Robotaxis At A Crossroads 20 Years After DARPA

Xingjian "XJ" Zhang, Head of Growth at Apex.AI, examines the state of the autonomous vehicle industry.

2024 marked a significant milestone for the autonomous vehicle (AV) industry: the 20th anniversary of the DARPA Grand Challenge. This event sparked the modern era of self-driving technology.

At first glance, it felt like a banner year for robotaxis:

• Waymo established new service areas and increased its funding.

• Tesla unveiled the Cybercab, a sub-$30,000, two-seater electric robotaxi with wireless charging.

• Companies like Pony.ai and WeRide went public, riding high on optimism for a driverless tomorrow.

Yet progress in this sector is rarely as straightforward as it appears.

Reality Behind The Hype

The fanfare around IPOs and Tesla's ambitious project masks how tough the business is. Consider TuSimple, a company that was struggling in autonomous trucking and, as a result, announced its intent to delist from Nasdaq and pivot to AI animation. Aurora Innovation remains pre-revenue more than three years after IPO and postponed its first commercial deployment of “up to 10 driverless trucks” to April 2025.

While deep-pocketed Tesla and Google seem prepared for continued investments, not every giant is immune. Cruise, backed by General Motors, saw its fortunes reversed in a single year. GM injected $850 million into Cruise in mid-2024, only to pull the plug by December. Apple, after a decade exploring autonomous vehicles, abandoned the “Project Titan” in 2024.

These setbacks sharply contrast with the seemingly boundless revenue opportunities. Consider these recent forecasts from 2023: ARK Invest predicted that "autonomous taxis could generate $4 trillion in revenue in 2027." McKinsey estimated that by 2030, shared autonomous vehicles could achieve as much as $410 billion in revenue. GM projected that Cruise, now defunct, would generate $50 billion by 2030.

These predictions seem more like mirages today. This growing disconnect between aspiration and reality raises a critical question: Why has the autonomous driving industry struggled to deliver on its enormous promise?

High Capital Intensity

Car manufacturing is inherently capital-intensive, but robotaxi companies have taken this challenge to a new level by simultaneously investing in custom-built vehicles and centralized fleet networks. Their competition? Ride-hailing platforms like Uber, which operate with minimal fixed assets.

Consider the hardware alone: lidar, cameras, radars and GPUs. Waymo’s co-CEO estimates these components add $100,000 per vehicle, excluding the base vehicle, software development and safety testing. During a Waymo trip, I got to speak with a remote monitor—an interesting experience but a reminder of the significant labor costs. While Waymo’s safety monitor ratio is undisclosed, GM reportedly maintained a 1.5-to-1 ratio, and Tesla has begun building a teleoperations team for its future robotaxi services.

Two decades on, the robotaxi industry remains both capital- and labor-intensive. In contrast, Uber operates without owning vehicles; most drivers use older cars and work part-time. Recent research highlights how drivers often finance their vehicles through personal loans, effectively subsidizing Uber’s rapid expansion.

Robotaxi firms may disrupt Uber in the long run, but they face an uphill battle: They are highly backward-integrated, controlling everything from hardware manufacturing (e.g., Waymo & Aurora making lidars) to software development and fleet operations. While the potential of the autonomous mobility market is vast—akin to an untapped oil reservoir—the “drilling costs” are prohibitive.

High Safety Requirements

Despite billions invested and millions of miles tested, the 2023 Cruise accident was a stark reminder that in safety, "good enough" is not enough.

A detailed technical assessment of the accident, released in 2024, highlighted multiple critical technical flaws in Cruise’s Automated Driving System (ADS):

• The ADS intermittently lost the ability to classify and track the pedestrian, misclassifying it as an "unknown object."

• Due to inconsistent LiDAR detections and misinterpretation of the pedestrian’s leg, the system mislabeled the collision type and continued moving after the collision.

• A semantic mapping error then caused the AV to misjudge its stopping location, dragging the pedestrian 20 feet. The vehicle ultimately stopped only when its traction control system detected resistance from the pedestrian trapped beneath it.

If a company like Cruise—second only to Waymo in testing miles—faces such fatal technical vulnerabilities, it raises a troubling question: How safe are the other players in this space? Waymo’s safety reports often compare their performance to human drivers, but public trust is not built on statistics alone. Consumers expect more from “centralized” brands like Waymo than they do from “decentralized” driver networks like Uber. A single incident in a Waymo fleet attracts outsized scrutiny, much like a food safety issue at a McDonald’s franchise generates more public alarm than a similar event at an independent eatery.

According to JD Power’s 2024 US Mobility Confidence Survey, the confidence levels for "comfort riding in fully automated, self-driving vehicles" were scored only 36 out of 100. Overcoming this heightened skepticism will require a high standard of technical maturity.

The Road Ahead

Two decades after DARPA, the AV industry today faces an uphill battle in advancing its ambitious vision. This requires a rethinking of how to address capital intensity, value creation and public trust.

Is there an industry we can learn from? Consider this one:

• It revolves around a safety-critical product that is highly capital-intensive to build.

• Customers often think a brand makes the entire product, but 50% to 75% of its value is outsourced to suppliers.

• The core strength of these companies lies in system integration (“knowing more than they make”).

• They retain value and control the sector by acting as guarantors of safety, meeting stringent regulatory requirements, managing product liabilities (e.g., recalls) and delivering customer experiences.

Sound familiar? This is the playbook of traditional automotive OEMs. While the AV ecosystem has already created new supplier industries—data labeling, mapping companies, lidar suppliers and more—they still try to capture as much of the value chain as possible, “making more than they know” instead of the other way around.

Robotaxis can’t succeed as a solo act. Thriving requires collaboration and ecosystem building. I remain optimistic that one day, AVs will become trusted companions in transportation, seamlessly integrating into our lives and leading us to a world where people and places are connected with ease, safety and joy.

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?