The new rumor has started again, can Tesla perform well this time?

Tesla is about to hold a Robotaxi launch event on October 10th, which is crucial and one of the important factors driving the recent continuous rise in stock price. However, whether Robotaxi can be Tesla's "turning point" or "failure" remains to be seen. The main focus and expectations of the market are expected to be on the following key points:

1) Is Robotaxi ready? Who can be the first to successfully implement the Robotaxi business model?

2) What kind of business and operation model will Tesla's Robotaxi adopt?

3) How is Tesla's next-generation Model 2 going to be? Can it achieve significant cost reductions?

I. Most important: Is Robotaxi ready? Who can be the first to successfully implement the Robotaxi business model?

Answering this question, both the technical and regulatory aspects are indispensable for Robotaxi. However, regulations are primarily due to safety concerns - the immaturity of technology, and technological advancements will further push regulatory relaxation. Therefore, it ultimately still boils down to a technological issue.

On the technical side, Robotaxi is still based on the hardware and software of FSD. It is well known that Tesla has always adopted a pure vision approach, and it is expected that Robotaxi will continue to adhere to this approach at this event.

However, L4 level of autonomous driving is a prerequisite for Robotaxi to provide services. Progressive route vehicle companies led by Tesla (transitioning from L1 to L4 gradually) are mostly shifting towards large-scale models + pure vision solutions (including WeRide, etc.). But the current autonomous driving services provided are still at the L2+ level. The autonomous driving providers that have already been implemented are taking a leap route (starting directly from L4) and all adopting a multi-sensor approach. The leap route seems to have a faster pace of implementation at the technological level compared to the progressive route.

Dolphin believes that the key factor in the competition of Robotaxi lies not in the initial operational scale, but more importantly, in which route can eventually become the mainstream route. This is also the core factor for Tesla to successfully establish a dominant position among the various players in the Robotaxi industry.

To determine the ultimate winner between these two routes, two aspects need to be considered:

1）Cost side: The current players in Robotaxi, whether domestic like Pony.ai, AutoX, or overseas like Waymo, Cruise, all adopt a multi-sensor approach. Although the implementation speed is faster, the vehicle cost remains high, which is also a core factor leading to significant losses in the UE model and the inability of current autonomous driving providers to seize the market through large-scale deployment.

Tesla, taking the pure vision route, relies mainly on cameras, supplemented by some low-cost sensors. The embedded cost of vehicle hardware is generally very low. Whether it is a prototype vehicle specially designed by Tesla for Robotaxi (possibly eliminating the steering wheel/pedals) to further reduce costs, or existing Tesla models equipped with FSD joining the Robotaxi fleet (Model 3/Y/S/X) , currently has an advantage in vehicle cost compared to the multi-sensor solution, and players using the UE model will run faster than those using the multi-sensor solution. (Dolphin Jun expects that Tesla's Robotaxi Day may announce the cost of the Robotaxi prototype and the UE model of the Robotaxi).

The three methods for reducing vehicle costs for unmanned driving providers using the multi-sensor solution are:

1. From retrofitting to original equipment manufacturing (OEM): Currently, most unmanned driving vehicles are purchased from OEMs and then retrofitted to be suitable for Robotaxi, resulting in high costs. Unmanned driving manufacturers have realized this issue and are collaborating with OEMs to develop unmanned driving models. For example, Baidu is about to launch RT6, Waymo and ZEEKR's collaboration M-vision are both original equipment manufacturing models.

Original equipment manufacturing models eliminate driver-related configurations such as dashboards, central screens, front row interiors, etc., while avoiding additional design and manufacturing costs for retrofitting existing vehicles, thereby reducing costs.

2. Improve algorithms based on collecting more data to reduce the number of sensors: Although unmanned driving providers cannot deploy vehicles on a large scale due to cost reasons, they have collected a large amount of data through road testing and commercial operations in the early stage (although they are far behind in data collection compared to progressive routes), promoting the improvement of software algorithm capabilities, thereby reducing the number of sensors. (For example, reducing the number of LiDARs, LiDARs transitioning from mechanical LiDARs to semi-solid-state LiDARs, and RT6 reducing the cost of the expensive Velodyne LiDAR in RT5 to the lower-cost Hesai Technology AT128-4).

3. Cost reduction through the sensor supply chain: With the popularization of domestic NOA intelligent driving, the increase in sensor usage promotes overall cost reduction in the supply chain.

From the current results, according to the latest cost of the RT6 vehicle published by Radish Express, the cost of the RT6 vehicle is only 205,000 yuan (excluding the intelligent driving kit). Assuming that the vehicle cost is around 250,000 yuan including the intelligent driving kit, the cost has decreased by 48% compared to the cost of 480,000 yuan for RT5. It is expected to start large-scale operations in 2025, which indirectly proves that the multi-sensor solution can also achieve a significant cost reduction, thereby running the UE model.

In the case where both solutions are expected to be feasible in terms of cost, although Tesla has a time advantage in running the UE model in terms of cost, the commercialization and landing speed on the technical side may be slower due to the adoption of a pure visual route. The fundamental core competitive factor still lies in the technological competition between the two routes: 1) Under the basic requirements of safety in commercial operations, who can land faster?; 2) Whose technological ceiling is higher - ensuring higher safety? Thus achieving true Level 4-5 autonomous driving?

2) Technical side: In addition to cost considerations, safety is still the most critical factor in whether the Robotaxi business model can be successful, which is related to whether policies can truly open up the operation of Robotaxi - affecting the supply side, and whether consumers can truly accept Robotaxi and be willing to ride in it - affecting the demand side. Safety depends on the progress and iteration of autonomous driving technology Robotaxi in the long term, due to saving driver costs and increasing operating hours (24 hours), will have a cost advantage compared to ride-hailing/taxis. However, the key point for Robotaxi to replace ride-hailing/taxis should first be on 1) surpassing human drivers in terms of autonomous driving safety; 2) followed by lower pricing based on higher safety levels.

The key indicator for measuring safety is MPI: how many miles an autonomous vehicle can travel before human intervention is needed. Generally, the longer the MPI, the higher the reliability of the autonomous driving system. This is also the most direct indicator to measure how far Tesla is from truly being ready for Robotaxi and commercial operation.

Elon Musk has previously stated that Tesla's autonomous driving computing power bottleneck has been resolved (the expansion of the Texas Gigafactory is nearing completion, which will accommodate Tesla's largest H100 cluster to date, with computing power expected to increase significantly by the end of the year). However, the current bottleneck for FSD lies in the testing and training methods, still limited by too many interventions in the mileage.

Currently, Tesla has not disclosed FSD intervention data, only revealing that FSD V12.5.2 was released in September, with a 3x increase in necessary intervention mileage, and the V13 version released in October with a 6x increase. According to third-party AMCI Testing, FSD requires human intervention every 21 kilometers on average, while Waymo only needs to take over once every 27,900 kilometers. It can be seen that Tesla is still far from commercializing Robotaxi operations. (Dolphin will focus on whether Tesla will disclose the actual takeover mileage on Robotaxi day)

The significant gap between Tesla and Waymo/Cruise is partly due to the fact that the pure vision-based end-to-end architecture is not mature enough. Transitioning from a multi-module architecture to an end-to-end architecture will initially experience a certain technological regression. Another reason is that Robotaxi's current road testing, passenger operations, and commercial pilot projects are limited to specific areas. High-precision maps can provide prior experience for vehicle operation, while Tesla's autonomous driving is map-free.

However, for Robotaxi to operate commercially in the long term, the region needs to be expanded to more public roads and cities, limiting the effectiveness of high-precision maps. Tesla's vision-based solution does not rely on high-precision maps. After the vision-based solution matures, the gap with multi-sensor fusion solutions may gradually become significant. The battle for routes will also become a key factor in whether Tesla can establish itself in the Robotaxi war. Therefore, Dolphin's focus remains on the progress of FSD In terms of the upper limit of technology from two different routes, the visual solution has higher accuracy because the essence of sensors is a bit stream, and the information rate of camera bits/second is several orders of magnitude higher than that of lidar and mmWave radar.

With the increase in training data scale and algorithm optimization, the performance ceiling of the visual solution may be higher compared to the multi-sensor solution.

As mentioned by Dolphin in the article " FSD Smart Driving: Can't Support Tesla's Next Valuation Miracle", Tesla has an absolute leading advantage in the field of pure visual solutions. It has formed a closed loop of data + computing power center + self-developed chips and algorithms, all of which are self-owned, self-built, and self-researched, truly achieving full-stack self-research.

Leading in data: Second only to BYD in stock vehicles, with a high accumulation of mileage.

Leading in vehicle-side hardware + software-side algorithms: FSD V12 end-to-end algorithms have been leading domestic players for nearly 2 years; Hardware HW 5.0 will be launched in December 2025, with performance 10 times that of HW 4.0, about 5000 TOPS+.

Leading in training computing power: Compared to peers, Tesla is the only company with a self-built supercomputing center in the backend, and has self-developed the D1 supercomputing chip and launched the DOJO supercomputer. While peers mostly rely on cloud computing service providers for computing power.

Once the route dispute is finally settled, if the pure visual route with strong generalization ability, fast technological progress, and higher technological ceiling can bring higher safety, and eventually become the mainstream trend, the core factor for Tesla to successfully establish itself as the king among the players in the Robotaxi field.

II. Secondly, focus on: What kind of business and operation model will Tesla's Robotaxi adopt?

From the perspective of the current landed autonomous driving providers, they mostly adopt a heavy asset operation model: Autonomous driving providers cooperate with car manufacturers to develop vehicle models/or purchase vehicles from car manufacturers for later modification and self-ownership of these vehicles for operation.

In previous earnings calls, Tesla also stated: In addition to building Robotaxi vehicles on its own, Tesla's Robotaxi will combine the operating model of Airbnb + Uber, which means that once the FSD technology matures, existing Tesla stock vehicles can join the Tesla Robotaxi fleet. Tesla's role in Robotaxi will shift towards a platform operator similar to Uber, charging a commission (possibly between 20% and 30%) to Tesla vehicles joining the Robotaxi fleet. Overall, it is expected that Tesla will initially adopt a heavy asset layout in the Robotaxi business model, and then extend to a light asset operation model as the FSD technology matures The core issue lies in:

1) Can all existing Tesla models be included in Robotaxi?

Tesla continues to iterate in autonomous driving, with software updates possible through remote OTA, but hardware updates cannot be made later. Currently, FSD has iterated to HW4.0, with the biggest improvement being the chip computing power, increasing from HW3.0 to HW4.0 by about 500 Tops. HW5.0 is also set to be released in 2025, with computing power expected to increase tenfold compared to HW4.0, reaching 5000 Tops. For Robotaxi based on L4 autonomous driving, the hardware requirements, especially chip computing power, may be higher. The key question Dolphin is concerned about is whether all existing Tesla models can be included in Robotaxi, or at least models based on the HW4.0 hardware version?

2) Will Tesla build its own platform similar to Uber or choose to cooperate with Uber and other ride-hailing platforms, or both?

Dolphin predicts that Tesla will create its own ride-hailing platform similar to Uber. Previously, Tesla has already demonstrated a model of its Robotaxi hailing app. Dolphin believes that the barrier to entry for a ride-hailing platform is not high, and the differentiation of the platform competition lies more in the ecosystem and traffic.

Although autonomous driving companies have built their own ride-hailing platforms (such as Robo-Taxi, Waymo, etc.), Waymo also cooperates with Uber. On one hand, due to the limited deployment of Waymo's autonomous vehicles, the advantage of the ride-hailing platform lies in having a sufficient density of operational vehicles on the ride-hailing platform. Therefore, in the initial stage of cold start, Didi can reduce the capital requirements for cold start by using a mixed dispatch model, even if only a small number of autonomous vehicles are deployed, without affecting passengers' hailing experience, thus reducing the need for capital for cold start. On the other hand, the ride-hailing platform has better advanced matching algorithms based on years of operational experience and data.

Will Tesla face the same issues when building its own Robotaxi platform? Dolphin believes that models built on heavy asset models will face limitations and similar issues. However, once the light asset model is activated, whether all existing models can join the Robotaxi fleet will become a key factor, but all these issues are premised on whether the pure visual route can be viable.

III. What is the progress of Tesla Model 2's launch? Can it achieve a significant cost reduction?

Amid significant uncertainties in the current Tesla Robotaxi business model, Dolphin believes that Tesla's biggest short- to medium-term growth point lies in the launch of the next-generation model, Model 2, which is also the most crucial factor that investors viewing Tesla as an automotive company are concerned about The key question remains whether Tesla's Model 2 can reduce costs as expected and lower the price to $25,000. When will the Model 2 be launched and become the next growth point for Tesla's sluggish car sales? Will Tesla's Model 2 be launched in the United States first and then gradually expand to the European/Chinese markets? Especially in the Chinese market, the 100,000-200,000 RMB low-end electric vehicle market has been firmly dominated by BYD. What are the core differentiation advantages of Tesla's Model 2 expanding into the Chinese market?