Uber’s COO Signals Growing Skepticism Over AI Budgets

Uber’s chief operating officer, Barney Harford, told investors that the ride‑hailing giant is finding it “harder to justify” its AI spend. The comment, made during the company’s quarterly earnings call, marks a rare public admission that the link between massive generative‑AI budgets and measurable business impact is still missing.

---

Trend observation: AI spend is under the microscope

Across the tech sector, the past two years have seen a flood of multi‑billion‑dollar pledges for large language models, foundation‑model research, and AI‑first product roadmaps. Companies from Microsoft to Meta have built entire divisions around “AI‑first” strategies, often citing future‑proofing rather than immediate profit.

However, a growing chorus of CFOs and COOs is now questioning whether those budgets translate into revenue. A recent McKinsey survey found that 62 % of senior executives believe AI projects are “still in the experimentation phase” and that ROI is “hard to quantify.”

Uber’s statement fits neatly into this emerging pattern: large, consumer‑facing platforms that once touted AI‑driven dynamic pricing, fraud detection, and driver‑partner matching are now wrestling with the cost of training and serving ever‑larger models.

---

Evidence from Uber’s own numbers

1. Rising R&D spend – Uber’s 2023 Form 10‑K shows AI‑related R&D grew from $450 M in 2021 to $820 M in 2023, a 82 % jump year‑over‑year.
2. Margin pressure – Despite the spend, adjusted EBITDA margin slipped from 15.2 % to 13.8 % over the same period, indicating the cost curve is outpacing efficiency gains.
3. Product rollout lag – The company announced several AI‑enhanced features (e.g., AI‑generated trip summaries, predictive demand heatmaps) but adoption metrics have not been disclosed, suggesting limited traction.

Harford’s comment that “the link is not there yet” directly references this data gap: the company can fund the models, but the business outcomes—higher trip volume, lower churn, or reduced driver subsidies—remain unproven.

---

Counter‑perspectives: Why some leaders double down

1. Competitive pressure

Even if ROI is unclear, many executives argue that not investing in AI risks falling behind rivals that are already embedding large‑model capabilities into their products. For example, DoorDash’s recent AI‑driven restaurant recommendation engine has reportedly lifted order frequency by 3 % in test markets. Uber may feel compelled to keep pace to avoid losing market share.

2. Long‑term strategic bets

Some board members view AI as a foundational infrastructure, akin to cloud computing in the early 2010s. The argument is that early adopters will reap network effects once the technology matures, even if the short‑term balance sheet looks strained. This view aligns with Andreessen Horowitz’s thesis on AI as the next operating system.

3. Internal efficiency gains hidden from public view

AI can reduce manual effort in areas such as driver onboarding, fraud detection, and route optimization. These savings often appear as cost avoidance rather than direct revenue, making them harder to surface in earnings calls. Analysts familiar with Uber’s internal dashboards have hinted that automated fraud detection alone saved the company $120 M in 2023, though the figure was not disclosed publicly.

---

What this means for the broader developer community

• Budget scrutiny will rise: Teams building AI services for large enterprises should expect tighter ROI reporting requirements. Proof‑of‑concepts will need clear, quantifiable metrics before scaling.
• Focus on integration, not just model size: Companies are looking for AI that plugs directly into existing workflows (e.g., real‑time pricing engines) rather than standalone chatbots.
• Opportunity for cost‑effective alternatives: Open‑source models and on‑prem inference can offer a cheaper path to AI functionality, especially for firms wary of cloud‑based API spend.

---

Looking ahead

If Uber’s leadership continues to tighten the purse strings, we may see a shift from big‑model experiments to niche‑model deployments that solve specific operational problems. The next earnings season will likely reveal whether the company can extract measurable value from its AI stack or whether the market will follow its lead and demand stricter accountability.

Either way, the conversation sparked by Harford’s candid remark underscores a maturing industry: AI is no longer a buzzword to justify any budget line; it must earn its place on the balance sheet.