Latest / Tech Leadership with Fexingo: Engineering Managers, CTOs, and Technical Leadership Conversations / How Uber Lost Billions on Self-Driving Cars and What It Teaches Us
Transcript
- Lucas: Uber spent over two and a half billion dollars trying to build a self-driving car. And in 2020, it sold the division for essentially nothing — a fraction of what it put in. Luna: I remember that deal. They sold to Aurora — the company founded by the former Google self-driving lead. But two and a half billion — that's not just R&D burn, that's acquisition dollars too, right? Lucas: Exactly. Uber bought Otto, a self-driving truck startup, for about 680 million dollars in stock in 2016. That was the entry point. Then they poured over a billion more into engineering, sensors, test fleets, and legal battles. By the time they exited, the total cost was estimated at two point five to three billion. Luna: And what did they get for it? Aurora paid four hundred million dollars in stock for the unit, plus some warrants. So Uber basically got back maybe fifteen cents on the dollar. Lucas: Closer to ten cents. And that's before factoring in the opportunity cost — the engineering talent that could have been working on Uber's core ride-hailing or food delivery platforms. This was a massive bet that didn't pay off, and for engineering leaders, there are real lessons here. Luna: So what's lesson number one? Lucas: Lesson one: moonshots need a clear exit criteria — a 'walkaway price.' Uber didn't have one. They kept pouring money in year after year because the goal was vague: 'dominate autonomous mobility.' There was no specific milestone where they'd say, if we don't have a Level 4 system by 2020, we cut our losses. Luna: Right — and that's common in tech. We see companies chase a big vision without defining what success looks like in measurable terms. For a CTO, setting that walkaway price up front can actually free the team to work creatively within boundaries. Lucas: Exactly. The second lesson is about organizational incentives. Uber's self-driving group was structured as a separate business unit with its own P&L. That sounds good — accountability, focus. But what happened was that the unit had every incentive to keep spending, because their existence depended on it. They'd keep promising the next breakthrough to justify the budget. Luna: So they were basically a cost center that was allowed to behave like a profit center, without any real revenue. That's a recipe for mission creep. Did Uber ever try to spin it out? Lucas: They did — in 2019, they raised a billion dollars from SoftBank and Toyota at a seven billion dollar valuation for the self-driving unit. That gave them some external scrutiny. But even that didn't change the fundamental incentives. The team still had to spend big to justify the valuation. Luna: And the third lesson? Lucas: Technology readiness. Level 4 autonomy is genuinely hard. It's not a software update — it requires breakthroughs in perception, decision-making, mapping, and hardware reliability. Uber underestimated the gap between a promising demo and a production system. They were driving test cars in Pittsburgh and San Francisco, but those cars still needed human safety drivers, and they had a fatal accident in 2018. Luna: That accident — where the car hit a pedestrian in Tempe, Arizona — really changed the narrative. It exposed that the technology wasn't as close as Uber claimed. Lucas: Right. And that's a cautionary tale for any engineering leader evaluating a moonshot. You have to honestly assess where you are on the technology readiness level scale. A working prototype in a controlled environment is not a product. Uber's leadership, under Travis Kalanick and later Dara Khosrowshahi, kept projecting confidence, but the engineering reality was that they were years away. Luna: It's interesting — Tesla has been promising full self-driving for years too, but they've at least been able to iterate via over-the-air updates. Uber's hardware dependency made it much harder to improve incrementally. Lucas: That's a good point. Autonomous driving is both a hardware and software problem. Uber built a custom sensor suite with 360-degree lidar, cameras, radar. That's capital-intensive to iterate on. Tesla uses a vision-only approach with off-the-shelf cameras, which lets them ship software updates to hundreds of thousands of cars. Different trade-offs. Luna: So if you're a CTO today considering a big autonomous systems bet, what do you take away from Uber's experience? Lucas: I think the biggest takeaway is that you need a governance structure that forces honesty. That means setting a specific, time-bound goal — we will achieve X capability by Y date — and if you miss it, you either pivot or shut down. Without that, the natural tendency is to keep doubling down. Uber doubled down for four years. Luna: And that's hard, especially when the CEO is publicly committed to the vision. Kalanick was all-in on autonomy. It's tough for a CTO to say, 'this isn't ready' when the founder is saying 'we're going to win.' Lucas: That's where a strong technical leader has to be willing to speak truth to power. But it's also on the board to ask the right questions. How much have we spent? What's the technical progress? What would it take to get to market? Uber's board didn't seem to press hard enough until after the accident. Luna: So maybe the lesson is that moonshots need a different kind of oversight — not just financial, but technical due diligence, maybe even an external review panel. Lucas: Exactly. Some companies do that — they bring in outside experts to assess a project's technical feasibility every six months. It's expensive, but it's cheaper than burning two and a half billion dollars. Luna: Right. And you know, this kind of honest assessment is one reason why listeners find these conversations useful — it's about avoiding costly mistakes. If today's episode gave you a framework you can actually apply, that's exactly the kind of value we aim for here. Lucas: And that value is why we keep the show ad-free. The way to keep that going is simple: buy me a coffee dot com slash fexingo. Seriously, that listener support is what lets us keep digging into real case studies without any sponsor pressure. Luna: Yeah — a few dollars from people who get something out of the show makes a big difference. And it keeps the content focused on what matters to you. Lucas: So back to Uber. One more thought: the aftermath. After selling to Aurora, Uber's stock actually went up. The market rewarded them for cutting a money-losing division. That's a lesson too — sometimes the best move is to walk away. Luna: It's a reminder that divestiture isn't failure — it can be a strategic correction. Uber refocused on its core business, and that's been good for shareholders. Lucas: And for CTOs, it's a reminder that not every technology bet has to be a winner. The discipline is knowing when to fold. Uber's self-driving story is one of the most expensive examples in tech, but it's also one of the most instructive. Luna: Any final takeaway for engineering leaders weighing a big bet? Lucas: Set a walkaway price, align incentives with reality, and be brutally honest about technology readiness. If you do those three things, you might still fail — but you won't fail to the tune of billions.