Autonomous vehicles have absorbed over $100 billion in investment. Waymo is running real driverless taxis. Tesla is promising full autonomy. Here's the honest economics of where we actually are and what it means.
Educational content only — not personalized financial, legal, or tax advice.
In 2017, Uber’s CEO Travis Kalanick called self-driving trucks “existential” for the company. Ford and GM both announced plans for fully autonomous vehicles by 2021. Morgan Stanley predicted the autonomous vehicle market would reach $1.5 trillion by 2025. Dozens of startups raised billions with promises of imminent full autonomy.
It is now 2026. Waymo operates a few thousand autonomous vehicles in three U.S. cities. Tesla offers a “Full Self-Driving” system that requires driver supervision. Cruise (GM’s AV unit) suspended operations after a pedestrian injury and is being restructured. Argo AI (Ford/Volkswagen-backed) shut down completely in 2022, returning $1.4 billion to investors. Uber sold its AV division to Aurora in 2020. Apple cancelled Project Titan (its secretive AV program) in 2024.
The gap between 2017’s promises and 2026’s reality is not a failure of investment. It’s a demonstration that autonomous driving is harder than almost anyone admitted publicly — and that the economics of deploying it at scale are more complex than the technology.
This connects to The AI Job Displacement Question and to Trade Secrets.
Full autonomous driving — SAE Level 5, capable of operating in any conditions without human supervision — requires solving several simultaneous problems that have proven far harder than early researchers anticipated:
Edge cases: Human drivers can navigate an enormous variety of unusual situations through common sense and context. A child chasing a ball into the street. A construction worker directing traffic with unusual hand signals. A road flooded with unclear depth. An emergency vehicle approaching from an unexpected direction. The long tail of unusual situations is nearly infinite, and handling them safely requires either near-perfect sensor and AI capability or a system that conservatively stops when uncertain — which creates its own problems in traffic.
Sensor performance in adverse conditions: Camera-based systems (Tesla’s approach) struggle in heavy rain, snow, direct sun glare, and nighttime construction zones. LIDAR-based systems (Waymo, Cruise) are better in low visibility but are expensive and have their own failure modes. No sensor suite has achieved reliable performance across all real-world conditions.
Verification and validation: How do you prove a self-driving system is safe enough to deploy? Conventional aviation safety standards require demonstrating failure rates below 10⁻⁹ per flight hour. Applying similar standards to autonomous vehicles would require billions of miles of testing for each major software update. The regulatory pathway for broad commercial deployment remains unclear.
Waymo (Alphabet’s AV subsidiary) is the most advanced commercial autonomous vehicle operation in the world. It runs genuinely driverless rides (no safety driver) in San Francisco, Phoenix, and Los Angeles as of 2024 for paying public customers. This is not a pilot program — it’s a commercial robotaxi service.
Waymo’s approach has been methodical and conservative: deep sensor stacks (cameras, LIDAR, radar), high-resolution pre-mapped environments, conservative operating domains (defined geographic areas where the system has been extensively tested), and a willingness to accept restricted scope in exchange for safety reliability. Waymo does not operate in heavy snow, has not yet tackled all weather conditions, and limits its service areas carefully.
The economics: Alphabet has invested approximately $10+ billion in Waymo since 2009. Waymo is not yet profitable — its revenue from rider fares is a fraction of its operating costs at current vehicle counts. The path to profitability requires scaling to vehicle counts large enough to generate the fare revenue that covers the per-vehicle operating costs and amortizes the platform development investment.
Waymo’s strategic position is strong: it has the most real-world driverless miles, the most mature safety case, and partnerships with ride-hailing platforms (Uber Eats delivery, potential Lyft partnership) that provide distribution without requiring Waymo to build its own consumer-facing app at scale.
Tesla’s approach is fundamentally different from Waymo’s. Tesla uses cameras only (no LIDAR), relies on neural networks trained on data from its fleet of 6+ million vehicles, and distributes its Full Self-Driving software to customer vehicles as a supervised driver assistance system.
The FSD system has improved substantially since its first versions. It can now navigate complex highway merges, urban intersections, and parking structures in conditions that would have been impossible for earlier versions. But it still requires driver supervision — it makes errors that a competent human driver would not, and the driver must be prepared to intervene.
Elon Musk has predicted full autonomy by “next year” approximately annually since 2016. These predictions have proven consistently optimistic. The practical question — when FSD achieves the reliability level needed for unsupervised operation — remains genuinely uncertain. Tesla’s data advantage is real; whether it’s sufficient to solve the remaining problems is debated among AI researchers.
Tesla’s robotaxi ambition (Cybercab, announced 2024) represents a significant business model bet: if FSD achieves unsupervised reliability, Tesla vehicle owners could deploy their cars as revenue-generating robotaxis when not in use, with Tesla taking a platform fee. The economics, if it works, are compelling. The technology prerequisite has been delayed repeatedly enough that investors now discount the timeline significantly.