When will Optimus 3 be revealed, when will production start, and what production rate is expected exiting the year?
The Optimus 3 unveil will likely be pushed closer to production because competitors do frame-by-frame analysis and copy everything shown. Start of production is assumed around late July or August, but the production rate this year is impossible to predict and will be quite slow at first.
Elon Musk explained that Optimus is a completely new product on a completely new production line with more than 10,000 unique items, and the ramp will move as fast as the slowest part. Initial skills will start simple, in the factory, and build up from there.
The V3 Optimus design is almost ready to demonstrate, with functional elements working but aesthetic elements still to be finalized, and a reveal is expected around the middle of the year. Tesla is also hesitant to show V3 early given competitor copying.
What milestones are targeted for unsupervised FSD and robotaxi expansion beyond Austin this year, and when will it drive recurring revenue?
Tesla hopes to have unsupervised FSD or robotaxi operating in roughly a dozen states by the end of this year, taking a cautious rollout approach. Robotaxi revenue will not be super material this year but is expected to become material, possibly in a significant way, next year.
The company reports no injuries or fatalities to date from unsupervised FSD and robotaxi expansion, and safety validation remains the deliberate limiting factor on expansion.
When will unsupervised FSD reach customer cars?
The estimate is probably the fourth quarter. Release will be gradual by geography, expanding as each area is confirmed safe.
Releasing everywhere at once is difficult because some cities contain uniquely complex situations, such as unsafe intersections, bad road markings, or heavy weather challenges, where accidents tend to happen.
How will Hardware 3 cars reach unsupervised FSD?
Hardware 3 does not have the capability to achieve unsupervised FSD, since it has only one-eighth of Hardware 4's memory bandwidth, which is a choke point for AI inference. Tesla is offering FSD buyers a discounted trade-in toward AI4-equipped cars and a computer-and-camera upgrade to Hardware 4.
Doing the upgrade efficiently requires small micro factories in major metropolitan areas rather than service centers, and over time converting all Hardware 3 cars to Hardware 4 makes sense because that enables entry into the robotaxi fleet.
In the meantime, a distilled V14 version for Hardware 3 with the same features as the Hardware 4 V14 release, including start from park state, is expected by end of June.
Why did the AI5 tape-out finish early, and has AI5 been dropped from the vehicle roadmap?
The tape-out finished early because the team worked every weekend for six months straight, including holidays, without major mistakes. AI5 is not dropped from the roadmap; it will go into Optimus and data centers first, since AI4 is expected to achieve unsupervised driving far above human safety levels.
Switching to AI5 in cars is not pressing, but an AI4 upgrade to newer generation RAM is planned, going from 16 GB to roughly 32 GB per SOC (64 GB total) with about a 10% compute increase, entering production around the middle of next year depending on Samsung's modifications.
Can you provide details on the Terrafab project, which party funds, builds, and operates each aspect, and what is Intel's role?
Tesla will build the roughly $3 billion research fab on the Giga Texas campus, capable of a few thousand wafers per month for testing new chip-making ideas, while SpaceX takes care of the initial phase of the scaled-up Terrafab. Intel is partnering on core manufacturing technologies, with Tesla planning to use Intel's state-of-the-art 14A process.
Details of the Terrafab deployment are still being worked out; any intracompany arrangement requires approval by both the Tesla and SpaceX boards through conflict-resolution and independent-director reviews.
The research fab plans to do memory, logic, packaging, and lithography mask creation under one roof, which management believes is unique in the world and enables the fastest recursive R&D loop, with some long-shot ideas that could radically improve chips if they pan out.
Is much of Terrafab motivated by getting better economics on midterm chip purchases from suppliers?
No. Terrafab is not a mechanism for leverage over chip suppliers; it exists because Tesla sees no path to a sufficient quantity of AI chips as production scales, anticipating hitting a wall in industry logic and especially memory supply.
Tesla has research ideas for radically better AI chips that are long shots but could yield giant improvements, which are easier to pursue with its own research lab and production technologies, and existing industry capacity could not keep up with concepts like AI satellites.
What is Tesla's robotaxi strategy for Europe now that FSD has been approved in the Netherlands?
Discussing robotaxi in Europe is jumping the gun, since approval of even supervised self-driving took an immense amount of time and the timeline for unsupervised approval is at the mercy of regulators. Supervised FSD goes to Brussels for EU review in May, and the same architecture and training used in the US should transfer once local data is added.
The Chief Financial Officer noted FSD approvals in the Netherlands set up EU-wide approval later in Q2, with additional approvals received in China and broader Chinese approval hoped for by Q3.
How does Tesla expect to resolve the battery pack capacity constraint limiting vehicle production growth?
The limiter is battery pack capacity rather than cells, and Tesla is actively adding capacity. Berlin's Model Y pack with in-house 4680 cells is ramping nicely, the Reno facility is being retooled with more efficient lines, and in-house LFP module and pack production is ramping in China.
Giga Berlin reached a record output of over 61,000 units in Q1, and the company plans to keep growing volumes across all factories, with battery pack capacity named the biggest limiter in the opening remarks.
What key safety metrics support robotaxi expansion, and what actually limits wider deployment?
Tesla tracks miles per intervention, miles per accident, and per-fatality metrics using a large QA fleet across the United States plus simulator analysis, and all expansions so far have gone according to expectations. The biggest limits on wider deployment are not safety issues but convenience problems, the car being over-cautious, getting stuck, or looping infinitely.
The NHTSA incident filings show zero incidents to date, and Tesla is scaling its QA fleet across the US while monitoring the customer fleet, which is approaching 10 billion FSD-driven miles, to validate safety and solve the long tail of scaling issues.