After reportedly burning through its entire annual AI budget by April 2026, Uber is publicly questioning whether the spending spree is actually paying off. In an interview with Rapid Response, Uber president and chief operating officer Andrew Macdonald said the company cannot yet draw a clear line between rising token consumption—especially for tools like Claude Code—and meaningful improvements to consumer-facing features.

“That link is not there yet, right? I think maybe implicitly there is more that is getting shipped, but it’s very hard to draw a line between one of those stats and, ‘Okay, now we’re actually producing 25 percent more useful consumer features,’” Macdonald told the podcast. He added that over the coming quarters and years the connection might become clearer, but today it remains difficult even as underlying metrics trend in “a really astronomical direction.”

Uber spent $3.4 billion on research and development in 2025, a 9% increase from the previous year. The company’s foray into artificial intelligence has been ambitious, with investments spanning generative AI, autonomous vehicle technology, and internal efficiency tools. Yet Macdonald’s comments reveal a growing tension between the hype around AI and the practical outcomes. “We’re going to have to start talking about token consumption and the associated cost versus headcount,” he said. “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.”

The remarks come just weeks after Uber CEO Dara Khosrowshahi told analysts that the company is compensating for rising AI spending by hiring fewer human employees. That trade-off—replacing some human roles with AI-driven automation—is central to the debate about whether big tech companies are overinvesting in artificial intelligence without commensurate returns.

The AI spending dilemma

Uber is not alone in facing this scrutiny. Across the tech industry, executives are under pressure to demonstrate that billions of dollars funneled into AI research and deployment are translating into real business value. While cloud providers like Microsoft, Google, and Amazon have seen revenue bumps from AI services, many user-facing companies have struggled to quantify the benefits. Macdonald’s comments highlight a common pain point: metrics like token consumption, model inference costs, and GPU utilization are soaring, but they do not directly correlate with product improvements that customers notice or pay for.

Token consumption refers to the number of units of text processed by large language models (LLMs) when generating responses. Tools like Claude Code—an AI coding assistant developed by Anthropic—consume tokens at a high rate. Uber has reportedly deployed such tools across its engineering teams to accelerate software development. However, Macdonald’s suggestion that fast token growth is not matched by a proportional increase in “useful consumer features” raises questions about whether the investment is efficient.

The underlying issue is that AI models are still imperfect. They can hallucinate, produce inconsistent results, and require careful prompt engineering to yield usable outputs. The cost of running these models at scale—both in terms of computational resources and developer time spent refining prompts—can outweigh the productivity gains they provide for certain tasks.

Historical context: Uber’s R&D trajectory

Uber’s R&D spending has climbed steadily over the past five years, driven largely by its autonomous vehicle division, Uber ATG (now part of Aurora Innovation), and later by its move into generative AI. In 2021, the company spent $2.4 billion on R&D. That figure rose to $2.8 billion in 2022, $3.1 billion in 2023, and $3.1 billion in 2024 before hitting $3.4 billion in 2025. The acceleration in 2025 aligns with the broader industry push into generative AI following the release of ChatGPT in late 2022.

By comparison, Uber’s total revenue in 2025 was $49.8 billion, meaning R&D represented about 6.8% of revenue. While that percentage is not out of line with other tech giants—Meta spent about 29% of its revenue on R&D in 2025—the absolute scale of Uber’s AI investments has drawn attention. The company has been an early adopter of Anthropic’s Claude models, using them for everything from customer support chatbots to internal code generation tools.

Macdonald’s interview suggests that Uber may be hitting a point of diminishing returns. The company reportedly exhausted its annual AI budget in just four months, implying that either the budget was too small relative to actual usage or that costs are growing faster than anticipated. The latter seems more likely given the explosive growth of token consumption—a phenomenon seen across the industry as LLMs become more deeply integrated into daily operations.

Broader industry implications

Uber’s skepticism is part of a growing chorus of voices questioning the ROI of AI. In recent months, executives from companies like Salesforce, Zoom, and even Microsoft have noted that while AI is promising, it has not yet produced the kind of productivity revolution that early hype suggested. A study from MIT in early 2026 found that only 23% of companies using generative AI reported significant productivity gains, while the majority saw marginal improvements or even decreases due to the time needed to verify AI outputs.

For Uber, the stakes are particularly high because the company operates on thin margins. Its ride-hailing and delivery businesses are competitive, and any cost overruns can quickly eat into profits. Shifting from human employees to AI tools requires careful cost-benefit analysis, especially when the tools themselves are expensive to run at scale. Macdonald’s focus on “token consumption versus headcount” suggests that Uber is considering making explicit trade-offs: either reduce the number of engineers and rely more on AI, or scale back AI spending and hire more humans.

Another factor is the rapidly evolving regulatory landscape. The European Union’s AI Act, which entered into force in August 2024, imposes strict requirements on high-risk AI systems. Companies that use AI for critical decisions—such as driver safety or dynamic pricing—must ensure transparency and accountability. Compliance costs add another layer to the spending justification puzzle. Uber has already faced regulatory challenges over its use of algorithms for pricing and surge, and any new AI systems will need to pass muster with regulators.

Moreover, the market for AI talent has inflated wages. Data scientists and machine learning engineers command salaries often exceeding $400,000 in total compensation at top tech firms. Uber has been competing for that talent, further driving up costs. While AI can augment existing teams, it also requires specialized personnel to manage and maintain models—a cost that is often hidden in R&D budgets.

The road ahead for Uber

Macdonald’s frank assessment does not mean Uber is abandoning AI. The company continues to invest in autonomous vehicle technology through its partnership with Aurora, and it relies on machine learning for route optimization, fraud detection, and demand forecasting. What it does signal is a more cautious approach. Uber may begin to impose stricter cost controls on AI usage, requiring internal teams to justify each deployment with clear metrics tied to user outcomes.

This could lead to a consolidation of AI tools. Rather than using multiple LLMs for different tasks, Uber might standardize on one or two models to reduce complexity and cost. Anthropic’s Claude, being a major partner, is likely to remain central. But the company may also explore on-device AI solutions that reduce cloud inference costs, especially for repetitive tasks like suggesting routes or processing driver feedback.

In the longer term, the challenge is to develop AI that is not only powerful but also cost-effective and reliable. Macdonald’s hope that “over the coming quarters and years, maybe that will become clearer” reflects a belief that the technology is still maturing. As models become more efficient—through techniques like quantization, distillation, and better hardware—the cost per token should fall, making the ROI equation more favorable. Until then, Uber will have to grapple with the uncomfortable reality that its AI spending is growing faster than its ability to measure the benefits.

The company’s next earnings report, expected in July 2026, will be closely watched for any signs of a pivot. Investors have already begun asking about AI ROI during quarterly calls. If Macdonald’s comments are any indication, Uber is preparing stakeholders for a more disciplined approach—one where every token consumed must earn its place in the budget.

Source: The Verge News