The robotaxi thread promised 20-cent miles. We pulled two Austin receipts: about $2.80 a mile.

A widely shared X thread — a Norwegian post by DNB Asset Management portfolio manager and Tesla bull Audun Wickstrand Iversen (@IversAudun), quote-tweeting a Tesla site in San Antonio — argues that robotaxis will make car ownership, parking, accidents, insurance, tickets, DUIs and even snow plowing 'gradually disappear' over 10 to 15 years. Its load-bearing number is a per-mile cost of 20 to 30 cents, set against roughly 70 cents for a private car and 200-plus cents for a human-driven taxi. We read the thread line by line against the record and pulled the actual fares. The 20-cent figure traces to Elon Musk at Tesla's 'We, Robot' Cybercab unveiling in October 2024, where he put operating cost at 'probably around 20 cents a mile,' explicitly 'over time' and 'at scale' [marketing]. It is a projection, which no fleet tracked here has demonstrated. The largest at-scale, fully driverless paid US fleet tracked here is Waymo, which reported about 500,000 paid rides per week across roughly 10 metros as of March 2026, up about tenfold from May 2024 [verified, company-disclosed]. Analyst estimates put Waymo's fully loaded cost near $1.40/mile today [modeled, analyst estimate, no company disclosure] — roughly 5 to 7 times the 20-to-30-cent target. Tesla's own Austin robotaxi fares, from two legible receipts in the X thread under review, were $6.99 for 2.5 miles and $6.84 for 2.4 miles — about $2.80/mile on those two short trips, base fare included [reported, operator-captured]. Tesla's metered Austin rate runs nearer $1.25 to $1.40/mile on longer trips [reported]. The private-car '70 cents' matches the IRS 2025 standard mileage rate of 70.0 cents and the AAA 2025 full-ownership average of 77 cents at 15,000 miles/year. Both are full-cost figures. The marginal cost of driving a mile in a car you already own — fuel plus maintenance — is about 24 cents [modeled from AAA components]. Rideshare and metered taxis run roughly $2 to $4/mile as a retail price, not a cost [reported]. Tesla's Austin service is geofenced to about 245 square miles and remains predominantly supervised: as of March 31, 2026, only about 4 to 8 of roughly 37 to 42 Austin vehicles ran without an in-car monitor, and those are remotely supervised. In San Antonio, a Zoning Commission voted 8-2 on July 21, 2026 to recommend rezoning a parking-and-charging depot near the Pearl in Tobin Hill — city staff had recommended against, City Council had not voted, and it authorizes no driverless operation.

The following is analysis, not fact. The thread's core comparison appears to mismatch its terms three ways ^mmodeled. First, cost versus price: the 70-cent car is a cost, the 200-cent taxi is a retail price that includes a driver's wages and a platform margin. Second, full versus marginal: the honest 'should I drive or ride' number for a car you own is about 24 cents, which is within the 20-to-30-cent target range, not above it. Third, demonstrated versus projected: the car and taxi numbers are measured today, while 20 to 30 cents remains an unproven future target ^mmodeled. Compare like with like and the story inverts — today a car costs about 77 cents/mile all-in versus an estimated $1.40 for a robotaxi, and two Austin receipts read about $2.80. The 20-to-30-cent figure is a forecast, not a measurement. At-scale manufacturing has driven costs down before, so the target is not far-fetched ^sspeculative. But no fleet in this review has demonstrated it, and the fleet closest to scale appears not yet profitable ^mmodeled. The 'everything disappears' framing reads as a category error ^sspeculative. These functions are more likely to restructure than vanish ^sspeculative: liability would shift from millions of individual auto policies toward concentrated operator and manufacturer product-liability — a different insurance market, not the absence of one ^sspeculative. Robotaxis still crash: Waymo recalled 3,067 vehicles in December 2025 over software that let them pass stopped school buses, under an open NHTSA investigation. Snow may be the thread's weakest claim ^sspeculative: heavy snow appears to remain a significant operating limitation for the systems tracked here ^mmodeled — Waymo pauses in heavy snow, and a robotaxi that stops in a blizzard could mean more plowing, not less ^sspeculative. And we identified no peer-reviewed measured decline in car ownership or miles driven attributable to robotaxis — the honest answer is that it is too early. Modeling actually points the other way, toward more miles via empty repositioning and induced demand ^mmodeled. The economics argue for a slower, ODD-limited path ^sspeculative. Alphabet's Other Bets, which houses Waymo, posted a 2025 operating loss of about $7.5 billion [verified, SEC segment]; Sundar Pichai has said he expects Waymo to contribute meaningfully to Alphabet's financials as soon as 2027 [reported] — a contribution target, not a profitability guarantee. Cruise, once among the largest US robotaxi operators, spent more than $10 billion, admitted a false report to regulators over an October 2023 pedestrian-dragging and paid a $500,000 criminal fine, and GM stopped funding it in December 2024. The timeline record is worse still: at Tesla's April 2019 Autonomy Day, Musk said there would be 'over a million robotaxis' on the road the next year. Waymo eventually delivered a real service — this is not vaporware — but the field's hard timelines have repeatedly slipped.

The strongest honest case for the thread is that the tech is real and improving fast, and that costs at scale historically fall. Waymo's roughly tenfold ridership growth in under two years is not a slide-deck number — it is paid rides on public streets, and peer-reviewed work (Kusano et al., 2025, in Traffic Injury Prevention) found about 79 to 81 percent fewer injury crashes for the driverless system, a large effect even after discounting for the authors being Waymo employees and the operating domain excluding freeways and severe weather. If Cybercab reaches purpose-built, mass-produced, single-purpose vehicles running dense city loops, per-mile costs genuinely could fall toward the marginal cost of a private car — and at that point ownership economics in a few dense metros really would come under pressure ^sspeculative. Fewer humans owning idle cars could reduce parking demand ^mmodeled, the one second-order effect with a plausible mechanism. The direction the thread points is not crazy; the record simply does not yet show the arrival. A lower crash rate within a geofence is not the elimination of crashes; a pilot in 10 metros is not a finished national transition; and the front-runner apparently not yet profitable ^mmodeled while a major competitor shut down points toward gradual, city-by-city, weather-limited expansion ^sspeculative — not the wholesale disappearance of cars, parking, tickets and plows on a 15-year clock.

From China, the framing looks less like a Tesla story and more like a two-country race ^sspeculative. Baidu's Apollo Go reported a weekly-order peak above 300,000 in late 2025 and more than 20 million cumulative rides as of its February 2026 earnings, across roughly 26 cities [reported] — a driverless network of a scale that appears comparable to Waymo's ^mmodeled, built under a different regulatory posture. The at-scale question the thread treats as settled looks, on the global evidence, still contested in two markets at once ^sspeculative.