are forcing that uncertainty into the open.

Companies that spent the past year racing to integrate AI everywhere are now asking harder questions: Which use cases actually justify the cost? How do you budget for a tool whose consumption is inherently unpredictable? What's the actual productivity gain, measured in something other than 'we're using AI'?

Accenture's internal struggles mirror what its clients are experiencing. The consulting firm that pushed employees toward AI is now trying to figure out the same questions everyone else is: what this technology costs, what it delivers, and whether the two numbers make any sense together.

Frequently Asked Questions

What is tokenmaxxing?

Tokenmaxxing refers to the corporate practice of using AI tools as aggressively as possible, often encouraged by internal mandates or leaderboards, without close attention to cost or measurable ROI.

Why are AI token costs so unpredictable?

Token consumption depends on prompt length, output verbosity, number of retries, and whether agentic AI systems call other AI services. None of these variables are fully controllable by the user.

How much did one company reportedly spend on AI tokens in a month?

One unnamed company, rumored to be Amazon, reportedly spent $500 million on AI tokens in a single month.

What is agentic AI and why does it increase costs?

Agentic AI refers to autonomous AI systems that can take actions, call other tools, and make decisions without human intervention. Each action in a chain consumes additional tokens, multiplying costs rapidly.

Are companies cutting back on AI spending?

Yes. Amazon killed its AI leaderboard, Uber is capping AI use, and multiple companies are switching to cheaper models and monitoring usage more closely.

Source: Latest from Tom's Hardware