AI/ ai · enterprise · costs · accenture

Enterprises Are Losing Track of AI Costs and Returns

Leaked audio from Accenture reveals that even the most AI-bullish firms can't predict token spend or prove the investment is paying off.

Corporate AI budgets are spiraling, and nobody has a reliable way to measure what they're getting back.

Leaked audio from a consulting firm internal meeting, first reported by 404Media, shows Accenture grappling with runaway token spend across its own operations and its clients'. Accenture's agentic AI strategy lead, Justive Kwak, described token costs rising "exponentially" as companies move from basic chatbots into agentic workflows and enterprise-wide tool deployments. Kwak put it plainly: "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." A separate data point from the same period: one unnamed company reportedly burned through $500 million in tokens in a single month — widely rumored to be Amazon, which also quietly shelved its internal AI leaderboard.

The deeper issue isn't just overspending — it's that the math is structurally broken. Token-based billing means companies pay for every input, every verbose output, every failed attempt, and every hallucination requiring correction. Kwak noted that CFO, COO, and CIO-level leadership are still asking whether AI spend is generating real value, and the honest answer from inside Accenture appears to be: we're not sure. Non-engineers, it turns out, are the bigger token consumers — using AI for tasks like converting PDFs to markdown, a workflow Accenture's own client lead flagged as a wasteful "token chewer."

This represents a meaningful reversal. Accenture previously tied promotion prospects to heavy AI adoption; Uber is now capping usage to cut costs; and OpenAI's CEO has publicly acknowledged that token costs are a serious concern. Nvidia's Jensen Huang said he'd be alarmed if engineers weren't spending at least 50% of their annual salary on AI tokens — a posture that looks increasingly like vendor enthusiasm rather than operational wisdom.

Accenture says it's building a tool called "Token IQ" to help clients manage costs, which is a tidy business pivot: first sell the AI maximalism, then sell the antidote.

TR

The Revision

Written by an AI system from the public sources credited above. How we write →