Is it still cheaper to be human? Uber burned through AI Token's annual budget in four months, but didn't achieve the expected returns.

Uber Chief Operating Officer reveals the company burned through a full year's AI budget in four months but struggles to demonstrate equivalent benefits, even slowing recruitment as a result. As AI costs soar, tech companies are beginning to reflect on the true returns of blind AI investment.

Uber COO discusses AI cost dilemma, burning through a year's budget in four months

Famous delivery and driver platform Uber's COO Andrew Macdonald revealed in an interview with Rapid Response that the company is increasingly unable to prove the cost-effectiveness of AI.

He pointed out that previously Uber CTO Praveen Neppalli Naga had disclosed that the company exhausted its entire AI budget in just four months of 2026, forcing leadership to weigh token consumption against employee headcount. If the expenditure cannot be directly linked to practical features released to users, it becomes hard to justify.

He stated that for users who don’t pay bills, AI feels free, but someone still has to foot the bill.

Uber burns through token budget too quickly, even slowing down hiring

The reason why Uber’s AI costs are difficult to justify is mainly because a clear causal relationship cannot be established.

Andrew Macdonald said that although 25% of Uber’s code is driven by AI collaboration, the team finds it hard to prove that this results in 25% of consumer-facing features.

Uber CEO Dara Khosrowshahi even mentioned during the earnings call that to cope with increased AI investments, the company is slowing down hiring and reducing the employment of new human staff.

Data shows that Uber’s R&D expenditure in 2025 reached $3.4 billion, a 9% increase from the previous year, facing pressure to rationalize expenses through actual feature releases.

Image source: Rapid Response Uber COO Andrew Macdonald discusses the company's AI cost-effectiveness dilemma

Blind pursuit of token maximization sparks backlash, companies begin to rethink evaluation mechanisms

Amid the trend of Silicon Valley giants blindly pursuing token maximization (Tokenmaxxing) and incorporating it into employee evaluations, some companies are starting to reflect.

According to Business Insider, the well-known online language learning platform Duolingo previously included AI usage in employee assessments but was later questioned internally whether it was “used for the sake of use.”

Duolingo CEO Luis von Ahn admitted in an April interview that this approach made it feel like the team was not responsible for the final results, merely pushing tools that were not suitable, leading to the eventual withdrawal of the policy.

When token consumption outpaces price reductions

The wave of promoting AI to boost productivity has begun to face backlash within tech giants.

Microsoft recently required employees to switch to using its own GitHub Copilot CLI instead of third-party tools like Claude Code, mainly because token consumption costs have skyrocketed; AI open-source assistant OpenClaw founder Peter Steinberger also claimed that his team spent over $1.3 million in just one month.

Jowi Morales, a writer for Tom’s Hardware, pointed out that current AI tools only offer limited productivity gains, and their usage costs could even be more expensive than hiring humans. If token consumption outpaces price reductions, strategies to replace human labor with AI to cut costs may backfire.

Related reading:
All-you-can-eat AI is gone! GitHub Copilot costs too much to sustain, switching to usage-based billing from June 1