Self-Driving Technology Enters a New Phase of Development

On a Tuesday morning, a Waymo car is parked outside a San Francisco coffee shop with the doors closed and no one inside, waiting. A woman passes it without giving it a second glance. Its bumper is cut by a delivery cyclist. A child on the pavement looks at it for a second before moving on. Ten years ago, the driverless car that was sitting there would have seemed like science fiction to most people. However, for some reason, the sector that created it is not rejoicing. It’s adjusting.

No single product launch or regulatory approval is the most obvious indication that self-driving technology has entered a new phase in 2026. The tone is what matters. The lofty goals of the past, such as having completely autonomous vehicles on every highway by the middle of the 2020s and making human drivers optional within ten years, have been replaced by something more measured, truthful, and, in some ways, more intriguing. The glitz is waning. Its replacement is more difficult to capture on camera, but it’s likely more long-lasting: a real assessment of the true costs of deploying and maintaining a driverless vehicle in commercial use.

This was explained somewhat clearly in McKinsey’s third biannual survey of leaders in the autonomous vehicle industry, which was published in early January. In comparison to 2023 projections, adoption timelines for almost all use cases have been delayed by one to two years. Previously anticipated to be widely implemented worldwide by 2029, level 4 robo-taxis are now scheduled for 2030. It is now anticipated that fully autonomous trucks, a category that recently attracted massive investment and breathless press releases, will become commercially viable by 2032 instead of 2031. These are not significant changes. However, they are significant because they show a pattern—a persistent propensity to underestimate how challenging the last stretch actually is.

It’s difficult to ignore the fact that the regional picture has also become more distinct. China and the United States are far ahead of everyone else, and the difference is growing rather than contracting. In the US, there are over 450,000 autonomous commercial rides per week; in China, there are over 250,000. In contrast, Europe is still mostly in the pilot stage and is carefully navigating regulatory frameworks that were not created with autonomous vehicles in mind. Faster development cycles, more accommodating road environments, and regulatory bodies willing to experiment are just a few of the factors contributing to the American and Chinese lead, but the end result is a two-tiered global industry with everyone else observing from a considerable distance.

Perhaps the most sobering development of all is the cost question. In the industry, there was a widely held belief that the development costs of autonomous vehicles would decline in a manner similar to that of other technological hardware. This belief was repeated so frequently that it was considered common knowledge. The reasoning made sense: more deployment, more data, less expensive sensors, and less expensive software.

That isn’t quite how things have turned out. Compared to previous estimates, the cost of attaining Level 4 autonomy in trucking has increased by 50–60%. Verifying performance across a vast array of edge cases—the odd, the improbable, and the truly uncommon—is necessary for the shift from testing to actual commercial deployment, and this verification is costly in ways that don’t neatly compress. Over $3 billion could be spent on software to enable fully autonomous driving. Conversations are halted by that number.

Where the industry believes the short-term money truly resides is changing, in part due to that cost reality. Instead of aggressively pursuing Level 3 systems, which permit hands-free driving under specific circumstances, the private passenger car market is moving back toward Level 2+—enhanced driver assistance that keeps humans in control but makes their job easier. 52% of experts predicted that Level 3 would dominate mass-market passenger cars in the 2023 survey. That percentage is now only 39%. According to nearly half of experts, Level 2+ will continue to be the commercial hub until 2035. Slower-than-anticipated cost reductions and the sheer difficulty of meeting regulatory standards for higher autonomy levels are the reasons behind this quiet but significant retrenchment.

Ford revealed in January that it intends to provide eyes-off driving technology in a $30,000 electric car by 2028, which is noteworthy for the goal and the price range. Democratizing this technology at that cost might actually alter public expectations in ways that the industry hasn’t yet fully anticipated. Or, as is common in this industry, the timeline may slip.

Observing all of this, one gets the impression that the story of autonomous vehicles is not coming to an end; rather, it is simply getting more complex than the initial iteration could handle. Waymo is growing. Later this year, Tesla plans to introduce cars without steering wheels. The robo-taxi is real; it operates real routes in real cities with real passengers. The effectiveness of this technology is no longer a question. It’s whether it operates at a scale and cost that supports a company, appeases authorities, and gains the confidence of the people it is meant to serve. Five years ago, no one publicly acknowledged how difficult that question is. To its credit, the industry appears to be asking the right question at last.