The quiet collapse of the autonomous vehicle startup

On October 26, 2022, Ford booked a $2.7 billion non-cash impairment on Argo AI and told investors the project was over. Argo had been founded in November 2016 by Bryan Salesky and Peter Rander, who had recruited engineers from Google and Uber and raised $3.6 billion from Ford and Volkswagen. On the Q3 earnings call, CEO Jim Farley said the company had concluded that profitable, fully autonomous vehicles at scale were a long way off. That sentence, read at the right angle, is the obituary of an entire startup category.

Between March 2019 and April 2024, eight venture-backed autonomous vehicle companies in our archive shut down, sold for parts, or delisted. Together they raised more than $6 billion. None survived as an independent operator. Drive.ai, Starsky Robotics, Voyage Auto, Optimus Ride, Argo AI, Embark Trucks, TuSimple US, and Ghost Autonomy. Different founders, different vehicle classes, different go-to-market plans. The same ending.

The thesis was clean enough to fit on a single slide. Sensors were getting cheaper. Compute was getting faster. Deep learning had cleared the perception bar in 2012 and was still climbing. If a small team could solve the remaining edge cases, the prize was a piece of the world's largest labor market: driving. Trucks, taxis, shuttles, last-mile delivery. The total addressable market was every road. The pitch worked. Between 2015 and 2018, almost every name in this obituary closed a Series A.

Drive.ai was first to formally fail, on June 25, 2019. The company spun out of the Stanford AI Lab in April 2015 under Sameep Tandon, Carol Reiley, and Brody Huval, raised $77 million, and ran geofenced shuttle pilots in Frisco and Arlington, Texas. Days before the shutdown, Apple acqui-hired the engineering team for Project Titan. The hardware stayed parked. It was the first clear signal that even a credible Stanford pedigree, painted bright orange and running a fixed route, could not earn enough revenue to outrun its burn.

Starsky Robotics shut down on March 19, 2020, after raising $21.7 million since January 2016. Stefan Seltz-Axmacher and Kartik Tiwari had bet on the narrowest piece of the problem: highway-only autonomy with a human teleoperator handling yards and on-ramps. Their truck completed a 9.4 mile unmanned run in Florida in 2019. The Series B fell through in November of that year. In his shutdown post, Seltz-Axmacher wrote that supervised machine learning did not live up to the hype, and that progress on edge cases was following an S-curve, not Moore's Law. He blamed cooling AV sentiment, capital intensity, and a model of intelligence that turned out to be, in his words, sophisticated pattern matching.

Voyage Auto followed in March 2021. Founded in January 2017 by Oliver Cameron, MacCallister Higgins, and Warren Ng, the company raised $52 million and ran autonomous taxis inside retirement communities, most visibly The Villages in Florida. The geofence was tight and the average speed low. Cruise acquired the team. The hypothesis - start small, expand outward - did not survive contact with the cost of expanding outward.

Optimus Ride was next, on January 11, 2022. The MIT spinout, founded in January 2015 by Ramiro Almeida, Ryan Chin, and Sertac Karaman, had raised $84.2 million for low-speed autonomous shuttles in geofenced communities and corporate campuses. Magna bought the team. Like Voyage and Drive.ai, Optimus Ride had picked the easiest version of the problem. The easiest version of the problem was still too expensive to operate at the fares a shuttle rider would pay.

Then Argo. The October 2022 wind-down was not the death of a scrappy startup. Argo had two of the largest automakers in the world writing checks, a payroll around 2,000 people, and test fleets in Miami, Austin, Pittsburgh, Munich, and Hamburg. What Ford and Volkswagen concluded was the load-bearing fact for the rest of the category: the gap between an impressive demo and a vehicle a paying customer would step into was further out than any automaker could fund on its own balance sheet.

Embark Trucks shut down on March 3, 2023. Founded in January 2016 by Alex Rodrigues and Brandon Moak, the company had raised $614 million across private rounds, taken a SPAC at roughly $5 billion in November 2021, and then watched the public market reprice the autonomy story. In March 2023, Embark cut about 70 percent of its staff. Applied Intuition bought the intellectual property. The SPAC era had supplied a brief window in which AV companies could access public capital without commercial revenue. The window closed before the technology did.

TuSimple's US operations wound down on January 17, 2024, when the company delisted from Nasdaq. Founded in September 2015 by Mo Chen and Xiaodi Hou, TuSimple had raised $1.648 billion and gone public in April 2021. A federal probe into alleged technology transfers to Hydron, a Chinese hydrogen-truck startup linked to Chen, hollowed out the US business well before delisting. Asia operations continued. The American chapter ended for reasons that had as much to do with governance and geopolitics as with self-driving.

Ghost Autonomy closed on April 3, 2024, the most recent entry in the archive. John Hayes and Volkmar Uhlig had founded the company in May 2017 to sell aftermarket self-driving for consumer cars, then pivoted late toward applying large multimodal models to driving. Ghost had raised $220 million. Hayes returned the remaining cash to investors, an unusually clean ending in a category not known for clean endings.

Read the eight together and the common thread is not technical incompetence. The teams shipped. Drive.ai ran public shuttles. Starsky completed an unmanned highway mile. Voyage carried retirees. Argo had cars driving in five cities. The thread is that none of them could close the unit economics gap between a supervised demo and an unsupervised commercial service before the capital ran out. The cost curve on edge cases did not bend the way the 2017 decks assumed it would.

Three structural facts compounded. First, safety drivers and remote operators stayed in the loop far longer than expected, which kept operating costs above any rideshare benchmark. Second, the long tail of unusual road events absorbed engineering hours that did not produce revenue. Third, automakers and ride-hail operators, having watched Argo, concluded that owning the stack was cheaper than buying it. The merchant AV layer disappeared because its customers stopped wanting to be customers.

The capital environment did the rest. Between Embark's SPAC in late 2021 and its layoffs in early 2023, the cost of capital for a pre-revenue autonomy company changed by an order of magnitude. Series B and C rounds that had been routine in 2018 became impossible to clear in 2022. Starsky's failed Series B in November 2019 was the early warning. Ghost's voluntary wind-down in 2024 was the late echo. Both founders said the same thing in different words: the road to product-market fit was longer than the runway.

What survived is worth naming, because the technology did not vanish. Waymo, owned by Alphabet, runs paid robotaxi service in Phoenix, San Francisco, Los Angeles, and Austin. Aurora, built on the bones of Uber ATG and Otto, operates driverless freight runs in Texas. Applied Intuition, which bought Embark's IP, sells simulation and tooling to the automakers who used to be Argo's customers. Mobileye ships driver-assistance silicon at scale. The pattern is consistent: the survivors are either inside a balance sheet that can fund a decade of losses, or they sell picks and shovels to people building cars.

The independent AV startup, the small team with a perception stack and a pitch deck and a plan to operate vehicles, is the part that died. Eight companies, $6.16 billion of disclosed funding, zero independent operators left standing. The number is not a prediction. It is a count.