Former Uber CEO Travis Kalanick once called automated vehicle (AV) adoption “existential” for the ridehailing company. Even after his ouster, the firm hasn’t changed course on this core position. The reason? An assumption that automated vehicles will produce substantially lower operations costs for TNCs.
But this assumption badly needs a reality check. AVs will not significantly reduce the cost of transportation for many years. Not for personal car ownership, fleet-based TNC services, or long-distance freight. And that’s because scaling network architecture is hard, especially when your system is not allowed to fail.
There are several factors I see frequently underestimated in tech media about connected automated vehicle (CAV) life-cycle costs:
- Operating costs will be substantially higher than currently assumed due to software development and security requirements.
- Vehicle system architecture will require years of refinement to reach anything resembling an efficient design, imposing higher capital costs and accelerating hardware obsolescence.
- Cities are quite different from one another, and the same scaling challenges impacting current TNC services will plague AV deployment.
Many industry leaders and analysts have been giving us bad back-the-napkin math in their AV “affordability” estimates underpinning aggressive adoption models. Even the more thoughtful costing analyses feature critical oversights that render their adoption models deeply flawed. UBS’s much-cited analysis ignores software and cybersecurity entirely. RethinkX discusses the former (dismissing software maintenance costs as “marginal”), but assumes linear five-year depreciation without accounting for growth in computing requirements.
Latency demands require AVs perform local computing (i.e. can’t just send all data to the cloud), and iterative software development will tax legacy hardware. Those “Gen 1” and “Gen 2” AVs won’t last long in a competitive environment, and may require hardware upgrades after just a year or two. Don’t expect an exponential decline in hardware costs if data collection and processing requirements escalate through early generations of the technology.
And no matter the capital cost, net profit will rely heavily on utilization rates. AV fleet management will prove more challenging than the legacy model, hence why prospective operators like Ford are already experimenting. Supply distribution is more challenging when your business model relies on every vehicle maximizing utilization. Compare with today’s services, where Uber and Lyft care little if they have too many drivers on the road. Driver recruitment and retention are minor costs compared with vehicle ownership, operations, and maintenance. If TNCs own the AVs on their platforms, they bear these costs themselves, and few US markets are equally suited as New York or San Francisco for fleet optimization.
Several firms have produced research warning of strong industry headwinds, including McKinsey on escalating software costs and automotive cybersecurity, and the R Street Institute on broader cybersecurity threats to connected vehicle infrastructure. On the hardware side, the CEO of LiDAR developer Luminar argued the first generation of AVs could cost several hundred thousand dollars due to reliability demands.
Moreover, we’re finally starting to see research about the energy requirements from the standard array of sensing and computing devices on CAVs, and the near-term implications are daunting. A recent study from the University of Michigan and Ford explored the energy impacts of alternative CAV architectures, and the results highlighted potentially dramatic shifts in energy requirements based on vehicle design alone. In mixed traffic with low CAV market penetration, potential energy efficiencies derived from coordinated vehicle movement (e.g. platooning) are entirely unrealistic.
The sum of these factors suggests AV technology is still a long way off from providing anything resembling cost parity with human-operated vehicles, let alone real cost savings. TNCs that can’t generate a profit today won’t flip a switch and see automation rescue failing business models.
That said, there has been no conclusive determination that TNCs will actually own the vehicles in their networks. Uber announced a potential deal in November where it would purchase up to 24,000 fully automated Volvo XC90 SUVs between 2019 and 2021, though there’s no commitment in the framework. But if it rejects that model, instead remaining an open platform for manufacturers/developers, it will cede to them a lion’s share of the mobility value chain.
I don’t doubt we’ll eventually see AV fleets reach profitability, but not for quite some time. Certainly not within the next five years, and likely closer to ten. In the meantime, how much patience will shareholders have? After watching Uber hemorrhage hundreds of millions of dollars quarterly for years on end, will they accept continued losses from TNCs in the transition towards automation? Or will increased pressure for fiscal restraint slow development and growth?
Better battery technology, efficiently designed vehicle and network architecture, and refined operations at the city and regional levels will ultimately shape the long-term viability of these business models. But today, market analysts and investors need to exercise greater caution when assessing the near-term prospects of industry players.
5 thoughts on “A Reality Check on AV Economics”
Spectacular analysis. An almost entirely different “take” on AVs. Very well done, thanks for adding these network perspectives!