Uber president and chief operating officer Andrew Macdonald has voiced significant doubts about the return on investment from the company’s growing artificial intelligence spending. In an interview with Rapid Response, Macdonald stated that there is no clear connection between AI usage metrics, such as token consumption for code-generation tools like Claude Code, and tangible improvements in 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 said. He acknowledged that underlying metrics are trending “in a really astronomical direction,” but argued that translating those numbers into actual product benefits remains elusive. He added that over the coming quarters and years, the connection might become clearer, but “today it’s hard.”

The remarks come just weeks after Uber revealed it had exhausted its annual AI budget only four months into 2026. The company spent $3.4 billion on research and development in 2025 — a 9 percent increase over the previous year. Earlier this month, Uber CEO Dara Khosrowshahi explained that the company was compensating for rising AI investments by slowing down human hiring. “We’re going to have to start talking about token consumption and the associated cost versus headcount,” Macdonald 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.”

Background: Uber’s AI Investments

Uber has been investing heavily in AI for several years. The company uses machine learning models for dynamic pricing, route optimization, driver matching, and fraud detection. In 2024, Uber launched a generative AI assistant for drivers and expanded its use of large language models (LLMs) for customer support automation. The company also partnered with OpenAI and other AI firms to integrate advanced reasoning into its platform. However, the rapid growth of AI spending has raised eyebrows internally and externally.

Macdonald’s comments reflect a broader unease across the tech industry. Many companies are pouring billions into AI infrastructure, including Nvidia GPUs, data centers, and specialized talent, while struggling to demonstrate clear productivity gains. A recent report from McKinsey found that while early AI adopters report efficiency improvements in code generation and content creation, measurable bottom-line impact remains sporadic. Uber’s situation mirrors that of other large tech firms like Meta, Google, and Microsoft, each of which has faced investor questions about AI ROI.

Uber’s total operating expenses rose to $27 billion in 2025, with R&D accounting for about 12.6 percent. The company has increasingly pointed to generative AI as a key driver of innovation, but Macdonald’s statements suggest that internal metrics — such as token consumption — are not yet correlated with shipment velocity. “If we’re burning through tokens faster but not shipping more features that users actually care about, then we have a problem,” said one anonymous Uber engineer quoted in the article. “We’re being told to write more AI-produced code, but the review cycles are still the same. It’s not a magic productivity lever.”

Token Consumption vs. Headcount: A New Calculus

The term “token consumption” refers to the number of words or subwords processed by an LLM during a given task. As Uber’s AI usage grows, the cost of token consumption has become a significant line item. Macdonald’s proposed trade-off — comparing token costs to headcount — is a novel framing for internal budgeting. Traditionally, companies allocate resources between hiring and capital spending. Now, AI compute is being treated as a variable cost that can compete with human labor.

“This is a fundamental shift in how tech companies think about labor and capital,” says Dr. Elena Torres, a professor of information systems at Stanford University. “For the first time, a company like Uber might choose to spend $1 million more on AI tokens instead of hiring five more engineers. The question is whether that $1 million in tokens yields the same output as those five engineers. Macdonald’s point is that the answer is not yet clear.”

Uber’s AI budget explosion was partly driven by the adoption of Claude Code, a coding assistant from Anthropic. Engineers at Uber use Claude Code to generate code snippets, write unit tests, debug, and even propose architectural changes. While the tool has accelerated certain tasks, Macdonald stressed that feature output — measured in shipped user-facing improvements — has not increased proportionally. This disconnect is especially concerning for a company that operates on thin margins in the ride-sharing and food delivery sectors.

Industry Context: Skepticism Grows

Uber is not alone in questioning AI ROI. In early 2026, a Deloitte survey of 2,000 CIOs found that only 28 percent reported a clear positive ROI from generative AI implementations. The remainder cited difficulties in quantifying benefits, integration challenges, and high costs. Meanwhile, AI spending across the Fortune 500 has surged by 60 percent year over year, reaching an estimated $480 billion in 2026. Analysts warn that a “AI spending bubble” could be forming, driven by hype and fear of missing out rather than measurable outcomes.

Macdonald’s interview comes at a time when Uber is also facing regulatory pressure, driver protests, and competition from Waymo and other autonomous vehicle operators. The company’s autonomous driving division, Uber ATG, was sold to Aurora in 2020, but Uber still partners with various AV companies. Some observers argue that Uber’s AI spending could be better directed toward foundational improvements in safety, efficiency, and driver satisfaction rather than experimental generative AI tools.

“The real AI opportunity for Uber might not be coding assistants,” says Raj Patel, an analyst at Morningstar. “It might be using AI to better predict demand, reduce wait times, or personalize promotions. Those are areas where the connection to user features is much clearer. But those kinds of applications require massive data pipelines and custom models, not just off-the-shelf LLMs.”

Macdonald’s Career and Perspective

Andrew Macdonald joined Uber in 2014 and rose through the ranks to become president in 2023. Previously, he led Uber’s Asia-Pacific operations and later oversaw global mobility. His background is in operations and growth, not AI, which may inform his focus on tangible results. Under his leadership, Uber achieved its first full-year profit in 2024, driven by cost controls and scaling of Uber Eats. Macdonald is known for a pragmatic approach, often pushing back against what he calls “vanity metrics.”

In the same interview, Macdonald also touched on Uber’s broader strategy. He emphasized that AI should serve the goal of improving user experience, not just internal efficiency. “The north star has to be the rider and the driver. If they don’t feel a difference, we’re wasting money,” he said. He also hinted that Uber might scale back some experimental AI projects if they don’t meet clear milestones within the next two quarters.

The stakes are high. Uber’s stock has been volatile, trading at around $68 in May 2026, down from a 2025 high of $92. Concerns about AI spending have been a contributing factor. Investors will be watching Uber’s next earnings call closely for signs of discipline. Macdonald’s message is clear: the era of unchecked AI spending may be coming to an end, at least at Uber.

As the interview concluded, Macdonald returned to the core challenge: linking rising token consumption to shipping valuable features. He called for better tracking and accountability. “We need to be able to say, ‘We spent X on tokens, and that directly led to rolling out this new feature that increased customer satisfaction by Y%. If we can’t do that, then we shouldn’t be spending that money.’ It’s that simple.”

Source: The Verge News