CyberCab, Tesla's two seat purpose built robotaxi, is already in limited production at Giga Texas near Austin. Tesla VP Lars Moravy argues that ARK Invest and Wells Fargo have built cost models without access to the company's cell to service stack.
Tesla has something analysts have not seen on paper: a robotaxi production line that has been quietly running for months, the first pieces of a humanoid-robot factory, and a regulatory shortcut that none of its competitors can use. On July 7, the automaker plans to disclose how that stack is actually performing, according to Tesla VP of Vehicle Engineering Lars Moravy, who has now teased the "Scaling Big" announcement twice in conversation with the Podcast Alpha show, including in the first 22 seconds of the episode.
The peg is not a launch. It is a production-scale disclosure that will put a number, or decline to put one, next to a thesis that public analyst models have been building without the data to test.
CyberCab is Tesla's purpose-built two-seat robotaxi, designed without a steering wheel or pedals for autonomous operation only. Production at the Giga Texas plant outside Austin started in February and moved into continuous mode in April, according to a recap by [Electrek](https://electrek.co/2026/04/23/tesla-cybercab-production-starts-no-nhtsa-2500-vehicle-cap/). Moravy told [Podcast Alpha](https://podcastalpha.substack.com/p/tesla-cybercab-july-7-av-cost-gap) that lines are running at 90 percent or higher automation, roughly 150 units have been built, and most of them have already been redirected to act as training vehicles for the next unsupervised build of Full Self-Driving. A perimeter burn-in track is live on the campus, where completed vehicles are run through suspension and noise checks before they leave the lot, and Tesla has wired in cabin microphones so completed cars self-inspect for squeaks and rattles on the way out the door.
The regulatory shape of the line matters as much as the production math. CyberCab units spotted at Giga Texas carried federal compliance stickers under the FMVSS self-certification pathway, which lets a manufacturer certify its own vehicles against federal motor-vehicle safety standards. This sidesteps the NHTSA exemption cap of 2,500 vehicles per year that applies to Waymo and Cruise, per [Electrek](https://electrek.co/2026/04/23/tesla-cybercab-production-starts-no-nhtsa-2500-vehicle-cap/) and Moravy's own X post. That asymmetry shows up in the regulatory rulebook rather than the unit-count race, and Tesla did not have to ship more cars to widen it.
Beyond cars, the first Optimus production line has landed at the Fremont factory in California, with roughly 40 lines planned and the same teams that built the car lines leading the build, Moravy said. Optimus is Tesla's humanoid robot project. [Basenor](https://www.basenor.com/blogs/news/teslas-first-optimus-production-line-has-landed) reported that Gen 3 Optimus mass production began at Fremont in January 2026, and the Gen 3 hand subsystem was production-ready by mid-February. Sawyer Merritt, a Tesla-watcher with a verified X account, shared an image described as Musk's first look at the line on July 1, 2026, and Moravy described the line as modular so it can be retooled as Optimus hardware evolves.
Put the campus and the cells together and the stack runs from aluminum and 4680-cell ingot, through casting and vehicle production at the same site, to the ride-hail service. A [NextBigFuture](https://www.nextbigfuture.com/2026/06/tesla-vp-lars-morevy-updates-unsupervised-robotaxi-and-cybercab.html) recap notes that Giga Texas has produced its 500,000th vehicle as of October 2025 and has capacity for up to roughly 10,000 vehicles a week, with 4680 cell production localized on the same campus. An outside analyst modeling the cost of a robotaxi mile has to estimate each of those steps separately, using public inputs that Tesla does not break out. That is the argument Moravy is making for why both ARK Invest and Wells Fargo, in his telling, have the cost models structurally off: their work, however careful, cannot see through walls Tesla has decided not to open.
That framing belongs to Tesla, not to independent cost analysis. [Podcast Alpha](https://podcastalpha.substack.com/p/tesla-cybercab-july-7-av-cost-gap) is the venue and Moravy is the speaker, and neither ARK nor Wells Fargo has published a public rebuttal as of this writing. The honest read is that public cost models are not wrong in the usual sense, they are built from the outside, and a vertically integrated manufacturer can hold a cost position no outsider can replicate from public inputs.
What July 7 might do is force Tesla to either publish something directional (cost per CyberCab mile, or per-vehicle cost at the full stack level) or keep the stack opaque. If the company discloses even a range, ARK and Wells Fargo can re-anchor their models on real data. If it declines, the thesis remains unprovable from outside and the event is a production milestone rather than an analytical resolution.
There is also a workforce caveat. [Electrek](https://electrek.co/2026/04/23/tesla-cybercab-production-starts-no-nhtsa-2500-vehicle-cap/) reported in March that vehicle program manager Victor Nechita, OTA and ride-hail director Thomas Dmytryk, and assembly leader Mark Lupkey all left between February and March 2026. Tesla no longer had any of the original program managers running its production vehicles as of that reporting, a fact that does not by itself break the ramp, but raises the question of who is carrying operational knowledge into the next phase. Confirmation of any current-state departures beyond March 2026 is not available.
A separate, smaller number to handle is the supervised-crash rate Tesla has been quoting on its own robotaxi fleet: roughly one crash per 57,000 miles, against a human-driver benchmark of about one per 229,000 miles, per [Electrek](https://electrek.co/2026/04/23/tesla-cybercab-production-starts-no-nhtsa-2500-vehicle-cap/) citing Tesla's own reporting. The Tesla figure has not been independently validated, and the human baseline depends on which dataset you start from.
For July 7, the watchlist is short. Look for disclosed line count and automation rate; any cadence framing for CyberCab output through the second half of 2026; any directional per-mile cost number; and any Optimus update that pins down how many of the planned 40 lines are now active. The single question that resolves the cost-gap debate is whether Tesla discloses, even directionally, what an unsupervised CyberCab mile costs the company at the full ingot-to-service stack level. If yes, the analyst models can be re-anchored. If no, the gap stays a black box on both sides of the argument.