How Uber’s AV Strategy Could Fail
Robotaxis, “self-driving as a service,” and the fragility of Uber’s AV partners
I recently wrote about how Uber is in a better position than most people think when it comes to autonomous vehicles (AVs) for two reasons. The first reason is that variable demand for ridesharing and high existing costs for AV hardware mean it will take a while for an all-AV fleet (like Waymo’s) to become price-competitive with Uber’s human driver fleet. This creates short-term defensibility, but eventually costs will come down.
The long-term defensibility comes from the second reason: Most AV companies are not going to build their own robotaxi networks like Waymo or Tesla and deal with all the challenges of running a marketplace. Instead, they will join Uber’s and Uber can become the aggregator not just for fragmented human driver supply but also for fragmented AV supply.
And on paper, Uber looks like it’s in a great position to do this. It’s signed a flurry of partnerships in recent weeks to deploy robotaxis from Volkswagen (2026 in LA, with safety drivers), May Mobility (2025 in Texas, with safety drivers) and Momenta (2026 in Europe, with safety drivers), and expand its partnership with WeRide (15 additional cities in the next five years). That adds to existing partnerships with Avride, Motional, and Waymo, which appears in the Uber app in two cities today.

But as I’ve dug in more, I’ve realized that the long list of Uber partners is masking a big risk: it’s not clear that many of them will survive! Uber has two key types of partners, and both face big challenges on the capital and technology fronts:
Traditional robotaxi companies, like May Mobility or Motional, have enormous capital requirements and still face large technology gaps to market leaders like Waymo (note: even though Waymo is an Uber partner today, for this piece I assume they won’t be one long-term given their ambitions).
"Self-driving as a service" companies like Mobileye and Wayve AI offer a more promising alternative path, but they too are behind the market leaders technologically.
Below, I’ll dig into these challenges, discuss why I think most Uber partners will fail, and also show why I’m still cautiously optimistic on Uber despite these risks.
The robotaxi business model is in question
The biggest problem for Uber is that most of its partners are traditional robotaxi1 companies that require enormous capital to survive, given huge costs and massively delayed revenue. Remember these companies’ predicament: they need to build out the software/hardware for L4 autonomous driving and manage fleets of vehicles. They only make money once they hit superhuman levels of safety and get regulatory approval to deploy their cars on the road. And if they make a mistake, regulators can (and will!) shut down their business instantly.
It’s no wonder a lot of companies have thrown in the towel. Cruise was shut down by GM in late 2023. Argo AI was shut down by Ford/Volkswagen in 2022. Uber and Lyft disbanded their units a few years ago, and Apple’s Project Titan was killed last year.
Almost all of the ones that have survived have done so by finding a parent company willing to fund the enormous capital requirements. But it’s easy to imagine many of them pulling the plug like GM did with Cruise recently:
Uber partner Motional (owned by Hyundai) has delayed its robotaxi launch to at least 2026 and saw its CEO recently step down.
Uber partner Avride’s parent company is hemorrhaging money, and Avride is far from rideshare commercialization.
Zoox, owned by Amazon and not a current Uber partner, is supposed to launch this year. But it’s had a series of delays over the last few years and you can imagine Amazon pulling the plug given it’s not core to their business.
None of Uber’s partners have the backing that Waymo does from Alphabet. Waymo has likely burned $15-25 billion in its history and is still burning ~$2 billion annually with no end in sight.2 “Capital moats” have fallen out of style in recent years, but this might be one market where they still matter.
Waymo’s widening tech gap
If Uber’s robotaxi partners had the best self-driving technology, they could probably find the right backers to fund their expansion. But they are also behind on their technology, making it harder for them to raise the money to catch up.
It’s hard to precisely measure these technology gaps, but we can proxy this by looking at who is furthest along in safely deploying commercial L4 AVs.3 There are only four companies today that have done this: Waymo and three Chinese companies (Baidu Apollo Go, WeRide, and Pony.AI).4
Waymo and Baidu have significant scale. Baidu was on track for 4.4m annualized in Q4 2024, and Waymo just announced 250k weekly rides (13m annualized). But both are building their own marketplaces and are unlikely to be Uber partners long term as mentioned (Baidu has never been one).
WeRide and Pony.AI are both Uber partners and are each doing a small number of rides in China (unclear how many). These are more promising partners because they’re unlikely to build their own networks globally, but they have other long-term questions. They are Chinese companies, so they are banned from being used as robotaxis in the US under a Biden administration rule (hence why their Uber partnerships are focused on Europe). They also don’t have large parent companies to support them, but are public today and valued at modest levels ($7b for Pony.AI, $2.75b for WeRide.5
All of Uber’s other partners today have yet to make it to L4 commercial operations and are seemingly years behind the market leaders. Given these capital and technology hurdles, there are a lot of challenges ahead for Uber's robotaxi partners.
“Self-Driving as a Service” to the rescue?
We’ve been talking exclusively about “robotaxi” companies, but there’s another set of Uber partners I’m more bullish on: “self-driving as a service” companies. These are companies that are only selling self-driving software/hardware to automakers and not handling any of the operations or fleet management. These companies will allow any automaker to follow Tesla’s playbook of selling self-driving cars for both personal and robotaxi use, but would not require them to build everything in-house like Tesla.6
Uber partners like Mobileye and Wayve AI fit in this category. Mobileye already sells Advanced Driver-Assistance Systems (L1) to most automakers (70% global share) and is now trying to sell similar L2, L3, and L4 software and hardware to them. Wayve AI is doing the same, which is why Uber partnered with them and invested in their Series C round. The press release says Uber “envisions future Wayve-powered self-driving vehicles being made available on the Uber network in multiple markets around the world.” So Uber is hoping that any self-driving car with Mobileye or Wayve AI technology would be able to automatically plug into Uber’s AV network (regardless of the automaker).
I’m more bullish on these companies because they require less capital to succeed, and have a clearer path to monetization by selling L2, L3, and L4 technology incrementally and by tapping into the even larger personal vehicle market for sales.
But they still face similar challenges. They are well behind Waymo and Tesla technologically, as Mobileye has yet to ship any L4 systems and Wayve AI is shipping L2 systems only in 2027.
They also face competition from the market leaders. Waymo announced a partnership with Toyota that points to a future where individuals buy “Waymo Driver” vehicles for personal use and to put on the Waymo robotaxi network. Tesla has also discussed selling its FSD to other automakers, but it seems more like an idea than a reality.
So on both fronts (traditional robotaxi companies and self-driving as a service companies), Uber is at risk of seeing its fragmented supply fall away if Waymo and/or Tesla can take over the markets.
The commoditization question
Despite the risks, I’m still cautiously optimistic about Uber’s ability to succeed. It comes back to a more fundamental question about whether self-driving technology will be commoditized.7
It’s hard to imagine today, given Waymo has a growing tech advantage and has a data flywheel from real-world miles that lets it (in theory at least) improve faster than competitors.
But in the long run, I’m guessing the tech gaps close for a few reasons.
The cost curves on cameras and sensors look promising, and open source projects like Comma.AI show how much capability can be squeezed out of low-cost components.
LLMs have shown us that generalized AI models can often outperform previous specialized ones. ChatGPT is better at translation, geolocation, and text summarization than the previous specialized models for these use cases. You can imagine a future in which GPT 7.5 can just drive your car.8
There are diminishing returns to safety. As long as Uber’s partners are substantially safer than humans, I don’t think it’ll matter that much if they are 98% safer and Waymo is 99.5%.
If self-driving tech does become more commoditized, you’d expect the capital requirements to fall, the technology gaps to shrink, and for Uber to be in a good position as the value accrues to the network (and not the tech leader). Uber only needs a few companies to last to ensure supply remains fragmented enough.
But my theory will be tested soon. We’ll need to see some real progress from Uber’s partners over the next year or two. If these partners struggle to catch up, then there’s a risk that Uber won’t have any supply left to aggregate.
By traditional robotaxi company, I mean they are developing the AV stack, owning the vehicles, and handling operations themselves.
“Other Bets” has operating losses of ~$43b through Q1 2025 and ~1bn last quarter. Analysts have estimated that ~50% of that in recent years was Waymo.
In theory, this could be a bad proxy if the main barrier to deploying L4 AVs is regulatory or operational, not technological. But based on everything I read, it seems like the biggest factor holding back AV deployment for most companies is getting the technology to be good enough.
Tesla has a good chance of winning the self-driving race, but they have yet to deploy L4 commercial operations like these companies, and so I won’t be discussing them in this section.
Both are up massively in the last week, seemingly on the announcement of the Uber partnerships.
In my last piece, I was more bearish on the prospects of Tesla building a robotaxi network through customers buying their cars and putting them on the network. I have since updated my views and think their model can work, even if most of the people doing this will be managing fleets of Tesla vehicles they purchased (as opposed to sending their personal cars).
Similar to the debate on whether LLM models will be commoditized, and value will instead accrue to the application layer.
You’d expect any reasonable definition of AGI to mean it could do something like this.
Well written