Uber President: AI Spending Is Getting 'Harder to Justify'
Uber president Andrew Macdonald says it's difficult to draw a direct line between AI investment and deliverable product features, signaling growing executive skepticism about AI ROI at major tech companies.
Original sourceUber president Andrew Macdonald has become one of the more prominent enterprise voices pushing back on the AI spending narrative, telling an interviewer that the company finds it increasingly 'hard to draw a line' between dollars spent on AI and concrete features delivered to users. The comments are notable not because Uber is pulling back from AI entirely, but because a sitting president at a top-tier tech-adjacent company is saying out loud what many CFOs have reportedly been whispering internally for the past year.
The tension Macdonald is describing is structural. Most large companies are buying AI capabilities as a bundle — cloud compute, API access, fine-tuning, tooling — and attempting to map that spend retroactively to shipped features or productivity gains. That accounting is genuinely hard, and unlike traditional software procurement, there's no clean license-per-seat metric to anchor ROI conversations. The cost centers are real; the attribution is fuzzy.
This is not an isolated skepticism. A wave of enterprise earnings calls in 2025 featured similar hedging language, even as AI infrastructure vendors continued posting record quarters. The disconnect between infrastructure spend and demonstrable application-layer value is becoming a board-level conversation across industries, not just at Uber. Macdonald's willingness to say it plainly may be more significant than the statement itself — it gives other executives cover to ask the same question.
The irony is that Uber, as a company built on real-time routing, dynamic pricing, and demand forecasting, is arguably better positioned than most to find concrete ML ROI. If they're struggling to justify the spend, the companies with less obvious AI use cases have a much harder problem.
Panel Takes
The Skeptic
Reality Check
“This is the ROI reckoning that was always coming, and the fact that it's Uber saying it matters — these are people who have been running production ML systems for a decade, not a legacy company that bought a Copilot license and got confused. The honest read is that 'AI investment' has been bundled with cloud infrastructure spend in ways that make the accounting deliberately opaque, and executives are finally demanding unbundled answers. My prediction: in 12 months, the companies that survive this scrutiny are the ones who can name a specific feature, attach a revenue or cost number, and point at the model that shipped it — everyone else gets their AI budget cut in half.”
The Founder
Business & Market
“The unit economics problem Macdonald is naming is real: AI spend flows into an infrastructure layer that gets shared across dozens of potential use cases, which means no single product team owns the cost, and no single product team can claim the credit. That's a classic internal accounting failure dressed up as an AI problem — and it's the kind of thing that gets fixed by re-orging the P&L, not by cutting AI investment. The companies that figure out how to assign AI costs to specific value-generating features will have a structural advantage over the ones still arguing about shared cloud bills in their quarterly business reviews.”
The Futurist
Big Picture
“The thesis embedded in Macdonald's comments is actually falsifiable: AI investment at the application layer only pays off if the capability compounds into workflow changes that wouldn't have happened otherwise — not if it's a faster version of something engineers were already doing. The second-order effect here is that this kind of public skepticism from operators accelerates pressure on AI vendors to build attribution tooling, not just capability benchmarks. Watch for 'AI ROI dashboards' to become a genuine product category in the next 18 months — not because enterprises need the data, but because the vendors need the talking points.”
The PM
Product Strategy
“Macdonald is describing a product strategy failure more than an AI failure — if your org can't draw a line from an investment to a shipped feature, the job-to-be-done was never clearly defined before the budget was allocated. The companies winning on AI right now hired product managers before they hired ML engineers: they identified the specific user problem, scoped the AI's role in solving it, and set a measurable outcome before writing a check. Uber clearly has the data and the use cases to justify AI spend; the problem is someone skipped the step where you write down what problem you're solving.”