Uber President Questions AI Spending as Burn Rate Outpaces Product Gains
Uber burned its full 2026 AI budget in four months, and president Andrew Macdonald says the spending is getting harder to justify without clearer links between token consumption and shipped product value.
Uber president Andrew Macdonald says AI spending is getting "harder to justify." The company burned through its entire annual AI budget in the first 4 months of 2026, and leadership is now interrogating the economics of token-heavy developer workflows. The question is simple: when does rising compute spend translate into commensurate product value?
The burn rate and the friction
Macdonald laid out the mechanics on the Rapid Response podcast. Uber consumed its full yearly AI allocation in the opening quarter of 2026, driven largely by developer adoption of Anthropic's Claude Code and the token consumption that followed.
Uber's broader R&D bill hit $3.4 billion in 2025, a 9% jump year over year. CEO Dara Khosrowshahi has signaled the company will absorb AI overhead by slowing hiring rather than pausing investment. Macdonald framed the trade-off bluntly as a choice between "token consumption and the associated cost versus headcount." Internal technical indicators are moving in what he called an "astronomical direction" — but the bridge to commercial impact has not yet been built.
The measurement trap
The core problem is visibility. Macdonald said it remains "very hard to draw a line" between AI usage spikes and shipped product improvements. When developers consume tokens faster than engineers ship features, the promised efficiency gain disappears.
The fix requires shifting KPIs away from raw API volume toward outcome-based signals: feature completion rates, user engagement, retained revenue per dollar of inference. Until that correlation tightens, Uber is stuck in a familiar automation paradox — scaling inputs without scaling outputs, which compresses margins. That dynamic also creates an opening for vendors who can show transparent ROI on LLM integration, and pressure on incumbent tool providers to defend their pricing.
Our read
We read this as calibration, not retreat. Uber is correctly identifying that developer velocity tools turn into cost centers the moment they stop being force multipliers. Trading headcount for compute capacity mirrors the structural shift visible elsewhere, including Meta's 8,000 layoffs.
The risk is execution. If the commercial translation keeps lagging, boards will demand sharper gates on tooling procurement. The winners here won't be the companies with the fastest models; they'll be the ones who can prove every dollar of inference flows into retained revenue. The question for Uber is whether it can tighten that feedback loop before the burn rate forces a hard stop on experimentation.
