Inside the unpredictable AI costs hitting corporate budgets

Australian corporations are confronting escalating AI costs as excessive token usage—dubbed *tokenmaxxing*—drains budgets and undermines productivity gains. Uber, for example, capped monthly employee spending on AI coding tools at US$1500 after burning through its entire 2026 AI budget in just four months due to high usage of agentic software. The initial wave of AI adoption prioritized experimentation over tangible outcomes, leading to inflated costs and misaligned priorities. Deloitte Australia’s chief technology officer, Stu Scotis, notes that organizations often judge AI success by token consumption rather than real business impact. This approach can distort value, as poorly managed AI spend diverts resources from measurable growth and efficiency. Token costs surge when AI systems process unnecessary or poorly structured data, forcing enterprises to pay for redundant computations. Elastic ANZ country manager Jeremy Pell argues that the issue stems from inefficient data retrieval, where low-quality or excessive data inflates token usage. He emphasizes that refining data inputs—using *hyper-contextualized* data—can drastically reduce costs by minimizing wasted compute power. To curb spending, companies are shifting focus from raw token usage to AI-driven outputs and measurable returns. This requires integrating enterprise data and implementing semantic layers to streamline AI queries. Without these adjustments, AI adoption risks becoming a financial burden rather than a strategic asset. The unpredictability of AI costs is also reshaping IT budgets, as variable spending contrasts with traditional SaaS licensing models. CFOs are now scrutinizing overlapping AI tools and rationalizing expenditures to align technology spend with business goals. The trend highlights a broader need for transparency in AI usage to forecast costs and justify investments effectively.