Uber's AI Spending Crisis: When Token Consumption Doesn't Translate to Features

Uber has hit a hard ceiling on AI spending ROI. Uber president and chief operating officer Andrew Macdonald revealed to Rapid Response that the company exhausted its full 2026 AI budget by late May—just four months into the year—and lacks evidence that accelerating token consumption is yielding proportional gains in shipped features. This admission surfaces a widening gap between infrastructure spending and measurable product impact across the AI economy.

The Budget Burn and Feature Gap

According to The Verge AI, Macdonald stated plainly that Uber cannot draw a causal line between token consumption for Claude Code and customer-visible improvements. “That link is not there yet,” he said, adding that while underlying metrics trend upward, the company struggles to correlate infrastructure spending with actual feature delivery. Uber spent $3.4 billion on research and development in 2025, a 9 percent year-over-year increase; the 2026 trajectory suggests an even steeper climb, given the four-month budget exhaustion.

The core tension Macdonald identified is quantitative: rising token counts do not automatically translate to measurable consumer utility. Some features ship, he acknowledged, but attribution remains elusive. “I think maybe implicitly there is more that is getting shipped,” Macdonald said, “but it’s very hard to draw a line between one of those stats and producing 25 percent more useful consumer features.”

The Headcount-for-Tokens Trade-Off Under Scrutiny

Uber CEO Dara Khosrowshahi had previously framed the company’s approach as a direct substitution: hire fewer humans, spend more on AI infrastructure. Macdonald’s pushback suggests that trade is harder to defend without demonstrated feature gains. “We’re going to have to start talking about token consumption and the associated cost versus headcount,” he said. “So if you’re not actually able to draw a direct line to how much useful features and functionality you’re shipping to your users, that trade becomes harder to justify.”

This reframing is significant: it moves the accountability from raw token growth to outcome-based metrics. Token consumption alone, in Uber’s experience, is not a proxy for product value.

Why This Matters

Uber’s public doubts about AI spending efficacy signal a broader inflection point in enterprise LLM adoption. If a company with Uber’s scale and resources cannot establish clear ROI on aggressive AI investment—despite using a state-of-the-art model like Claude Code—other organizations will face similar pressure to demonstrate impact before scaling further. This may constrain near-term growth in token consumption and cloud spend for inference, particularly among companies still in early stages of LLM integration. The result could be a recalibration of AI budgets away from open-ended token growth toward task-specific, metrics-driven deployments with clear feature attributions.