Uber
Last reviewed
May 2, 2026
Sources
20 citations
Review status
Source-backed
Revision
v2 · 4,633 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
May 2, 2026
Sources
20 citations
Review status
Source-backed
Revision
v2 · 4,633 words
Add missing citations, update stale details, or suggest a clearer explanation.
Uber Technologies, Inc. is an American multinational company that runs ride-hailing, food delivery, and freight platforms through a single mobile app. It is headquartered in San Francisco, California, and is incorporated in Delaware. The company began as UberCab in March 2009, founded by Travis Kalanick and Garrett Camp, and now operates in roughly seventy countries with more than ten thousand cities on its platform. Uber went public on the New York Stock Exchange on May 10, 2019 under the ticker UBER.
For an AI wiki the interesting thing about Uber is that it is, beneath the surface, an enormous machine learning company. The matchmaking that pairs a driver with a rider, the ETA shown when the car is two minutes out, the surge multiplier that bumps your fare on a Friday night, the fraud check that quietly cancels a stolen-card payment, and the recommendation banner at the top of Uber Eats are all running through internal models that engineers at Uber have been training and retraining for more than a decade. The company also produced one of the more peculiar AI research labs of the late 2010s, Uber AI Labs, and in the same period operated one of the more troubled autonomous vehicle programs in the industry, Uber ATG, which it eventually sold to Aurora in late 2020 after a fatal pedestrian crash in Arizona.
The origin story has been told often enough to feel like a parable. On a snowy December evening in Paris in 2008, Garrett Camp, the cofounder of StumbleUpon, could not find a cab. He started sketching ideas for a button you could press on a phone to summon a black car. He kept refining the concept after he came back to San Francisco. On New Year's Eve 2008, Camp and his friend Travis Kalanick spent around eight hundred dollars on a chartered driver across the city, which Camp later said convinced him that the cost could be cut by an order of magnitude if you let one driver pick up successive passengers via app. Kalanick was a serial entrepreneur who had sold his last company, Red Swoosh, to Akamai for around nineteen million dollars in 2007.
UberCab was incorporated in March 2009. Camp and two early collaborators, Oscar Salazar and Conrad Whelan, built the prototype. Kalanick joined as what he called a "chief incubator" and angel investor. Ryan Graves answered a tweet from Kalanick about a startup hire and became the first employee in February 2010, and was briefly the company's first CEO before Kalanick took the role in late 2010. The first ride happened in San Francisco in July 2010, after a private beta in May. By October the city's taxi regulator had sent a cease-and-desist letter complaining about the word "cab" in the name, and the company quietly dropped it. From then on it was just Uber.
Funding came fast. First Round Capital led a seed round of around 1.25 million dollars in 2010. Benchmark led an eleven-million-dollar Series A in February 2011. Menlo Ventures led a thirty-seven-million-dollar Series B in late 2011 with TPG and Goldman Sachs participating. Google Ventures led a 258-million-dollar Series C in August 2013, with David Drummond, then a senior Google executive, taking a board seat. By 2015 Uber had raised more than a billion in a single round, with private valuations that climbed past sixty billion dollars before any of it was tested by the public market.
The original product was simply called Uber and was built around black town cars driven by licensed limousine drivers. In July 2012 the company introduced UberX, a low-cost option using ordinary drivers in their own cars. UberX was the first time the company moved decisively into the gray area between licensed transport and what regulators kept calling "unlawful taxi service." The legal fights that followed defined Uber's relationship with city governments for years.
The core product line later expanded into a long list of variations. UberPOOL (2014, later renamed UberX Share), Uber Comfort, Uber Black, Uber Premier, Uber XL, Uber Pet, Uber Reserve, Uber Connect, Uber Teen, and Uber Health are all built on the same dispatch and pricing engines. The matching algorithm assigns the closest available driver to each request, with adjustments for vehicle category, driver rating, predicted route, and pickup viability. The pricing engine layers a base fare with per-mile and per-minute rates and applies a surge multiplier when local demand outruns supply.
Uber Eats started as a small experiment called UberFRESH in Santa Monica in August 2014, delivering a curated lunch menu. It rebranded as Uber Eats in 2015 and launched as a separate app in Toronto in late 2015. By the early 2020s Uber Eats had grown into the company's second business line and was, in many quarters, the single fastest-growing part of the company. After the December 2020 acquisition of Postmates for 2.65 billion dollars in stock, Uber Eats became the largest food delivery service in the United States by some measures, competing with DoorDash, Grubhub, and Instacart. The company also bought the alcohol delivery service Drizly in February 2021 for 1.1 billion dollars, although it shut Drizly down in early 2024 after deciding to fold the alcohol business directly into the Uber Eats app.
Uber Freight launched in May 2017, initially in Texas, as a kind of trucking equivalent of UberX. Shippers list loads, carriers and owner-operators bid through the app, and Uber Freight handles the brokerage in between. The unit was led by Lior Ron, who had been one of the cofounders of Otto, the autonomous trucking startup Uber acquired in 2016. In November 2021 Uber Freight bought Transplace, a managed transportation provider, for around 2.25 billion dollars in cash and stock, vaulting the unit into the top tier of US freight brokers.
The most aggressive period of Uber's international growth ran from 2013 through about 2017. The company entered hundreds of cities, fought running regulatory and legal battles in many of them, and burned through billions of dollars in subsidies trying to outlast local incumbents. The strategy worked in some markets and failed loudly in others.
This pattern (enter a market, burn cash, then trade the local business for stock in the eventual winner) was eventually formalized by Khosrowshahi as a strategy. By 2020 the company had a portfolio of minority stakes in regional ride-hailing leaders that quietly accounted for billions of dollars on its balance sheet.
The Saudi Arabian Public Investment Fund put 3.5 billion dollars into Uber in June 2016 for roughly a five percent stake, taking a board seat in the process. It was at the time the largest single investment a foreign government had ever made into a venture-backed startup.
The biggest single private financing round arrived at the end of 2017. After months of board fights, SoftBank's Vision Fund led a roughly nine-billion-dollar deal that valued Uber at 48 billion dollars, a significant discount to its previous 68-billion-dollar valuation. Most of the money was a tender offer to existing shareholders rather than fresh capital. The deal closed in January 2018 and made SoftBank, run by Masayoshi Son, Uber's largest shareholder.
Uber filed its S-1 in April 2019 and listed on the New York Stock Exchange on May 10, 2019. The IPO priced 180 million shares at 45 dollars each, the bottom of the indicated range, raising about 8.1 billion dollars and giving the company a non-diluted valuation of roughly 75.5 billion dollars. The opening trade was 42 dollars and the stock closed its first day at 41.57, down 7.6 percent. The debut was widely called the worst large-cap IPO in years on a first-day basis. By June 2019 the stock had finally crossed back above 45.
This is the part that justifies Uber's place on an AI wiki, and it is much larger than most outside observers realize. The company's machine learning footprint runs across three rough categories: an internal ML platform that powers nearly every consumer-facing feature; a research lab that produced influential open-source tools and some genuinely novel academic work; and a long, expensive, ultimately abandoned bet on developing self-driving cars in-house.
In December 2016 Uber announced the formation of Uber AI Labs, built around the acquisition of Geometric Intelligence. The acquisition price was never disclosed officially, but reporting at the time put it in the tens of millions. Geometric Intelligence had been founded in 2014 by four researchers: Gary Marcus, then a professor of psychology and neural science at NYU; Zoubin Ghahramani, a Cambridge machine learning professor and Royal Society Fellow; Ken Stanley, a computer science professor at the University of Central Florida; and Douglas Bemis, a recent NYU PhD. Marcus was the founding director of Uber AI Labs, and the original team brought roughly fifteen researchers into Uber.
The lab's mandate was unusually broad for an industrial AI group. The team published in NeurIPS and ICML on subjects that had nothing obvious to do with ride-hailing, including neuroevolution, open-ended learning, probabilistic programming, and deep generative modeling. Marcus left in early 2017 after only a few months, and Ghahramani took over as Chief Scientist before himself moving on to Google in 2020. Jeff Clune, Ken Stanley, and Joel Lehman led much of the lab's later research output before the group was wound down through a series of layoffs in 2020.
A short list of the better-known research projects to come out of Uber AI:
The lab and the broader engineering organization shipped a string of open-source releases that genuinely shaped the open-source machine learning ecosystem. The two most consequential were Pyro and Horovod.
| Project | Released | Purpose | License | Status |
|---|---|---|---|---|
| Pyro | November 2017 | Deep probabilistic programming on PyTorch | Apache 2.0 | Donated to LF Deep Learning, 2019 |
| Horovod | October 2017 | Distributed training for TensorFlow, Keras, PyTorch, MXNet | Apache 2.0 | Donated to LF Deep Learning, December 2018 |
| Ludwig | February 2019 | Declarative deep learning toolbox, no-code training | Apache 2.0 | Donated to LF AI & Data, 2020; commercialized by Predibase |
| Petastorm | 2018 | Apache Parquet-based dataset library for distributed deep learning | Apache 2.0 | Maintained on GitHub |
| Manifold | 2019 | Model-agnostic ML diagnostics | Apache 2.0 | Maintained on GitHub |
| Pyro POET | 2019 | Open-ended environment generation | MIT | Reference implementation |
Pyro was the most influential of the lot. It built on PyTorch and provided variational inference and stochastic computation primitives in a way that let researchers express complex generative models without writing inference algorithms by hand. After Uber donated it to the Linux Foundation, much of the original Pyro team eventually moved to other organizations and the project's center of gravity shifted to academia.
Horovod solved a much more practical problem. Distributed training in TensorFlow was painful in 2017, requiring hand-written parameter servers. Horovod, named after a Russian folk dance in which dancers form a ring and step together, used MPI-style ring-allreduce communication to scale data-parallel training across many GPUs with a few lines of code. Real-world Uber workloads using Horovod included self-driving training, fraud detection, and trip forecasting. The project was widely adopted across the industry and became a baseline that later frameworks like PyTorch DDP and DeepSpeed had to beat.
Ludwig, released in 2019, took the opposite approach. Where Pyro and Horovod were aimed at researchers and platform engineers, Ludwig let domain experts train competitive deep learning models from a YAML file. The author, Piero Molino, eventually left Uber and cofounded Predibase in 2021 along with Travis Addair, who had led Horovod inside Uber. Predibase has continued to maintain Ludwig as the open-source foundation under its commercial product.
While Uber AI Labs was publishing papers, the larger ML organization at Uber was building the platform that actually served predictions to riders and drivers. Michelangelo is Uber's internal end-to-end ML platform, first introduced publicly in a 2017 engineering blog post by Jeremy Hermann and Mike Del Balso. The team had been building it since mid-2015. Michelangelo handles the full lifecycle: data management, training, evaluation, deployment, online serving, and monitoring. By 2017 it was already serving production traffic for fraud detection, ETA prediction, marketplace forecasting, and Eats recommendations.
Michelangelo's architecture has gone through several public iterations. The platform is built on a mix of Apache Spark, Apache Kafka, Cassandra, and a custom feature store called Palette. By the early 2020s Uber engineers were publishing follow-up posts about scaling Michelangelo to deep learning models served at low latency, and by 2024 Uber was talking publicly about extending Michelangelo into the generative AI era with support for foundation model fine-tuning and serving.
Del Balso left Uber in 2018 and cofounded Tecton, a feature store startup, with several other Michelangelo alumni. The Michelangelo lineage runs through a surprising number of ML platform companies in the late 2010s and early 2020s.
The interesting thing about Uber's product machine learning is the sheer breadth of where models show up.
The self-driving program is its own long, expensive story. Uber Advanced Technologies Group (Uber ATG) was formed in early 2015 in Pittsburgh after Uber struck a partnership with Carnegie Mellon University and then hired roughly fifty researchers and engineers out of CMU's National Robotics Engineering Center. The hiring spree gutted parts of CMU's robotics program and was treated, fairly or not, as the moment Uber went all-in on building the self-driving stack.
In August 2016 Uber acquired Otto, an autonomous trucking startup founded by Anthony Levandowski earlier that year, for roughly 680 million dollars in stock and incentive payments. Levandowski had recently left Google's self-driving project, which would soon spin out as Waymo. In February 2017 Waymo sued Uber, alleging that Levandowski had downloaded more than fourteen thousand confidential files from Google before leaving and that those files made it into Uber's LiDAR program. The lawsuit, which produced reams of damaging discovery for Uber, ended in February 2018 with a settlement: Uber gave Waymo equity worth about 245 million dollars and agreed not to use the disputed technology. Levandowski himself was fired by Uber in 2017 and was later sentenced to eighteen months in prison for trade secret theft, though he was pardoned by President Trump in January 2021.
Then, on the night of March 18, 2018, an Uber ATG test vehicle traveling at 39 miles per hour in autonomous mode hit and killed Elaine Herzberg, a 49-year-old woman pushing a bicycle across a four-lane road in Tempe, Arizona. It was the first known pedestrian fatality involving a self-driving car. The NTSB's final 2019 report was scathing. The system's perception stack detected Herzberg 5.6 seconds before impact but oscillated between classifying her as a pedestrian, a bicyclist, and an unknown object. A subsystem called "action suppression" suppressed automatic emergency braking for a full second to avoid false alarms during normal driving. The safety driver, Rafaela Vasquez, was watching a video on her phone in the moments before the crash. The NTSB's probable cause finding cited Vasquez's distraction but added that contributing factors included "the Uber Advanced Technologies Group's inadequate safety risk assessment procedures, ineffective oversight of vehicle operators, and lack of adequate mechanisms for addressing operators' automation complacency, all a consequence of its inadequate safety culture." Uber suspended testing in Arizona, and ATG never fully recovered.
After Khosrowshahi took over as CEO and rounds of cost cuts hit the unit, ATG was sold on December 7, 2020 to Aurora Innovation, a startup founded by Sterling Anderson, Drew Bagnell, and Chris Urmson. The deal was structured around an enterprise valuation of roughly four billion dollars in Aurora stock plus a 400-million-dollar equity investment from Uber into Aurora. Khosrowshahi joined Aurora's board. The combined company subsequently went public via SPAC and is now a publicly traded autonomy company focused on trucking.
After the ATG sale Uber pivoted to a partnership-only autonomy strategy. The company stopped trying to build the self-driving stack and instead positioned itself as the network on top of which other people's robots can run.
| Partner | Type | Geography (announced or live) | Status |
|---|---|---|---|
| Waymo | Robotaxi | Phoenix (Oct 2023), Austin (Mar 2025), Atlanta (Jun 2025) | Live |
| Aurora | Autonomous trucking | Texas | Pilot via Uber Freight |
| Wayve | Robotaxi | London | Trial planned for spring 2026 |
| Pony.ai | Robotaxi | Middle East and Asia | Pilot |
| WeRide | Robotaxi | Abu Dhabi, Dubai | Live |
| Baidu Apollo (Apollo Go) | Robotaxi | Asia and Middle East ex-China | Multi-year deal, July 2025 |
| May Mobility | Robotaxi/shuttle | Arlington, Texas | Live |
| Nuro | Autonomous delivery | US food delivery | Pilot |
| Avride | Robotaxi and delivery | Austin, Dallas | Live |
| Volkswagen ADMT | Robotaxi (ID. Buzz AD) | Los Angeles | Multi-year |
| Momenta | Robotaxi | International | Multi-year |
The arrangement that has gotten the most public attention is with Waymo. The Phoenix integration started in October 2023 with autonomous Jaguar I-PACEs being dispatched against UberX, Uber Comfort, and Uber Green requests inside the Waymo coverage area. The companies expanded the integration to Austin in March 2025 and to Atlanta in June 2025, splitting operational responsibilities such that Uber handles charging, cleaning, and fleet management while Waymo retains responsibility for the autonomous driving stack itself. As of mid-2025 Waymo said it was running roughly 250,000 paid robotaxi rides per week across its full footprint.
The shift from Kalanick to Khosrowshahi is one of the most studied management transitions in modern Silicon Valley. Kalanick was, in retrospect, ill-suited for an organization the size that Uber had become by 2017. A series of incidents in early 2017 (Susan Fowler's blog post about sexual harassment, the Greyball revelations, a video of Kalanick himself berating an Uber driver, the death of an executive who had been at the company only briefly, and the publication of an internal Holder report calling for sweeping changes) culminated in pressure from a group of major investors led by Benchmark. Kalanick took an indefinite leave of absence in June 2017 and resigned as CEO a week later under direct pressure from those investors.
The board picked Dara Khosrowshahi, then CEO of Expedia and one of three finalists, in late August 2017. He took the job without having been the obvious leak-to-the-press favorite, which was apparently part of why he won the vote. Khosrowshahi spent his first eighteen months on what amounted to a public reputation tour: settling lawsuits, hiring a new general counsel and CFO, restoring Uber's London license after it was revoked, and rolling out a new set of corporate values. Kalanick stayed on the board for several more years and resigned the seat in December 2019 after selling roughly 2.5 billion dollars of his Uber shares. Khosrowshahi was still CEO as of mid-2026.
The full list is long. The ones that mattered most to the company's later trajectory:
A partial overview of consolidated annual figures, in US dollars. The 2017 and 2018 numbers cover Uber's first two full years as a privately held but quasi-public reporter; the 2019 figures cover the year of the IPO; and 2024 covers the company's first calendar year of full-year GAAP profitability.
| Year | Revenue | Net income (loss) | Gross bookings | Trips |
|---|---|---|---|---|
| 2017 | $7.9B | ($4.0B) | $34B | 4B |
| 2018 | $11.3B | $1.0B (one-time gain) | $50B | 5.2B |
| 2019 | $13.0B | ($8.5B) | $65B | 6.9B |
| 2020 | $11.1B | ($6.8B) | $58B | 5.0B |
| 2021 | $17.5B | ($0.5B) | $90B | 6.3B |
| 2022 | $31.9B | ($9.1B) | $115B | 7.6B |
| 2023 | $37.3B | $1.9B | $138B | 9.4B |
| 2024 | $43.9B | $9.8B | $162B | 11.3B |
The 2019 net loss was inflated by IPO-related stock-based compensation. The 2022 loss was driven by mark-to-market write-downs on Uber's equity stakes in Didi, Grab, and Aurora. Beginning in 2023 Uber posted its first full-year operating profit, and 2024 was the first year of substantial GAAP net income, much of which came from revaluation gains on those same minority stakes.