Autonomous vehicles: Arriving soon in autopia

Executive summary: Autonomous vehicles (AVs) are here and more will arrive in 2026. Over 500,000 weekly rides are being currently taken in Waymos across America. With both technology and safety levels improving rapidly, scepticism towards driverless cars is diminishing. They can help save both time and lives as well as improving urban environments. Greater AV supply should fuel increased demand. By 2035, the overall autonomous vehicle market could be worth up to $400bn. The journey forward will not be linear. Inevitable accidents, increased regulatory scrutiny and fears of potential cyber-compromise may delay progress. Regardless, significant sums of (often private) capital are being ploughed into the industry. Waymo appears to be an early leader, but many other players are developing rival businesses, while companies such as Uber are seeking to position themselves as platforms for demand aggregation. Buckle up for an exciting journey.

The road to autonomous driving, or autopia (a phrase we first coined in 2015), has been long and winding. However, the foot is now truly on the accelerator. 2026 is shaping up to be the year of the robotaxi. With their emergence, cities globally may soon start to be reshaped. The industry is now at an inflection point. Expect more autonomous vehicles on a street near you soon.

Your author first took a ride in a Waymo in San Francisco last September. Users must rely entirely on technology to get them safely to their destination. It was both an exciting and unnerving experience. One ride led to more, in different conditions too (at night, when raining, and so on). In the short period since then, the industry has evolved rapidly. Indeed, such is the rate of progress that many of the statistics cited below may already represent a backward-looking perspective.

Waymo trials began in San Francisco in late 2022. The service formally launched in June 2024. The company reported that its vehicles had undertaken 14m trips in 2025, triple the levels of the year prior. By March this year, Waymo was seeing over 500,000 weekly rides across the ten US cities where it is currently operational, a tenfold increase since launch. There are currently over 3,000 Waymos in service across the States. The business has announced plans to enter at least a further ten cities before the end of 2027 including several outside the US. Your author hopes to be using Waymo in London shortly.

Not to be outdone, multiple other players have begun trials and services. These span a range of locations (further US cities, Europe, the Middle East and Asia) and formats (partial versus full autonomy, partnerships with ride hailing businesses, OEM rollouts and so on). In China, more than 10 commercial companies are currently operational, with testing underway in 50 cities across the country. Dubai has said that it aims to have 25% of all road transportation trips undertaken by autonomous vehicles by 2030.

Major improvements in technology and corresponding vehicle safety levels have helped win over the sceptics. Although the tech stack will vary by autonomous vehicle operator, most AVs are equipped with a complex array of hardware and software. These would typically include LiDAR (i.e. laser-driven detection sensors), cameras and radars, as well as AI and deep-learning, high-definition mapping and GPS positioning. Fear not, redundant systems are built in for safety, so if one sensor breaks, the car can still pull in data from others. Humans also typically oversee AV fleets and can intervene if needed.

Crucially, AVs benefit from a constant learning, or reinforcement environment. The vehicles use artificial intelligence in both the car and the cloud to mimic the way in which human drivers process (new) information and react to it. Logically, as robotaxi services expand, the algorithm becomes self-reinforcing. This explains why NVIDIA’s Jensen Huang said in January at the Consumer Electronics Show that “the ChatGPT moment for physical AI is here – when machines begin to understand, reason and act in the real world… robotaxis are among the first to benefit.”

When robotaxis first launched in San Francisco, two-thirds of residents were opposed to them. Two years on, the ratio had flipped. A November 2025 poll (reported in The Economist) suggested that more than 65% of San Franciscans were now in favour of unmanned vehicles. A separate study, by JD Power (a consumer research firm) found that confidence in AVs generally was 56 percentage points higher among those who had ridden a robotaxi versus those who had not. More use-cases in different conditions – say, snowy cities such as Denver, or those with old and narrow streets, like London – help further legitimise the technology.

The advantages of using a robotaxi ought to be clear: they are time- and lifesaving. Driving should become safer, more convenient and more enjoyable. Unlike humans, robotaxis do not get distracted, text, or drink and drive – and they have a 360-degree field of vision. Globally, car accidents result in around 1.2m annual fatalities and are the leading cause of death among people aged 5-29, based on data from the World Health Organisation. Furthermore, it is estimated that around 90% of accidents involve human error (this latter figure relates to the US market, and is according to information provided by the NHTSA).

Waymo reports that its autonomous vehicles have now driven more than 200m miles cumulatively without human assistance. During this period, the company notes that there have been 92% fewer serious injury (or worse)-causing crashes compared with a human driver over the same distance, and 81% fewer injury-causing crashes. Another study, by Waymo and Swiss Re, a reinsurer, finds that Waymos are involved in 88% fewer property-damage claims and 90% fewer bodily injuries per mile than humans.

Consider the other benefits. Since autonomous vehicles are in almost constant use, the more popular they become, the more they will free up space and make urban life more comfortable. With fewer cars and parking spaces needed, in time, cities could be redesigned to be more pedestrian-friendly and to have more green spaces. Uber has reported that Waymo AVs operating on its network complete more daily trips than 99% of the company’s human drivers. Vehicle arrival time is typically 25% faster in cities where present, relative to a trip undertaken in a conventional taxi. For context, the average working American spends just under an hour commuting each day, versus eight hours on the job.

The cost of both taking and operating an autonomous vehicle continues to fall. This is a function of scale economics and improved vehicle design. At present, self-driving vehicles cost about $7-9 a mile to operate, compared with $2-3 a mile for traditional ride-hailers and $1 a mile for personal cars, according to data from consultants BCG. At launch, Waymo’s vehicles were said to cost ~$150,000. However, its latest generation of AVs will likely see a more than 50% cost reduction. Not only do they use fewer sensors – since they have become more efficient – but the cost per sensor continues to drop. As with other technologies that start to reach maturity, sensors have seen a marked decrease in cost, equivalent to around a 100-fold fall over the last five years.

Unlike regular cars, robotaxis do not sit idle for most of the day, meaning that their main cost – the investment to build them – can be spread over many journeys. The key, then, is to deploy robotaxis in areas where there is demand, and cars can be kept occupied for long periods of time. Furthermore, as time goes on and the experience improves, AV companies will also need to rely less on remote vehicle operators. For a ride-hailing business, human drivers are estimated to account for 50-70% of the cost of a traditional taxi journey, so with no-one at the wheel, costs should fall rapidly.

Assume robotaxis become sufficiently safe and no more expensive than a human-driven cab, then demand should accelerate rapidly. More vehicle supply should also drive increased customer demand. Uber reports that in both Atlanta and Austin, following the launch of initial AV services, there was a 9 percentage-point year-on-year growth acceleration in first-time riders relative to pre- its AV launch, and an additional 3 percentage-point year-on-year increase in trips per rider, using the same parameters.

Project ahead and there could be up to 100,000 driverless taxis on the road by 2030. By then, they may have captured 30% of the US ridesharing market and be generating $7bn in revenues (data from Wood MacKenzie, Morgan Stanley and Goldman Sachs respectively). The overall autonomous vehicle market could be worth up to $400bn a decade from now, according to McKinsey. China alone could be a $180bn market by 2040, says UBS.

Many factors could, of course, undermine the journey to autopia. The market is still nascent. AVs still struggle with unpredictable scenarios (road conditions as well as human behaviour). Operating robotaxis at scale is untested. For now, the industry may be just one high profile accident away from being taken off the road. Consider, for example, the negative media coverage that accompanied the power outage experienced by Waymo’s fleet in San Francisco last December, and also the recent systems failure of Baidu’s robotaxis in Wuhan. Reputational risk remains significant.

Furthermore, governments almost everywhere have been cautious about approving fully autonomous vehicles due to safety and liability concerns. Self-driving cars are only currently permitted in around half of America’s states. The Governor of New York State blocked AVs in February 2026. Other major cities such as Boston and Seattle are currently considering whether to permit them. In Germany, all AVs today must be always monitored by a human.

There remain further legal, ethical and cyber concerns. The general rule that underpins AVs at present is collision avoidance at all costs. However, who decides the ethical rules: car manufacturers, governments or consumers? Further, were an AV to kill someone owing to a programmed decision, who would be liable: the developer, the owner or the algorithm? Embedding ethics into code inevitably creates accountability issues. Different cultures and legal systems have different moral priorities (such as minimising total harm, protecting passengers or following the law).

In addition, a fully autonomous car relies upon (as well as generating) huge amounts of data. There remain major uncertainties about the legal ownership of such data as well as the risk of it being corrupted or hacked. If AVs were compromised, this could lead to catastrophic and fatal failures.

The most likely reality is that the pace and scale of adoption will be uneven. Many people enjoy driving (or being driven, especially if they value the company) and others may resist giving up control. All vehicles – whether shared or personal – will increasingly embed greater levels of autonomy. The near future may be a hybrid model, where conventional and autonomous vehicles coexist.

Given the size of the prize, many players are seeking to gain exposure to the expanding autonomous vehicle market. There are also multiple (and overlapping) elements to the AV value stack. These comprise self-driving technology or software platforms; demand aggregation solutions (the role currently fulfilled by the likes of Uber and Lyft); asset ownership and financing; and vehicle manufacturing plus other adjacent hardware. Assume the AV industry follows others, then more value should accrue higher up the stack.

From a practical perspective, only one operator – Waymo – has been currently approved by US regulators for multiple use cases: both supervised and unsupervised operations, running 24-7, permitted on highways and at airports, and allowed to operate in rain and snow. Waymo, then, may have an early mover advantage, partly enabled by its safety positioning (with redundant sensors). It is also gaining major insights from continued real-world driving experience. Notably, Waymo has pursued partnerships in several geographies, including working with Uber in Nashville and Avis in Dallas.

Have no doubt, a lot of private capital is going into the autonomous vehicle industry, even though the ultimate winners may not yet be known. Consider the precedent of AOL’s dominance at the start of the dotcom boom and where the business is now (essentially defunct). Waymo’s January 2026 funding round implied a $110bn valuation for the business. For context, however, this is equivalent to less than 5% of its parent company, Alphabet’s market capitalisation, and so is unlikely to be a major driver of share price performance in the near-term.

Tesla is perhaps Waymo’s closest challenger from a tech stack perspective, although its approach to autonomous driving is very different. Whereas Waymo has moved straight into L4 autonomy (i.e. no driver required), Tesla has opted for an approach that requires less up-front investment. Using only cameras and no LiDAR, its vehicles currently permit less autonomy and so can be used in more limited circumstances, although Tesla does win out in terms of cost per mile and production advantages (from having vertically integrated manufacturing). The business has said that it hopes to be launching fully driverless services shortly. However, Elon Musk has a poor track record of meeting deadlines.

Amazon, via its Zoox subsidiary, has taken a different approach. Its driverless vehicles, active in San Francisco and Las Vegas trials currently, have been purpose-built (Tesla’s existing vehicles are being adapted, whereas Waymo uses cars provided by Jaguar and Hyundai) and are vertically integrated. They contain no steering wheel, no pedals and no driver controls at all. The seats face inward. The system is the driver. Such a model may be the future, even if there is little visibility on the economics of Zoox at present.

Major technology businesses such as NVIDIA, ARM Holdings and Western Digital also look well-placed, irrespective of which players ultimately claim the top positions in the AV market. The former is positioning itself as an essential supplier to the industry. NVIDIA’s Alpamayo-R1 model (launched in December 2025) allows vehicles to interpret their surroundings, explain their observations and outline their planned actions using natural language. ARM’s IP architecture will also be essential for many AV players. It already has long-standing partnerships with several auto businesses including Tesla and Nuro (a US AV tech company). Meanwhile, since every mile driven by AVs will need to be stored for model training and safety purposes, storage companies such as Western Digital should benefit.

The role that Uber and Lyft will play in the development of the AV could also be crucial. Both businesses see themselves as demand aggregators that benefit from being technology agnostic. Uber (the larger of the two) aims to be “the largest facilitator of AV trips in the world," per its Chief Executive. The company has signed partnerships with over 20 companies and sees itself as a platform business, acting as the intersection between hardware players (Lucid, Stellantis, Mercedes, VW etc) and self-driving technology businesses (Baidu, Motional, Nuro, NVIDIA, Pony, Waymo, Wayve etc).

Uber and Lyft clearly hope to leverage their network of riders to provide higher AV utilisation as well as fleet management capabilities, and in turn drive better returns for AV tech companies and AV fleet owners. Such a strategy is inevitably self-serving, since removing the driver could be seen as an existential threat to these businesses. Neither Uber nor Lyft has an interest in the market narrowing to just two players (Waymo, Tesla), hence why they are supporting so many self-driving tech companies.

In the near-term, a hybrid mix of humans and AVs will be beneficial to consumers, minimising pick-up times and maximising coverage. The road forward for the autonomous vehicle industry will not be straight, but it will certainly be exciting.

Alexander Gunz

April 2026

The above does not constitute investment advice and is the sole opinion of the author at the time of publication. Heptagon Capital is an investor in ARM Holdings and Uber. The author of this piece has no personal direct investment in the business. Past performance does not predict future returns, the value of investments and income from them can fall as well as rise.