Why Uber has Already Burned Through its AI Budget

· Source: AI Magazine · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Corporate Strategy & Leadership · Depth: Fundamental Awareness, short

Summary

Uber has depleted its 2026 AI budget four months into the year due to extensive adoption of AI coding tools like Claude Code and Cursor, according to CTO Praveen Neppalli Naga. The company's R&D expenses rose 9% year-over-year in 2025 to US$3.4 billion, with AI being a primary cost driver. Engineers' widespread use of these tools, partly driven by internal leaderboards and "tokenmaxxing" trends, has resulted in monthly API costs between $500 and $2,000 per person. Approximately 11% of Uber's live backend code updates are now fully written by AI agents, and nearly 95% of engineers use AI tools monthly, contributing to 70% of committed code. Uber is now exploring OpenAI's Codex to expand its AI stack and is shifting towards an "agent engineering" model where AI systems manage coding, testing, and deployment.

Key takeaway

For VPs of Engineering managing AI initiatives, your teams' enthusiastic adoption of AI coding tools can quickly exhaust budgets, as seen with Uber's 2026 budget depletion. You should implement robust cost tracking for AI API usage and critically evaluate the actual return on investment of high token consumption, rather than solely incentivizing usage, to prevent unexpected financial strain and ensure sustainable AI integration.

Key insights

Aggressive AI tool adoption can rapidly deplete budgets, necessitating cost management and strategic expansion.

Principles

Method

Uber encouraged AI tool adoption through internal leaderboards, leading to high usage of Claude Code and a plateau in Cursor interest, ultimately exhausting their AI budget.

In practice

Topics

Best for: VP of Engineering/Data, Director of AI/ML, CTO, Executive

Related on AIssential

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI Magazine.