Uber Spent Millions on AI and Now Doesn’t Know What It Got

Uber Spent Millions on AI and Now Doesn’t Know What It Got

Ondřej Barták
Ondřej Barták
Entrepreneur and Programmer
27. 5. 2026
4 minutes reading · 51 views
Uber Spent Millions on AI and Now Doesn’t Know What It Got

    A company that has transported tens of millions of people around the world has run into an unexpected problem. It is not regulators, drivers, or competitors. It is artificial intelligence—or, more precisely, the question of whether the enormous sums invested in it are producing anything tangible at all. Andrew Macdonald, Uber's chief operating officer, said so openly in an interview for the Rapid Response podcast released in late May. According to him, the cost of artificial intelligence is becoming increasingly difficult to justify. And that is not something we often hear from the head of one of the world's best-known technology companies.

    It is May, and Uber has already exhausted its annual AI budget

    Back in April, Uber Chief Technology Officer Praveen Neppalli Naga gave an interview to The Information. By then, Uber had used up its entire Claude Code budget allocated for 2026. And it was only March or April.

    Internally, this triggered a wave of discussions about how many tokens the company was actually consuming, what it was getting in return, and what trade-offs this entailed. Including staffing trade-offs. Shortly afterward, CEO Dara Khosrowshahi announced during an earnings call that Uber was slowing its hiring of new employees to cover the costs of artificial intelligence. But then came the uncomfortable question: what are we actually getting for that money?

    Macdonald spoke with the company's most experienced engineers. And what he found did not please him. Higher token consumption is not translating into a better product for users. "The connection just isn't there yet," he said. "I think perhaps implicitly more is being delivered, but it's very difficult to draw a line between one of those statistics and the idea that we're now actually producing 25% more useful features for customers."

    In other words: yes, 25% of code commits in the last quarter came from artificial intelligence. But how many projects that had previously been shelved were actually completed as a result? No one can say, even though the figures themselves are described as fantastic. Companies report that AI writes dozens of percent of their code, that token consumption is skyrocketing, and that employees are using AI more than ever before. But Macdonald points out that headline statistics and actual value for customers are two different things. And the gap between them is still enormous.

    Tokenmaxxing: a new corporate sport that may not make sense

    The term "tokenmaxxing" has recently caught on in the technology world. It refers to an effort to maximize the use of artificial intelligence at any cost, and some employers even evaluate their employees based on how much they use AI. Meta, Google, and JPMorgan are among the companies that have incorporated this metric into performance reviews.

    But Uber is beginning to wonder whether it is approaching the whole thing incorrectly. AI may seem free if you are a user experimenting with interesting things and not paying the bills. But someone is paying those bills. And at Uber, it was the company itself, without a clear idea of exactly what it was getting for its money.

    And it is not just Uber. Duolingo backed away from its decision to include the use of artificial intelligence in employee performance reviews. Management encountered resistance: people asked whether they were supposed to use AI because it made sense or simply to comply with a directive.  "It seemed that instead of holding people accountable for actual outcomes, we were pushing something that simply wasn't a good fit in some cases," said CEO Luis von Ahn.

    Uber and autonomous cars

    While Macdonald is putting the brakes on internal AI tools, he speaks very differently about one area. According to him, autonomous vehicles are not a question of if, but when.

    Uber currently has 25 partnerships with companies developing autonomous vehicles around the world and is actively helping them build the physical infrastructure needed to scale, from charging stations and remote vehicle management to customer support. Macdonald says the technological problems have largely been solved. What remains is the business model and mastering the process of scaling from five vehicles to five hundred and then to five thousand.

    "For Uber, autonomy is existential," he said bluntly. "It will be somewhere between an existential necessity and an enormous unlocking of value. It won't happen in a few years, but it won't take a few decades either." According to him, his own children, who are still young today, will probably never get a driver's license.

    Sources: aol.com, finance.yahoo.com and businessinsider.com

    Advertisement

    Content created with help from UpTier.

    SEO and GEO on autopilot. UpTier’s multi-agent systems write and optimize content for search engines and AI answers.

    Discover UpTier ↗

    Category:AI
    Did you enjoy this article?
    Discover more interesting posts on our blog
    Back to blog

    Related posts

    Google makes Tasks free for Gemini 4 Hydrogen usersGoogle makes Tasks free for Gemini 4 Hydrogen users
    Tasks is becoming available for free to hundreds of millions of Gemini 4 Hydrogen users. Google says it can call businesses, book appointments and complete purchases on their behalf.
    1 min read
    18. 11. 2026
    OpenAI introduces GPT-6.2 Astra and Sol for professional and everyday workOpenAI introduces GPT-6.2 Astra and Sol for professional and everyday work
    OpenAI has introduced two models: Astra for complex professional tasks and Sol for everyday work. Both are available in ChatGPT Work, Codex and through the API.
    3 min read
    17. 11. 2026
    Tencent releases Hy4 Flash with image support and a context window of up to 1 million tokensTencent releases Hy4 Flash with image support and a context window of up to 1 million tokens
    The lighter Hy4 Flash model targets chat, documents and routine code edits. It accepts text and images, supports tool use and, according to Tencent, matched Claude Haiku 5.5 in tests of everyday tasks.
    1 min read
    16. 11. 2026
    Přihlaste se k odběru našeho newsletteru
    Zůstaňte informováni o nejnovějších příspěvcích, exkluzivních nabídkách, a aktualizacích.
    CodedTrip

    Operated by CodedTrip LLC, USA.

    YouTube
    TikTok