Nvidia has found a driver for the largest robotaxi order on Europe's books. On October 21, 2026, Bolt announced that its planned fleet of at least 25,000 autonomous Lucid vehicles will run Nvidia's Alpamayo Stack, starting with a data-collection program in Madrid in 2027, according to Electrek's report. The cars won't carry riders at first, and even the data collection needs regulatory approval.
The headline sounds like a launch. It is closer to a software choice for a program that is still years from public service. This post separates what is confirmed from what is undated, and puts it in context with Nvidia's growing automotive footprint. We covered the earlier pieces in our write-ups of Alpamayo 2 Super, Volvo's Hyperion and Alpamayo plans and the Waymo 20,000-vehicle report.
TL;DR
| Question | Answer |
|---|---|
| What was announced? | Bolt's Lucid robotaxis will use Nvidia Alpamayo Stack |
| Fleet size | At least 25,000 vehicles, announced in September |
| First step | Data collection in Madrid in 2027, no riders at the start |
| Regulatory status | Even data collection still needs regulatory approval |
| Next markets | Germany and France named |
| Which Lucid model? | Not stated by Bolt |
| Driverless rider date | None |
| Firm order? | Reported as not yet placed; treat as a plan |
What was missing in September
When Lucid and Bolt announced the 25,000-vehicle deal in September, nobody said who would write the driving software. The vehicles were to be built around Nvidia Hyperion, which Electrek describes as a reference design for compute and sensors, not a self-driving system. Bolt said it would bring in autonomous-driving partners later.
Now it has picked Nvidia again. Alpamayo Stack, per Electrek, is Nvidia's full-stack driving software: "reasoning AI" combined with classical perception and planning, running on Hyperion with Halos safety tools layered on top. Bolt and Nvidia will tune it for European driving, and Lucid is only the first automaker Bolt plans to equip with it.
Confirmed versus unconfirmed
| Item | Status |
|---|---|
| Alpamayo Stack selected for Bolt's Lucid fleet | Announced by Bolt, per Electrek |
| Madrid data collection in 2027 | Announced, still subject to regulatory approval |
| No riders in the first Madrid cars | Stated |
| Germany and France as next markets | Named, no dates |
| Which Lucid model goes to Madrid | Not stated |
| Date for driverless paid rides | None |
| 25,000 vehicles | Plan; Electrek says "at least"; a TechCrunch summary says no order or payment yet |
| Bolt's 100,000-vehicle target | Stated goal for 2035 across Europe and the UNECE region |
The vehicle question matters. Electrek notes the September deal was designed around Lucid's upcoming Midsize platform, whose first vehicle, the Cosmos, was pushed to 2027 in Lucid's August operating reset, most likely the second half. If the purpose-built vehicle is not ready, Madrid's first data-collection cars would have to be existing Gravity or Air vehicles. That is Electrek's reasoning, not Bolt's statement.
Why Nvidia selling the driver matters
For years Nvidia sold chips and sensor reference designs to automakers and left the driving software to others. With Alpamayo Stack, it is selling the driver too. Electrek says that puts Nvidia up against Nuro, Waymo and Tesla's software.
Car on a curving road with a green path ahead, illustrating robotaxi autonomous driving software on Nvidia Alpamayo Stack
Lucid now has two robotaxi customers on two different autonomy systems: Nuro's software for Uber's Gravity robotaxis in the US, and Nvidia's for Bolt in Europe. Electrek totals Lucid's robotaxi commitments at at least 35,000 Gravity vehicles for the Uber program, up from Uber's original 20,000, plus at least 25,000 for Bolt, or 60,000 vehicles. For scale, it notes Lucid delivered 15,841 vehicles in all of 2025 and 10,852 in the first nine months of 2026. The orders on paper equal almost four years of current deliveries.
That mismatch is the sober context. Software choices are cheap to announce; building tens of thousands of purpose-built vehicles is not.
Where this sits in Nvidia's autonomous-driving push
Alpamayo is Nvidia's reasoning-style driving model family. Our August write-up of Alpamayo 2 Super describes a 34B open-weight model licensed for commercial robotaxi use that explains its own decisions. Volvo's expanded partnership puts Hyperion, Halos OS and Alpamayo in next-generation cars before 2030, with supervised features first. The Bolt deal adds a robotaxi operator, which is a different customer: fleet operators want a driver they can scale across cities and update centrally.
Bolt and Nvidia first announced a collaboration at GTC in March 2026, per Engadget's report of the partnership: fleet data feeding a training pipeline on Nvidia tools, with Cosmos for searching driving data and Omniverse for digital twins. Bolt promised its fleet data would comply with GDPR and said it would provide open-source access to European universities and small businesses. Those are statements from the March announcement, not from this one.
The competitive picture
- Waymo operates the largest US robotaxi service, as in our coverage of its expansion to 14 US cities and the unconfirmed 20,000-vehicle fleet report.
- Nuro supplies the software for the Uber and Lucid Gravity program. Uber employees have been taking test rides in San Francisco since April with a safety operator, per Electrek, with a public Bay Area launch still promised for 2026 and Houston in mid-2027.
- Tesla sells its own driving software.
- Wayve is the UK-based competitor in the Uber UK robotaxi launch.
- Bolt's hedge: it runs a separate track with Pony.ai and Stellantis, which announced a pilot in Luxembourg in June.
Why data collection comes first
A driving stack trained mostly on other countries' roads does not transfer cleanly. Lane conventions, signage, roundabouts, narrow historic streets, scooter and cyclist behavior and even how pedestrians negotiate crossings differ across Spain, Germany and France. Starting with a rider-free data-collection fleet lets Bolt and Nvidia record real European driving, replay it in simulation, find the cases where the model's choices diverge from a human's, and tune and validate before any passenger is involved.
This is the standard staged path for robotaxi programs: collect and validate with safety drivers or without passengers, then supervised pilots, then driverless service in a geofenced area, then expansion. Electrek's reporting puts Bolt at step one with no date for step three. For Nvidia the data loop is also strategic: its tools for searching driving data and building digital twins, mentioned in the March announcement, are most valuable when a large fleet feeds them continuously.
Reasoning-based driving versus the classic stack
Nvidia describes Alpamayo Stack as combining "reasoning AI" with classical perception and planning. In plain terms, older autonomous stacks split driving into separate modules, perceive, predict, plan, control, each hand-engineered and tested on its own. Newer designs add or replace parts with large learned models that look at the scene and decide, and some, like the Alpamayo family per our earlier coverage, can explain the choice in language.
The trade-off is verification. A modular stack can be unit-tested; a learned model must be tested statistically, across many scenarios. That is why Nvidia pairs the stack with Halos safety tools and why regulators ask for validation evidence. A hybrid keeps classical planning as a guard rail around the learned part, which is the direction the Electrek description suggests. Readers should expect the interesting disclosures to be validation methods and disengagement or incident data, not the model's benchmark scores.
What to check in the coming months
| Signal | Why it matters |
|---|---|
| Spanish and Madrid regulatory approval for the 2027 program | Gates the first step |
| Vehicle announcement for the Madrid cars | Shows whether the Lucid Midsize schedule holds |
| A formal order or contract between Bolt and Lucid | Turns a plan into a commitment |
| Safety validation disclosures for Alpamayo Stack in Europe | Determines trust with regulators and riders |
| Progress of Uber and Nuro's Lucid Gravity rollout | The nearer-term test of whether Lucid can deliver robotaxi volume |
| Bolt's Pony.ai and Stellantis pilot in Luxembourg | Shows whether Bolt runs multiple stacks in parallel |
What this means for builders and watchers
- Treat "robotaxis on Nvidia" as a roadmap, not a service. The first Madrid vehicles carry no riders.
- Watch the regulators. Madrid's region is described by Bolt's Jevgeni Kabanov as pioneering a structured framework for commercial autonomous mobility; approvals decide the timeline.
- Watch vehicle readiness. The Lucid Midsize platform's schedule affects which cars start collecting data.
- Watch for data and safety disclosures. A reasoning-based driver should publish how it is validated. Halos is Nvidia's safety layer; independent assessments would carry more weight.
- For AI practitioners: this is a large real-world deployment test for vision-language-action style driving models, which matters for the broader physical-AI field.
Open questions
- Which Lucid vehicles start in Madrid, and when exactly in 2027?
- How is Alpamayo Stack validated for European roads and rules?
- What commercial terms does Bolt have with Nvidia and Lucid, given reports that no order or payment has been made?
- Will Bolt keep its parallel Pony.ai and Stellantis track, and how will they interact?
Related reading
- Nvidia Alpamayo 2 Super
- Volvo, Nvidia Hyperion, Halos and Alpamayo
- Waymo 20,000 robotaxi fleet report: claimed vs verified
- Waymo expands to 14 US cities
- Uber and Wayve UK robotaxi launch
- GTC Berlin 2026 keynote recap
- Sources: Electrek on Lucid, Bolt and Alpamayo · Engadget on Nvidia and Bolt
Based on Electrek's October 21, 2026 report and earlier coverage; timelines depend on regulatory approval and may change.