Elon Musk’s robotaxi story contains two separate forecasts. Confusing them makes it harder to judge either his credibility or the technology’s prospects. One deadline has long passed. The other is still ahead. Neither, by itself, answers the more interesting question: could autonomous electric taxis eventually make convenient transport dramatically cheaper?

At Tesla’s Autonomy Investor Day on 22 April 2019, Musk forecast deployment the following year, meaning 2020. In the published Quartr transcript, he said: “we will have over a million Robotaxis on the road”. That forecast was missed. Tesla subsequently dated its first Robotaxi service to June 2025 in Austin, with a limited launch that included a safety rider. [1]; [2].

In his CNBC interview with David Faber on 20 May 2025, discussing US self-driving deployment, Musk placed a different forecast “probably by the end of next year”, with “probably hundreds of thousands, if not over a million Teslas”, according to the published transcript. That means end-2026, a deadline that has not elapsed. The discussion concerned unsupervised driving; those cars would not necessarily all be actively carrying paying robotaxi passengers. [3].

The missed 2020 date matters because timing is part of a business proposition. Years of delay postpone customer benefits and revenue, leave capital committed for longer and give competitors time to improve. Repeatedly optimistic forecasts also justify demanding stronger evidence before accepting the next one. A prediction can eventually become directionally right while having been materially wrong for people making decisions against its original timetable.

But a missed deadline cannot establish that the destination is worthless. The strongest bullish argument starts with an unusually valuable possibility: combining inexpensive electric vehicles with the removal of the onboard driver could reduce the cost of supplying a journey. High utilisation could spread vehicle ownership costs across more miles. Reliable software could be replicated more quickly than a human workforce can be recruited and trained.

ARK Invest’s long-standing bullish research illustrates the distinction: it modelled very low fares at scale, while also arguing that limited competition could support higher prices. Even a successful low-cost platform need not pass every saving to passengers. Models remain scenarios rather than achieved prices. [4].

Tesla’s manufacturing experience makes that argument worth considering. If it can build durable vehicles cheaply, validate autonomy and operate a busy fleet efficiently, those strengths could reinforce each other. More driving data might improve the system, provided the data are useful and improvements are validated. This is a conditional business case, however, rather than an established account of profitable operations.

There is now more to assess than a demonstration. Tesla’s latest quarterly update reports nearly two million cumulative unsupervised Robotaxi miles by 30 September, and Cybercab rides beginning in Austin on 3 September. Those are company-reported milestones, not independent proof of nationwide readiness or profitable economics. Its consumer Full Self-Driving product still requires active supervision. [5]; [6].

Other operators also show that driverless passenger services are possible. Waymo reports substantial rider-only experience and favourable safety results in its operating areas. Its evidence deserves attention within those geographic and methodological limits; it does not establish Tesla’s safety or show that every existing Tesla can become an unattended taxi. Industry success and Tesla winning the market are separate propositions. [7].

The strongest bearish argument concerns the distance between bounded success and dependable service across America. A taxi must handle unusual road layouts, emergency scenes, poor weather, awkward pickups and failures that passengers cannot resolve themselves. Removing the driver also leaves cleaning, repairs, insurance, customer support and remote assistance to organise and pay for. Competition and regulation can constrain expansion and margins even when the driving technology works.

Safety is particularly difficult to establish through mileage alone. Rare severe events require large samples, while routes, conditions and comparison groups affect the result. RAND’s research explains why conventional road testing alone cannot efficiently settle every safety question. That supports combining operational evidence with other assurance methods; it is not an instruction to wait for billions of miles before allowing any deployment. [8].

Then there is the seductive figure of 30 cents a mile. At Tesla’s We, Robot presentation on 10 October 2024, Musk said the “operating cost is probably gonna be around $0.20 a mile”. For customer pricing including taxes, he added: “probably ends up being $0.30 or $0.40 a mile”, according to the published Quartr transcript. These were projections, not demonstrated nationwide fares. All figures here are US cents. [9].

Three quantities need separating: the fare a passenger pays, the operator’s total cost per vehicle-mile, and its cost per occupied or billable mile. An empty vehicle travelling to collect someone still consumes energy, tyres and useful vehicle life. A cheap operating mile therefore need not produce an equally cheap passenger mile, and a low cost need not translate into a low price if demand supports higher fares.

Consider a simple hypothetical. If every vehicle-mile costs 30 cents, and passengers occupy the car for 70% of its mileage, the cost is 42.9 cents per occupied mile. If only half the mileage carries passengers, it becomes 60 cents, before profit or taxes. These are arithmetic illustrations, not estimates of Tesla’s actual accounts. For a separately billed pooled service, passenger numbers and billing arrangements would add further complications.

Empty mileage is a real operational issue. An independent analysis of Waymo’s California filings found passengers aboard for approximately 53.6% of vehicle-miles between August 2023 and December 2025. That describes one operator in one state during a past period. It cannot determine Tesla’s future utilisation, but it shows why the denominator matters. [10].

For another illustration, assume a $30,000 vehicle lasts 300,000 miles with no resale value. Recovering its purchase price contributes 10 cents per vehicle-mile before financing. Electricity at 16 cents per kilowatt-hour, used at 0.25 kilowatt-hours per mile, adds four cents. A 30-cent vehicle-mile cost target would leave 16 cents for every other expense; a 20-cent target would leave six. These are hypothetical assumptions, not Tesla operating data. Insurance, cleaning, repairs, remote support and financing decide whether the sums work.

Credible 30-cent fares would require low vehicle costs, long useful lives, manageable insurance and support costs, and enough demand to keep cars productively occupied. The available evidence does not justify a calendar promise for nationwide availability. Watch sustained unsubsidised pricing alongside complete operating costs, service reliability and empty-mile rates. A promotional fare can demonstrate willingness to subsidise growth without demonstrating a sustainable transport model.

For British readers, there is already a route to commercial driverless passenger pilots before the full Automated Vehicles Act framework, expected in late 2027. Permit regulations took effect on 15 May 2026: they cover taxi-like services in England and bus-like services across Great Britain. This is not automatic approval for any operator or a UK-wide taxi licence. Vehicle approval, operational assessment and relevant local consent remain material. [11]; [12].

The EU has a separate approval system; Europe is not a single legal market. RDW’s April 2026 approval concerned FSD Supervised, whose driver remains responsible, rather than an unattended taxi. EU rules for fully automated vehicles also preserve member states’ control over circulation and safe operation in local transport services. An assistance-system approval therefore cannot establish permission to launch a driverless fleet, and EU approval does not itself authorise British deployment. [13]; [14].

Practical adaptation matters too: British left-side traffic, right-hand-drive customer cars, local roundabouts, narrow streets and safe kerbside boarding require validation, not assumptions carried over from Austin. For purpose-built driverless vehicles, passenger access matters more than steering-wheel position. British pilot guidance calls for insurance covering the operation and an accessibility plan, including booking, emergencies and assistance dogs. Local electricity, insurance, tax and support costs also mean a US-cent target cannot simply become an equivalent penny or euro-cent fare. [15].

If reliable, affordable robotaxis spread, the benefits could be substantial. Non-drivers and some older people could gain independence. Households might manage with one fewer car. Travellers could recover time, and reduced private parking demand could release useful urban space. But accessibility requires suitable vehicles, boarding arrangements, booking options and human help when needed; removing the driver can also remove assistance that some passengers rely on.

Cheaper journeys could also encourage extra trips, longer journeys and a switch away from buses or trains. Empty repositioning adds traffic. Access could remain unequal between profitable urban districts and less attractive routes. Electric propulsion eliminates exhaust emissions at the vehicle, while the wider climate impact depends on electricity supply and total travel. Cars still occupy road space. Automation alone cannot guarantee less congestion or a better transport system; pricing, pooling and integration with public transport will shape the outcome. [16].

Judge progress through active driverless vehicles, paid trips, geographic coverage, waiting times, independently scrutinised safety and cost per occupied mile. Musk’s missed deadline is evidence about forecasting reliability. Successful, affordable service would be evidence about eventual value. Both deserve to count.

Sources

The links below support the statements attributed to them. A source’s own claims are identified in the article.

  1. 2019 transcriptQuartr transcript via Stock Analysis · Source evidence; company projections and transcript limitations attributed in text
  2. Tesla’s Q2 2025 updateTesla via SEC · Source evidence; company projections and transcript limitations attributed in text
  3. Published interview transcriptElon’s FAQ published interview transcript · Source evidence; company projections and transcript limitations attributed in text
  4. ARK’s ride-hailing analysisARK Invest · Source evidence; company projections and transcript limitations attributed in text
  5. Tesla’s Q3 2026 updateTesla · Source evidence; company projections and transcript limitations attributed in text
  6. FSD supportTesla · Source evidence; company projections and transcript limitations attributed in text
  7. Waymo’s June safety updateWaymo · Source evidence; company projections and transcript limitations attributed in text
  8. RAND’s autonomous-vehicle testing studyRAND · Source evidence; company projections and transcript limitations attributed in text
  9. 2024 transcriptQuartr transcript via Stock Analysis · Source evidence; company projections and transcript limitations attributed in text
  10. Independent mileage studyFindings · Source evidence; company projections and transcript limitations attributed in text
  11. Permit regulationsUK legislation · Source evidence; company projections and transcript limitations attributed in text
  12. VCA pilot requirementsVehicle Certification Agency · Source evidence; company projections and transcript limitations attributed in text
  13. RDW explanationRDW · Source evidence; company projections and transcript limitations attributed in text
  14. EU automated-driving regulationEUR-Lex · Source evidence; company projections and transcript limitations attributed in text
  15. UK pilot guidanceUK Government · Source evidence; company projections and transcript limitations attributed in text
  16. US Department of Energy transport overviewUS Department of Energy · Source evidence; company projections and transcript limitations attributed in text