The real problem: token economics, not AI hype
Enterprise AI costs refer to the total spending businesses incur on AI tools, infrastructure, and usage-based token pricing models, and those costs are now exposing a harsh reality: many organisations are burning more money on tokens than they are gaining in productivity, forcing a sharp rethink of AI adoption strategies and budgets. In other words, the AI boom has run into basic unit economics. Boards signed off ambitious AI programs on the promise of transformation; finance teams are now staring at invoices that tell a different story. Microsoft’s chief executive has framed the issue bluntly: the marginal cost of productivity improvement has to match the marginal cost of the token. When that balance breaks, AI turns from strategic asset into uncontrolled operating expense, and that is exactly what is happening.
Uber, Accenture and the office work that broke the bank
The poster child for runaway enterprise AI costs is Uber. The company 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). Another report confirms Uber blew through its entire AI budget in four months, with the company’s chief technology officer acknowledging the overrun. That alone should unsettle any CIO. But the more surprising story is where the spending is coming from. Leaked audio from a major consulting firm describes “soaring token spend” driven not by elite engineers but by office workers converting PDFs into slides, reformatting documents into markdown, and automating tasks that used to cost nothing but time. Routine office tasks, scaled across thousands of employees, are quietly generating the biggest AI invoices.
Token pricing models are misaligned with enterprise value
This cost explosion is not an accident; it is baked into current token pricing models. A prominent security chief executive, who also sits on Uber’s board, has warned that high token pricing for enterprises while consumers get AI for free is a trap that will drive businesses toward open‑source alternatives. The logic is clear: leading AI labs need enormous compute to chase AGI, and free consumer AI has become a data pipeline they are reluctant to turn off. Monetisation pressure is therefore pushed onto enterprises, which end up paying the bill. Pricing is also shifting from simple subscriptions to seat‑fee‑plus‑pre‑committed‑token structures, locking companies into consumption whether or not it yields real value. Token costs are real and compound fast; Uber’s experience shows how easy it is to vaporise an annual budget in one quarter.
Marginal productivity AI and the pullback from unlimited use
The harsh lesson surfacing in boardrooms is that marginal productivity AI must be measured, not assumed. Microsoft’s chief executive has said, “The marginal cost of productivity improvement has to match the marginal cost of the token. That’s a management discipline”. Inside his own company there has been “a lot” of token maxing, with people over‑consuming tokens because AI tools are easy to use and usage numbers look impressive. Token spending without that discipline is just cost, and he is clear that companies pulling back from unconstrained AI use are reacting to this mismatch. The consulting giant seeing “soaring token spend” is experiencing the same pattern: the tab has arrived, finance teams are choking on it, and the biggest bills are coming from everyday workers automating mundane tasks.
What needs to change: cheaper tokens and tighter controls
If enterprises keep paying top‑shelf prices for marginal gains, this wave of AI adoption will stall. The Palo Alto Networks chief executive has a three‑part prescription. First, cut token pricing now so enterprises can experiment without budget panic. Second, show companies how their own context and data can become a competitive advantage by building proprietary layers on top of base models. Third, build tools for rapid edge‑case learning and fewer false positives, so enterprise deployment becomes an operational capability rather than a research project. Meanwhile, governance is tightening. High token prices already create a chilling effect, pushing CIOs to restrict AI use and focus on efficient deployments instead of open‑ended experimentation. Expect token quotas, role‑based access, consumption dashboards, and chargeback models to become standard in AI‑enabled workplaces. The next phase will not be about more AI; it will be about cheaper tokens, clearer outcomes, and strict discipline.
