The enterprise AI cost hangover has arrived
Enterprise AI costs are the recurring expenses companies incur when employees and systems use large language models, measured in metered tokens rather than flat licenses, and they are now large enough to force executives to rethink how, where, and even whether they deploy AI at scale across everyday work. Enthusiasm-heavy, spreadsheet-light enterprise AI adoption is slamming into the hard reality of token-based billing. Companies rolled out AI tools across the org chart, celebrated high usage, and are now discovering that the meter has been running faster than the value. Uber burned through its entire 2026 AI coding budget in roughly four months, with per-engineer monthly API costs between USD 500 (approx. RM2300) and USD 2,000 (approx. RM9200). Accenture executives, in leaked audio, described “soaring token spend” as finance teams finally saw the tab.
The surprise money pit: office workflows, not engineers
The biggest shock isn’t that engineers are expensive AI users; it’s that they aren’t the main problem. Routine office tasks, scaled across thousands of employees, are quietly generating the biggest AI invoices. In leaked remarks, Accenture’s Justice Kwak said the heaviest consumers aren’t coders but office workers converting PDFs into slides, reformatting documents into markdown, and automating small chores that previously cost time, not money. This is the dark side of “AI everywhere”: each micro-task seems harmless, but token meters don’t care that it’s a status report instead of a novel feature. When the per-token cost is high, enterprises either don’t attempt ambitious AI projects, or they scope them so conservatively that the results are underwhelming. The math is ugly: lots of low-stakes usage, high aggregate cost, thin measurable benefit.
The economics test most deployments are failing
Satya Nadella’s framing is blunt: “The marginal cost of productivity improvement has to match the marginal cost of the token”. That single sentence exposes why early enterprise AI adoption is stalling. First-wave rollouts were judged by volume—how many tokens consumed, how much code AI generated, how many people turned on Copilot—rather than whether those tokens produced business value. Uber’s experience is emblematic: 95% of engineers used AI tools monthly, 70% of code commits were AI-driven, and yet the company burned its entire coding budget in about four months with weak evidence of consumer impact. Token spending without that discipline is just cost. The obstacle isn’t model quality; it is organizational discipline and cost awareness. The aggressive AI adoption mandate is quietly getting walked back at companies that haven’t admitted it publicly yet.
From open bar to metered service: the new AI budget discipline
We’ve been here before. Token governance is shaping up to be the cloud cost crisis of the AI era. Finance leaders, burned by surprise invoices, are importing FinOps-style rules into AI: token economics must connect consumption directly to business outcomes, or you are paying too much with no receipt. Deloitte’s guidance argues organizations should manage AI spend like a carefully budgeted financial resource, not an open bar with no receipt. In practice, that means usage constraints. Uber is capping employee AI tool usage after its budget blowout, according to reporting of its internal decisions. Expect what comes next to feel familiar: token quotas, role-based access tiers, consumption dashboards, and chargeback models arriving in your department. Accenture is even developing a product called Token IQ to help clients manage token consumption, according to leaked details.
Vendors under pressure: cut prices or lose the enterprise
On the other side of the table, AI vendors are facing their own economic bind. Leading models need enormous compute to stay ahead, and free consumer AI has become a data pipeline to feed those systems. Labs can’t easily stop offering free consumer access, so they push monetization onto enterprises, who end up paying the bill. Nikesh Arora calls this a trap: high token pricing for enterprises while consumers get AI for free will drive businesses toward cheaper open-source models and away from frontier offerings. He argues vendors must cut token pricing now and forward-price cheaper tokens for enterprises to unlock serious experimentation and workflow redesign. That warning isn’t theoretical. Open-source and regional models already hold a sizeable share of token volume, helped by cost differentials that can exceed 4x. If frontier providers do not fix their token pricing models, enterprise AI adoption will plateau long before it transforms productivity.
