Tesla today announced the company's financial report for the first quarter of fiscal year 2024: total revenue for the first quarter was US$21.301 billion, down 9% year on year, the biggest drop since 2012; net profit was US$1,144 billion, a sharp drop compared to net profit of US$2,539 billion in the same period last year; and net profit attributable to common shareholders was US$1,129 million, a sharp drop of 55% year on year.

Tesla's first-quarter revenue exceeded Wall Street analysts' previous expectations, but the adjusted diluted earnings per share fell short of analysts' expectations. Meanwhile, Tesla Tesla CEO Elon Musk said that the company plans to “start production of new models in early 2025, possibly later this year,” while production was previously expected to begin in the second half of 2025. After the earnings report was released, Tesla's stock price rose sharply by more than 10% after the market.

After the financial report was released, executives such as Tesla CEO Elon Musk, CFO Waibaf Tania, Drew Baglino, Senior Vice President of Power and Energy Engineering, Kahn Budilaj, Vice President of Supply Chain, Russ Moravi, Vice President of Automotive Engineering, and Martin Wicha, Head of Investor Relations, read the financial reports and answered questions from analysts.

The following are the main topics of this conference call with analysts in this question-and-answer session:

Individual investors: How are 4680 batteries currently progressing? What is the current production capacity?

Russ Moravi: Tesla 4680 battery production has increased 18% to 20% since the fourth quarter. We also shared it with you on X before. Currently, the weekly output of 4,680 batteries can be equipped with more than 1,000 Cybertrucks (Tesla pickups), and the annual production capacity is about 7 gigawatt-hours. With the third and fourth production lines of the first phase of the plant being put into operation, we expect that the battery production capacity in the second quarter will continue to grow faster than Cybertruck's production capacity, ensuring that we always have several weeks of battery inventory to cope with the subsequent rise in vehicle production capacity.

As production capacity climbs and production line output increases, our cost of goods sold (COGs) is falling rapidly every week, and we hope to offset supplier costs for nickel-based batteries.

Individual investors: The next question concerns Tesla's humanoid robot Optimus. What's new with Optimus? Is it already in use in the factory? When does management expect Optimus to be mass-produced?

Elon Musk: Currently, Optimus has been able to perform some simple factory tasks, or I should say that judging from current laboratory progress, Optimus is already able to perform simple factory work. We think Optimus will probably be launched in the factory by the end of this year and officially put into use within the factory. As for the timing of external sales, I expect it to be around the end of next year.

But these are just predictions. I've shared with you before that, in my opinion, Optimus is worth more than all other technologies combined. Take sentient, sentient humanoid robots, which may not be able to perform tasks exactly as required. Not so with Optimus. I believe in the field of humanoid robots, Tesla is more suitable for mass production than other peers — our robot's reasoning ability is very efficient. This is worth everyone's attention. Our AI reasoning ability and reasoning efficiency are far superior to those of other companies. It can be said that no company's robot's reasoning efficiency is comparable to Tesla. This (improving reasoning efficiency and ability) is also something we have no choice but to do; after all, we are limited by inference hardware in train production.

Institutional investors: What are Tesla's internal expectations for regulatory approval of unregulated FSD (fully autonomous driving) in the US? Is there an appropriate safety threshold for unregulated FSD compared to human drivers?

Drew Baglino: Currently in the US, several states have passed regulations related to autonomous driving. Arguably, these states are paving the way for the future development of autonomous driving technology, and related operational data will also encourage more and more people to choose driverless cars. I won't go into detail on safety performance issues here, but I believe that the driverless driving experience in these states, along with the relevant data we can provide, will pave the way for regulatory approval for fully autonomous driving in the US, while also building blocks for relevant technology approvals in other countries.

Elon Musk: In fact, some autonomous car companies are already trying to expand their regulatory horizons, which is very helpful to us. At present, these companies have been operating in San Francisco for some time, and it seems that they have also been approved to operate in Los Angeles. It can be said that government regulations and approvals are extremely fast, and changes are changing rapidly.

In my opinion, as the scale increases and the amount of data is accumulated, if the final results show that the accident rate of autonomous vehicles is half that of human drivers, this will not be underestimated, but it is a problem that cannot be ignored. Because if autonomous driving is banned at that time, it will mean disguised murder. Therefore, as long as there is definitive data to prove that autonomous cars are safer than human-driven cars, I believe there won't be much of a barrier in terms of regulation.

For example, fully autonomous driving technology is like elevators back then. In the past, elevators were operated by humans to switch elevators. However, people get tired, get drunk, and make mistakes, which may cause elevator passengers to get stuck in the middle of the floor. Today, we only need to enter the elevator and press the button ourselves; you don't need to think about how the elevator actually works. The same is how autonomous cars work. Tell the car your destination via your phone, and it will automatically take you to your destination once you get on the bus. It all comes as it should, just like an elevator — it automatically takes you to the floor you want to go to. It's that simple; you don't need to think about how elevators work at all.

Also, I need to be clear. In the future, Tesla will launch and operate its own fleet. You can think of us as Airbnb (Airbnb) or Uber (Uber) in a car company. In other words, Tesla will have a certain number of self-operated cars, and some cars in the fleet will come from end users, and users can join or leave the fleet at any time. They can choose whether their car will only be used by friends and family, or only for five-star users. Furthermore, their cars can be withdrawn from the fleet at any time and are for the user's own use only. Just like Airbnb. We believe that as the fleet continues to grow, we may have 7 million, 9 million, or even tens of millions of vehicles worldwide in the future. As the number of fleets increases, the data will also achieve continuous feedback and continuous circulation, and every time a problem occurs, it will be added to the training database, thus achieving the “flywheel effect” of the training data. However, Tesla's training data “flywheel” is actually very similar to a Google search. You need to know that it is very difficult to compete with Google, because users' searches and clicks are uninterrupted, and there is also a steady stream of data feedback from Google, so Google's data forms a closed loop of feedback. The same goes for Tesla. The scale of our data can reach tens of millions.

Also, I think Tesla will have considerable potential for future development in the direction of AWS (Amazon Cloud), which will help us achieve more powerful inference and calculation capabilities. Tesla's hardware has evolved from the previous Hardware 3.0 to today's Hardware 4.0. The Hardware 5.0 design has basically been completed, and it is expected to be applied to cars by the end of next year. We imagine that if the vehicles in the fleet don't move for a while, we can use it for distributed inference, just like AWS. In other words, you can imagine that in the future, if the Tesla fleet reaches 100 million units, their reasoning power will reach 1 kilowatt on average, which means that globally we have 100 gigawatts of deductive computing power. Integrating this huge amount of computing power isn't easy, but autonomous cars don't work all the time. It probably works 10 hours a day, and only needs to work 50 hours a week; the rest of the time can be used for reasoning, otherwise wouldn't it be a waste?

Vaibaf Tania: A question about safety.

Almost every week, Tesla uses our neural network system to generate different trajectories to drive cars. Through millions of user data and our internal training data, we continuously train for some key accidents and accidents, such as when someone suddenly jumps in front of a car, etc. Additionally, our training database includes real-world operational data we've collected over the years. We will continue to repeatedly train and test these elements in the simulation system to ensure that vehicle safety is truly improved. In addition, we also conduct data training in different cities, such as San Francisco, Los Angeles, Austin, New York, etc., to ensure the diversity of training quality. At the same time, we also collect key incidents or some major incidents and integrate this content into our training library. We pay a lot of attention to training performance and often ask ourselves if our performance has improved compared to previous weeks?

Only when we have confidence in our internal data training network will we begin delivering products to advanced users (such as to Tesla's 2,000 employees). These users will provide us with feedback after using the product, such as where improvements are needed? Or whether we discovered problems during use that we hadn't seen before, etc. And only after these processes are foolproof will we promote the product to external users. Furthermore, even if the product is finally launched to market, Tesla will have a real-time dashboard to monitor product operation to ensure that no key issues are missed.

Along the way, it can be said that Tesla has always attached great importance to the quality and safety of its products and hopes for continuous improvement. We will continuously obtain new training data at every stage, add it to the next data training cycle, and continuously improve the product model. It's a process of continuous feedback, circulation, and evaluation. In our new architecture, previous issues will continue to be improved. In a sense, our engineers don't necessarily need creativity, nor do they have to constantly improve the way they write algorithms; what is important is that they need to constantly train and learn from data. Therefore, if we face a problem at some point, such as how to drive through an intersection, the relevant training data will be immediately fed back to the training library, and the technology will be continuously optimized through repeated automatic training and automatic learning.

Elon Musk: We are very confident about our products in the next three to four months. The reason for saying this is because we currently have models that are more powerful and better than at this stage, but there are still some problems that need to be solved. I believe the new model will bring dramatic changes to the car's features, but we will not release it until the specific issues have been resolved.

We take great care every time we launch a product to our users. The upcoming FSD V12.4, V12.5, or even V13 versions, can be said to be disrupting neural network systems, and we have made many improvements and changes. Therefore, I believe that once these technologies are applied to vehicles, they will greatly improve product performance. Please be patient for 3 to 4 months.

Vaibaf Tania: Within Tesla, we will continue to increase model capacity and improve product performance accordingly. In addition, we will continue to expand training data and increase training time to improve product performance. Furthermore, improvements in product architecture, changes in model size, and changes in data training combinations will all bring improvements to product performance, and we are very optimistic about this. But everyone should know that whether it's experiments or training, it takes time. Collecting data, training data, processing data, and even watching tens of millions of videos over and over takes time. However, from Tesla's development trends in the past, we can trust that our future will not disappoint anyone.

(Continuously updated...)