


Tesla: The AI Narrative Continues to Surge (4Q24 Earnings Call Summary)
1. Business Performance Review
Core Information Review of Tesla's 4Q24 Financial Report:
Please refer to the article titled "Tearing Apart Tesla: The Grand AI Narrative 'Clashing' with the Collapsing Fundamentals of Car Manufacturing" for an analysis of the financial report.
1. Elon’s Remarks
In Q4 2024, we set new records with annualized deliveries exceeding 2 million vehicles, both production and deliveries reached new highs. The Model Y was a bestseller in 2024, not just in electric vehicles, but the top-selling vehicle globally. We have also achieved significant improvements in many aspects of FSD, and in 2024, we made many important investments in manufacturing, AI, and robotics technology to support the company's future development, and we will continue to move forward. We are one of the most valuable companies in the world, with Tesla's total value exceeding the combined value of the next five companies. Much of our success is attributed to autonomous driving and humanoid robots, and our focus is to continue developing in these directions. In 2024, we laid the foundation for future success, and in 2025, we will further strive to build product structure, expand product lines, and drive capacity ramp-up.
From 2025 to 2026-2028, we will achieve further epic leaps. Even now, people do not fully understand the value of our entire fleet and its monetization capability. Initially, people thought we were like toddlers learning to walk, but we will gradually run. People need to look at the essence of our technology and see what efforts we are making in technological advancement. Relying solely on linear intuition to perceive our exponential progress does not allow for complete understanding.
(1) Cars and Autonomous Driving
A car typically drives about 10 hours a week. If we can significantly increase the utilization rate to over 100 hours, the value of the car will multiply. If this car is autonomous, I estimate it could be in use for at least 1/3 of the week, or at least 50 to 55 hours. What will it be used for? Transporting people, delivering goods, and even when people are asleep, the vehicle can continue to operate, such as delivering packages at midnight or supplying restaurants. The utilization rate is greatly increased without a significant rise in costs. We are simply updating the software; people may look for past precedents, but I believe there has been no precedent like Tesla in the past. Vehicles can drive themselves, and our entire model operates well in the U.S., and over time it will be equally effective anywhere in the world, with efforts to increase production every year, and we can achieve good results this year.
Currently, our limiting factor is the battery pack. We have been working hard to resolve this constraint. The battery pack is where we hope to make better progress, and next year this constraint will further ease, and 2027-2028 will be fantastic.
The vehicle safety report for Q4 2024 shows that the safety of vehicles with FSD has improved year by year. If we compare the mileage between two FSD takeovers and the mileage between two human driving collisions, we can see that it is only a matter of time before vehicle safety surpasses human safety. FSD V13 and other incremental versions have made significant improvements, and the leap from V13 to V14 will be another major advancementWe launched the cortex training cluster at the Austin factory, which has made a further significant contribution to the progress of FSD from an AI perspective. We will continue to invest in infrastructure outside of Texas to support AI-related training.
(2) FSD Timeline
We are confident that we will launch fully autonomous FSD services in Austin as planned in June 2025. We have conducted further testing at our Fremont factory and will soon validate it at the Austin factory itself, with thousands of vehicles operating under FSD driving around our facilities every day, especially at the Fremont factory. In California, we will also see production and testing further advance.
These vehicles do not just operate on fixed routes; they have autonomy and can go to different points as needed. In terms of collision prevention, they are indeed very safe and will not collide just because there is a conflict in the points and paths they want to go.
FSD performs excellently in various states, from parking to starting, and in reversing. From the end of the production line, they can start and go to the corresponding positions, transporting components to various locations. We execute such operations thousands of times every day.
In the coming months, we will prepare to hope to implement everything without any risks. It's like having a basin of water; we want to take a bath and test the water temperature first by dipping our toes in. If it's suitable, we will then put our whole foot in. We have always prioritized public safety and hope to provide them with the best service.
Returning to FSD, in the future, people will look back at Tesla and say that 2025 was a special year for Tesla, when it launched truly effective real-world AI-based FSD, perhaps this milestone is more important than any single product launched before. Currently, no other company in the world can do real-world AI as well as Tesla, and this is the cornerstone of our further success. I don't even know which company will follow us in real-world AI.
(3) Humanoid Robots
Humanoid robots can play many different roles for various scenarios, and the training volume for so many scenarios is enormous. Perhaps their utility will be 1,000 times more than vehicles. We will not achieve this all at once but will do it gradually; it may start at 10 times and eventually reach 1,000 times.
However, this does not mean that the money we spend on training will be in such multiples, as we have been deploying better infrastructure for training, improving efficiency while matching more tasks for humanoid robots to complete, thus further realizing their value and spreading the costs.
Over time, training costs will certainly drop sharply, and this is indeed the case. The revenue from humanoid robots could potentially exceed $10 trillion, even with an investment of $500 billion, the training volume is still very high. The future of training may be drastically different from the past, as the entire infrastructure will improve further, and training efficiency can be significantly enhanced, which may bring a turning point in AI(4) Humanoid Robot Guidance
We are also making further revenue forecasts for humanoid robots. It all sounds very crazy, but I realize that our forecast numbers are actually achievable. Humanoid robots can provide value in different scenarios and tasks, and while the exact timeline may still be a bit vague, we hope to further advance the development of humanoid robots. With proper training, they can indeed accomplish many different tasks and functions.
According to our plan, we aim to build about 10,000 humanoid robots internally this year, achieving such a large number by the end of December. By the end of this year, several thousand humanoid robots can provide significant value in our factories, and we believe that once they are introduced to the public, they will definitely provide even greater value.
We are currently in the production design phase of the 1.0 humanoid robot, and we will launch the production phase 2.0 robot next year. Our goal is to make the production of humanoid robots faster than ever before, increasing the quantity by an order of magnitude each year.
In terms of humanoid robots, we are cautiously advancing this work at all costs. We will utilize any existing motors, actuators, sensors, etc., and we will continue to push this work forward at all costs. Humanoid robots must respect the basic principles of physics, and we will consider these principles in our design and production processes. So far, humanoid robots can perform complex hand tasks that have historically been challenging for robots; they can thread a needle with precision that is beyond human capability. My long-term prediction is that humanoid robots will become an overwhelming part of our company's overall value.
(5) Energy
In terms of energy, the importance of our entire energy business has been gradually increasing, and its significance will become even more apparent in the future. We can interconnect with the power grid and better coordinate in energy output, production, and usage. Before that, we hope to improve our performance in energy storage, as the vast majority of power grids currently lack storage capacity, making it necessary to design accordingly. Without storage, it is very difficult to operate. If we can excel in energy storage, the total energy of the power grid can potentially double. The second factory in Shanghai has already begun operations, and we are building a third factory.
We will gradually analyze what the demand looks like, what the processes are, and how we can respond to supply and demand challenges. We should have a significant output in GWh each year. We place great importance on fixed energy storage while balancing it with mobile energy storage. For the total gigawatt-hours of batteries, we will see that both fixed and mobile demand will grow rapidly over time. The year 2025 will be a year of significant development for our energy business. We must seize the opportunity to further enhance products like megapack and powerwall, ensuring that production meets demand.
2. Teneja's Speech
2.1 Automotive Business:a. Sales: In the fourth quarter of 2024, Tesla's vehicle deliveries reached a record high. The Model Y became the best-selling vehicle globally in 2024, not only as the best-selling electric vehicle but also as the sales champion among all types of vehicles.
b. Reasons for Margin Decline: The focus of the automotive business in the fourth quarter was to reduce inventory levels, achieving the lowest finished goods inventory level in the past two years through attractive financing options, discounts, and promotional activities, but these affected the average selling price (ASP).
However, the automotive profit margin declined quarter-over-quarter, primarily due to the decrease in average selling price (ASP) and the recognition of revenue related to Full Self-Driving (FSD) confirmed by feature releases in the third quarter. Nevertheless, due to the reduction in raw material costs, the overall cost per vehicle fell below $35,000.
2.2 Energy Storage Business:
In the fourth quarter, energy storage deployment reached a historical high, with growth mainly coming from Megapack and Powerwall products, both of which are currently constrained by supply. It is expected that capacity will be accelerated through the operation of the Shanghai Gigafactory, with a third factory also under construction.
Although quarterly deployment may continue to fluctuate, energy storage deployment is expected to grow by at least 50% year-over-year by 2025.
2.3 Services and Other Businesses
Gross profit and profit margins for services and other businesses increased year-over-year but declined quarter-over-quarter. This was mainly due to rising costs at service centers and declining profits from the used car business.
These businesses primarily support the new car business, particularly by influencing the total cost of ownership. While the company strives to maintain profitability on a GAAP basis, it does not expect these businesses to achieve the same profit margins as other operations.
2.4. 2025 Outlook - Product Planning:
Multiple new products will be launched in 2025, with the affordable Model 2.5 expected to debut in the first half of the year, continuing to enrich the product line. All products will be equipped with industry-leading software and autonomous driving features, with the potential for full self-driving in the future.
Additionally, all factories will begin production of the new Model Y next month, but this production transition will result in several weeks of downtime this quarter, leading to a loss in output and subsequently affecting profit margins. However, these issues will be resolved as production increases.
2.5 Financial Situation:
Reasons for Q4 Operating Expense Growth: Our operating expenses increased year-over-year and quarter-over-quarter in the fourth quarter of 2024, primarily due to further strengthening of R&D, as we will continue to invest in AI-related programs. The remaining growth came from our increased sales capabilities and further optimization of marketing efforts.
We expect operating expenses to increase in 2025 to support our growth cycle.
The net profit for the fourth quarter of 2024 was affected by fluctuations in the Bitcoin market, and we adopted new digital asset accounting standards, resulting in gains of approximately $600 million. We will conduct corresponding data statistics and will adjust the value of our investments in digital assets through each future reporting period
Capital Expenditure: Regarding Cortex, we hope to further strengthen the AI training infrastructure. We have also made more infrastructure investments in FSD version 13. For capital expenditure (capex) in 2025, we expect it to remain flat year-on-year.
The uncertainty of tariffs will still affect profit margins, although we have localized production as much as possible.
II. Q&A
Q: Is the unsupervised FSD still planned to be launched this year in Texas and California? What obstacles are there to achieving this goal?
We are confident that we will release the unsupervised FSD in California this year, and in fact, it is likely to be released in many areas of the United States by the end of this year.
The only reason currently hindering the release is excessive caution, as we need to ensure that its safety level is significantly higher than the average level of human drivers. After all, when autonomous vehicles have accidents, they become global headlines, while approximately 40,000 people die in regular car accidents in the U.S. each year, which mostly goes unnoticed. People can let the car drive autonomously in the city and observe how often they need to intervene for safety, experiencing the performance of unsupervised FSD. In reality, such interventions are rare and almost do not occur over time.
Q: Have you discussed FSD licensing with other car companies?
Currently, some major car companies have shown strong interest in licensing Tesla's full self-driving technology. The general response is that the best way to understand it is to disassemble a Tesla vehicle, which allows them to see the camera placements, the cooling requirements of Tesla's AI inference computer, and so on.
We will only consider situations with very large licensing volumes, and we will not let the engineering team get bogged down in tedious discussions with other engineering teams until unsupervised full self-driving is fully operational in the U.S. Once it becomes clear that there is no FSD, other manufacturers will have a high interest in licensing FSD.
Q: Is the design of Optimus basically confirmed for production in 2025?
A: The design of Optimus is not locked in; it may change when mass production occurs in 2025 or at another time. It is rapidly evolving and heading in a good direction. Trains appeared over 100 years ago or even earlier, but there were also design modifications.
Tesla has the best team of humanoid robot engineers in the world and possesses many of the elements needed to manufacture humanoid robots, such as excellent battery packs, power electronics, charging capabilities, communication capabilities, as well as real-world AI and large-scale production capabilities.
Many companies lack real-world AI and the capability to manufacture millions of products each year at scale. The industrialization of design and pure design are completely different concepts; making prototypes is easy, but achieving mass production is difficult.
Q: When will humanoid robots start being sold? What will the price be?
This year, Optimus is expected to be primarily used internally at Tesla, with thousands of robots performing the most boring, tedious, and dangerous tasks.
Initially, it should involve some simple use cases, such as loading materials into hoppers, transporting metal sheets on the body production line, welding body lines, etc. We just need to take out different materials and allow the robot to access them in the corresponding way, after which the robot can work quickly. Humanoid robots will not feel boredProduction version 2 may be launched next year, hopefully in early next year, but more likely in mid-next year. At that time, the production line design capacity will be 10,000 units instead of 1,000 units.
Q: What is the plan for the Optimus robot production line? When will deliveries start to companies outside of Tesla?
A: The currently designed production line produces about 1,000 Optimus robots per month, and the next production line will produce 10,000 per month, possibly starting in early next year or mid-next year. The following line will produce 100,000 per month.
It is roughly estimated that deliveries of Optimus robots to companies outside of Tesla may begin in the second half of next year. When stable production exceeds 1 million units per year, the production cost of Optimus will be below $20,000.
Compared to cars, Optimus has much lower total mass and complexity. When its annual production is comparable to that of the Model Y (which exceeds 1 million units), its cost is expected to be about half that of the Model Y. The price of Optimus will be determined by market demand.
Q: What is the status of Semi mass production? What is the impact of capacity ramp-up on business scale?
Last week, the topping and wall construction of the semi-truck factory in Reno was completed. The timeline for what comes next is still not very clear. The weather in Reno is unpredictable, and we will further carry out mechanical installations in the coming months.
The first manufacturing of high-volume design for Semi will begin at the end of 2025, with production expansion starting in early 2026. The total cost of ownership for Semi is much cheaper than any other mode of transportation, and large-scale production will make a meaningful contribution to Tesla's revenue, although the specific contribution amount is difficult to determine.
In the U.S., we lack truck drivers, which is also a constraint on transportation logistics. We have great respect for truck drivers, but not many people want to become truck drivers. We hope to address this issue with products like Semi.
Autonomous driving will also play a significant role in the trucking sector, with annual value creation expected to exceed several billion dollars. Initially, the volume may be small, and value creation will be at a relatively low level, but over time, I believe it should reach several billion or even $10 billion annually.
However, all of this pales in comparison to humanoid robots; $1 billion per month seems like a lot, but it may only reach 1% of Optimus's value.
Q: Does Tesla need to upgrade vehicles with HW3.0 for the future? If so, what is the timeline? What are the impacts on capex and costs?
We are still researching, and software versions are continuously being released; development of HW3.0 has not stopped. We have recently made a very good version, V12.6, which is a simplified version of V13 but has significant improvements compared to previous versions.
We are still trying to distill larger models into smaller ones. Therefore, we will not give up on software development for vehicles equipped with HW3.0; we are still working hard, but the release will lag behind the software version for vehicles equipped with HW4.0Q: Will you abandon the solar roof product?
The solar roof product remains a core part of the residential energy storage product lineup, attracting the attention of many customers. We have made many engineering iterations and have optimized installation and distribution significantly.
Recently, we have not focused on direct installations but rather on achieving growth through our network of certified installers, who have been collaborating with us for many years. This is a better approach that allows our products to benefit a larger audience.
Installing solar roofs makes a lot of sense when building new homes or replacing roofs. It is a high-end product that, when combined with the Tesla Powerwall battery, can achieve energy self-sufficiency for households.
Q: FSD V13 has achieved advancements in photon control levels and technical precision. What groundbreaking technologies can we expect in FSD V14?
We will go further than the photon control level and are making further improvements to neural networks, with some current enhancements looking very promising. We are surprised by the results achieved with aggressive transformers for solving different problems.
We will continue to expand in multiple areas, including scaling up the model (there is still room for growth beyond V13), increasing memory (currently limited, we will extend the duration of driving context), better handling of audio emergency situations, and incorporating tricky relevant data obtained from the entire fleet (such as user interventions). We will expand in various aspects, including computation, asset scale, model size, model context, and reinforcement learning objectives.
Q: Tesla's current stock price already includes many favorable factors mentioned today. Will these favorable factors drive further increases in company value? What measures will be taken to accelerate innovation to achieve the value described by the company?
We have been committed to real-world AI, achieving good results every week. We aim to fundamentally solve problems rather than just creating flashy surface-level solutions. Some engineering teams can make decisions independently, and our executive team highly respects their opinions.
Humanoid robots and FSD currently face many challenges, but we have very good solutions to address these challenges. At that time, we will also reflect on where we could have done better. Our team is continuously evolving, and we hope to improve through constant examination of our team's work. We believe the predictions made here are very accurate. Greater value will be created in the future.
Elon mentions a lot of very optimistic information in some contexts, and we believe it is not overly optimistic but based on solid information. We will not make predictions without any clear basis in the early stages. We believe it is possible to achieve fully unmanned FSD in Austin and several other cities in the U.S. by the end of this year. We will also expand further in North America, with regulatory constraints being the only remaining factor by 2026.
Europe has stricter regulations. The U.S. excels in innovation, while Europe excels in regulation. For example, although our unsupervised version of FSD has performed well, we have to submit a lot of documents to comply with regulations in the Netherlands. In May, we will submit the relevant documents to the European Commission, hoping for approval by then. Our FSD can be deployed in Europe, but with so many countries, many may require separate document submissions, which is quite challenging. It is uncertain when unsupervised FSD can be realized in EuropeIt may be next year or later. We need to understand when the relevant meetings of the European Union will be held and what their meeting rhythm is like.
China is a huge market, and we face some challenges. The Chinese regulatory authorities do not allow us to transfer training video data abroad, while the U.S. government does not permit us to conduct training within China, putting us in a dilemma. We hope to further consider how to respond to this issue.
Currently, we are training by using publicly available street videos from China on the internet and putting them into precise simulators to address this issue. This way, we can use publicly available videos of Chinese street signs and traffic rules for training.
The bus lanes in China are very complex, with limited access times each day. If you enter the bus lane at the wrong time, you will receive a fine, which is a significant challenge. Therefore, we must input the data into the simulator for training, allowing the car to read the signs.
From a technical standpoint this year, we believe we have the capability to launch an unsupervised FSD version in every market, only constrained by regulatory limitations. Our plan is to roll out unsupervised FSD in many cities in the U.S. this year, with a nationwide rollout expected next year. We hope that by the end of next year, most countries will have launched unsupervised FSD. This is our best prediction based on the existing data.
Q: Previously, you said that using LiDAR was very foolish, and that even if LiDAR were free, Tesla would not consider it. Do you still think this way?
Yes, when humans work, they do not emit lasers with their eyes unless you are Superman. Humans drive using their eyes, neural networks, and brains, which is a clear biological advantage. We hope to replicate this advantage on a digital level, allowing our cars to see, perceive, and analyze through cameras for visual processing, and then analyze the data through digital neural networks and AI. I think this is our fundamental principle. This is the cornerstone of our FSD, and we are firmly convinced of this technological route. LiDAR does not work well in foggy weather and is also difficult to maintain.
At SpaceX, we will also consider whether to use LiDAR or what kind of products to use. I am not specifically opposed to LiDAR; SpaceX has actually considered using products like LiDAR. We need to understand the principles of LiDAR to know its applicable scenarios.
The entire road system is designed for passive optical neural networks. Additionally, LiDAR has many issues; Tesla has designed and manufactured LiDAR, and practice has proven that it is not the correct solution for automotive autonomous driving.
Such products are very different from vehicles, and we hope to conduct further analysis. The results we are currently analyzing on vehicles indicate that the best value and experience can be achieved through the current visual route.
Q: Before Trump's term ends, artificial intelligence will rapidly enter the physical world. What progress can Elon achieve in AI through X company?
I think manufacturing is very cool, and I hope to better integrate AI with manufacturing. I hope that manufacturing can help us turn our electronics into significant realities. From a legal and financial perspective, we should attract more talent to engage in manufacturing, which could lead to better development for our countryTesla will ensure that it can continue to produce products even in the face of escalating geopolitical tensions.
Q: Will the deployment of robotaxis in Austin in June mean that I can go to Austin in June and have my car take me to different corners based on unsupervised FSD?
Our fleet is currently testing, and we hope to further review and identify any minor issues. We expect to start trial operations with our internal fleet before the end of the year.
As for the plan for Tesla owners' vehicles to join the fleet, similar to the Airbnb model, they can add their vehicles to the fleet and withdraw at any time. We need to wait until next year to launch this because we need to ensure that most of the issues we have resolved allow the entire fleet to operate efficiently.
Q: If I have a Tesla and FSD, do I need to keep my eyes on the road all the time? This is very boring. If it is truly unsupervised, does that mean I really don't need to look at the road anymore, and I can check my emails or do anything else? If I need to take over, will I be notified 5 seconds in advance?
This is indeed the direction of future development. We hope that unsupervised FSD will allow everyone to do anything they want in the vehicle. We need to be very confident about the future; the likelihood of our vehicles having any accidents is very low. You can do anything on the road, like looking at your phone, texting, or writing emails.
Before allowing owners to check emails and texts, we need to ensure that the probability of injury is very low. Currently, the car's autonomous driving capabilities have not fully met the requirements. Although we are not far from achieving this, we still need to prove to ourselves and regulatory agencies that the car is absolutely safer in autonomous driving mode than in non-autonomous driving mode.
In fact, in traditional vehicles, there were people who texted while driving, which is unsafe. However, with FSD, it becomes safer. In a few months, our unsupervised FSD will gradually roll out on a large scale. Sometimes people want to use it but end up turning off autopilot. This is because some safety features may occasionally send reminders, and if they want to turn it off, they certainly have that option. We actually hope that these features can always provide safety information to help people reach their destinations safely.
On the regulatory side, as long as our vehicles are safe enough, regulators will definitely support us, even in fully autonomous driving mode. We released a vehicle safety report today, showing that there is one accident for every 5.9 million miles driven, while human driving has one accident for every 700,000 miles.
Q: You mentioned the necessity of popularizing sustainable transportation in the past? We heard that President Trump plans to revoke mandatory policies related to electric vehicles, and many people have different opinions. What do the government and regulators think about the popularization of electric vehicles? If President Trump really goes ahead with plans to revoke mandatory policies related to electric vehicles, how might this slow down the popularization of electric vehicles in the U.S., and what are your comments on this?
Sustainable transportation is an inevitable trend and will definitely be realized. The future of transportation should be automated, electrified, and sustainable. In the future, even airplanes could be electricNo one can stop this trend. It started with the steam engine, then the internal combustion engine replaced the steam engine, and now the transition from internal combustion engines to electrification is an inevitable trend.
Q: Regarding the issue of automotive profits, FSD is advancing rapidly. What impact will the rapid advancement of FSD have on sales? Do you believe FSD will help you launch more vehicles in the market? Currently, there are millions of vehicles; will this be linked to the future growth of the fleet and the growth of FSD? Will you also actively sell electric vehicles?
Currently, Tesla is trying to address the limiting factor of battery production, rather than demand. For example, in the first quarter of this year, the factory underwent a large-scale equipment overhaul for the new Model Y, which affected short-term production. The key issue now is to find ways to increase the total gigawatt-hours of battery production this year, as this is the limiting factor for output