Expected annual capital expenditures to exceed USD 10 billion, will continue to strengthen investment in AI chips (Tesla Q2 earnings call minutes)

The following is a summary of Tesla's 2024 second quarter earnings conference call. For an interpretation of the financial report, please refer to " Tesla: "AI Pie" sounds good, but reality is too tough "

I. Review of Core Financial Information:

II. Detailed Content of the Earnings Conference Call

2.1. Key Points from Management's Statements:

1) Operational Highlights:

① Overall Business Situation

a. The adoption rate of electric vehicles has significantly increased, but facing intense market competition, other manufacturers offering substantial discounts pose challenges to Tesla

② Autonomous Driving and Robotaxi

a. FSD has made significant progress, launching version 12.5, with 5 times the parameters of 12.4, combining highway and city driving stacks

b. Demonstrating system usage methods, expected to be a significant demand driver, transitioning from supervised full self-driving to unsupervised full self-driving

c. The Robotaxi product launch event has been postponed to October 10th for important modifications and improvements to the main showcase content

d. Expansion of the Giga Texas factory is nearing completion, the supercomputing center will house the largest training cluster, adding 50,000 Nvidia H100 AI chips, 20,000 FSD computing hardware AI5

③ Optimus Robot

a. The Optimus Robot has been performing tasks in the factory, expected to start production early next year and limited production

b. Several thousand Optimus Robots will be produced by the end of next year and perform useful tasks, with a significant increase in production by 2026, provided to external customers

④ Energy Business

a. Energy storage business is growing fastest, mainly due to demand constraints rather than production constraints

b. Expanding production in the United States and building a super factory in China, significantly increasing capacity, possibly doubling or tripling

2) Financial Highlights:

① Revenue and Profit

Second-quarter energy storage deployment and energy business profits hit a record high, vehicle delivery volume increased, regulatory credit revenue hit a record, and automotive gross margin remained stable

② Costs and Expenses

a. Providing low-interest rate car purchase schemes to counter high rates, impacting second-quarter car sales revenue, with the impact continuing into the third quarter

b. Excluding the impact of Cybertruck ramp-up, vehicle costs decreased compared to the previous quarter, while some material costs decreased with a lagging reflection

c. In the second half of the year, Cybertruck and Model 3 are still affected by tariffs on raw materials (batteries) and finished products (cars), and short-term costs may rise.

③ Capital Expenditure and Free Cash Flow

a. Capital expenditure decreased in the second quarter, but is expected to exceed $10 billion for the full year due to increased investment in an additional 50,000 Nvidia H100 AI chips.

b. Free cash flow returned to positive in the second quarter, reaching $1.3 billion, with cash exceeding $30 billion at the end of the quarter.

3) Market and Customer Strategy

Competitive financing rates are offered, and after the price reduction of FSD in North America, utilization has significantly increased. Models with a range of over 300 miles are also provided, expanding the supercharging network to address range anxiety.

Carbon credit revenue is affected by other OEM manufacturing and sales plans as well as regulatory changes.

4) Other Businesses

Energy Business: Energy storage shipments have more than doubled, contributing not only from Megapack but also from Powerwall, bringing historical highs in revenue and profit to the energy business.

There is a large backlog of energy storage orders, with shipments fluctuating at different times, significantly increasing in some quarters while decreasing in others. Over time, the recognition of storage revenue depends on various factors, including logistics time to send equipment from a factory to the global market, customer readiness, and the status of EPC projects or construction activities.

Service and Other Businesses: Gross margin improvement on a sequential basis, growth in service utilization and collision repair business. Excluding one-time restructuring costs of $622 million, operating expenses decreased sequentially, despite increases in AI-related expenses and legal fees.

2.2. Q&A Analyst Q&A

Q: What is the latest situation with Roadster?

A: Regarding Roadster, we have completed most of the engineering work. We still want to make some upgrades to it, but we expect production to start next year. It will be a special product.

Q: When do you expect the first Robotaxi to hit the road?

A: I think this question is actually asking when we can achieve unsupervised full self-driving. Obviously, my past predictions were too optimistic. However, based on current trends, we expect to achieve unsupervised level by the end of this year. If we still can't achieve unsupervised driving next year, I would be very surprised. So, the likelihood of achieving it next year is high, based on the trend curve of intervention miles.

Q: Cybertruck is an iconic product, and everyone who sees it praises it. Do you plan to expand the Cyber vehicle series, such as launching a Cyber SUV or Cyber van?

A: I think we want to limit product announcements to specific product launches rather than financial earnings conference calls.

Q: What is the current production status of the 4680 battery and progress in increasing production capacity?

A: 4680 battery production significantly increased in the second quarter, up 51% from the first quarter, while sales costs decreased significantly. We are currently producing over 1,400 4680 battery cells per week for CybertruckWe will continue to increase production capacity and further reduce costs, moving towards the cost parity target set for the year-end. We have completed the first dry-process negative electrode process verification line for large-scale production, which is an important milestone in the cycle, and we are very proud of this.

We are progressing with the mass production launch of the dry-process negative electrode according to plan, which will significantly lower battery costs compared to existing solutions, achieving the initial target for the 4680 product.

Q: Progress on Dojo?

A: Yes, what I want to point out is that I have great admiration for NVIDIA's execution and hardware capabilities. We see a very high demand for NVIDIA hardware, to the point where we struggle to obtain enough GPUs. This also prompts us to put more effort into Dojo to ensure we have the training capability we need.

Therefore, we will double our efforts to develop Dojo, and we do see our competitive supercomputing center. I think we have no choice because the demand for NVIDIA is so high that they will naturally raise GPU prices to a level the market can bear. So we must make Dojo successful, and we will.

Q: What types of accessories will Optimus provide?

A: Optimus is a general-purpose intelligent humanoid robot that will be compatible with any tools or accessories designed for humans.

Q: Do you think not advertising will make people lose the joy of owning a Tesla?

A: We will do some advertising. Our basic belief is that we need to offer consumers the best products at a reasonable price. In the United States alone, in the second quarter, over two-thirds of our deliveries were to people who had never owned a Tesla before, which is encouraging. We have invested in advertising and other promotional activities and adjusted our strategies. We are not completely averse to advertising, but we see it as a dynamic process where we have not exhausted all options and plan to continue adjusting in the second half of this year.

Q: What is the latest timetable for the Mexico factory? What major vehicles will initially be produced?

A: We have currently paused the Mexico factory project. I think we need to see how things go after the election, as Trump has indicated imposing heavy taxes on vehicles produced in Mexico. If that is indeed the case, then investing heavily in Mexico would not make sense. Therefore, we need to see how the political situation develops. However, we are significantly increasing the capacity of existing factories. Cybertruck and Robotaxi will be produced at our headquarters in Texas. In addition, the production version of Optimus will be produced in Texas by the end of next year.

Q: Is Tesla still negotiating FSD licensing with OEMs?

A: Indeed, several major OEMs have expressed interest in licensing Tesla's full self-driving system. I suspect there will be more companies like this in the future. However, we cannot disclose the specific details of these discussions.

Q: Latest progress on investing in XAI and integrating it into Tesla's software?A: Tesla has learned a lot from XAI, which actually helps drive the development of fully autonomous driving and establish a new Tesla data center. Regarding investing in XAI, I believe any such investment needs shareholder approval. But if shareholders support it, I will certainly support it. We believe there is an opportunity to integrate it into Tesla's software.

Q: An article mentioned that you transferred some AI computing systems originally intended for Tesla to XAI. When it comes to resource allocation between Tesla and XAI involving capital investment, AI research and AI engineers, how do you make allocation decisions between these companies to ensure the benefit of Tesla's shareholders?

A: The article you mentioned about GPUs was six to seven months ago. At that time, Tesla had no place to deploy these GPUs, so it was a waste for Tesla. I want to make it clear that this is not Tesla's support for XAI. We have been working on completing the south expansion project of the Texas Gigafactory, which will accommodate 50,000 H100 GPUs. We will move these GPUs in and activate them, all for the benefit of Tesla.

AI is a broad field. We focus on fully autonomous driving and Optimus, but there are also other companies working in the AI field, such as XAI. This is not just a specific thing. Regarding XAI, some people are only interested in AGI, so they will not join Tesla. We tried to recruit these people, but they declined. That's how XAI came about.

Q: Regarding the new car models you plan to launch next year, will the main changes be adjustments and upgrades based on the current models, or will there be other forms of changes? Will these become a strategic important part in the next year or two?

A: I don't want to discuss product announcements in detail. We need to be careful about the timing of announcements. If we announce something great, it may affect recent sales. We will continue to make great products, just as we have done in the past.

Q: Considering the rapid growth of the energy business and the launch of Optimus, how should we view the overall business sector in the long term? Do you think car revenue will drop to less than 50% of total revenue? Also, can you update the timeline for the new hardware for distributed computing and how to reward customers for participation?

A: In the long term, the value of Optimus will exceed the sum of all other Tesla businesses. We believe that global demand for general-purpose robots will exceed 2 billion units. Tesla has the most advanced robot technology and large-scale production experience, enabling efficient production of general-purpose robots. Therefore, AKA Investment estimates Tesla's long-term market value to be $50 trillion or more, and the market value of general-purpose robots may be several times that figure. In a benign AI scenario, we will enter an era of abundance where everyone can have almost anything.

Distributed computing is the obvious next step. Its value will be more pronounced on the next generation hardware, which we call the AI5 system, with performance comparable to or even surpassing the B200. We plan to start mass production next year and scale up production in 2026. If you have a fleet of autonomous driving vehicles running 50 to 60 hours a day, and there are 168 hours in a week, then each vehicle has about 100 hours of neural network computing timeThis is an obvious next step, using these computing capabilities for distributed reasoning. If we have a fleet of 100 million AI5 vehicles, plus tens of billions of humanoid robots such as AI6, AI7, etc., this is a huge computing power that can be used for general computing, not just for robots or cars.

Q: Regarding the mechanism of FSD licensing, I guess this is not a simple plug-and-play solution. An OEM needs several years to develop a vehicle platform based on FSD, using Tesla's electrical architecture, computing, and sensor suite, etc. So, if you cooperate with an OEM, it may take several years from signing the agreement to generating licensing revenue, right?

A: Yes, OEMs are not moving fast. Our sensor suite actually only requires cameras, but they need to integrate our AI computer and equip it with 360-degree cameras, as well as a gateway computer for communication with the Tesla system. Therefore, they need a gateway computer with cellular and WiFi connections and 7 cameras to get a 360-degree view. But considering the speed of the automotive industry, it may take several years to see mass production.

Q: So, if you sign an FSD licensing agreement with another automaker, do you think it will be made public at the time of signing the agreement or when the vehicles are ready to be launched in a few years?

A: It depends on the OEM's situation and the agreement we reach, there are many factors to consider. But obviously, this is only for some OEMs that can mass-produce. If the production scale is only 10,000 or 100,000 vehicles per year, we can do it ourselves.

Q: Considering the implementation of tariffs in Europe, will Tesla change its strategy of using Shanghai as an export center? Will there be any changes in your strategy of importing batteries from China to the US?

A: We are adjusting our strategy for importing to Europe from China. Europe has conducted tariff investigations on some OEMs, and we are cooperating with them, providing all the information. In the second quarter, we started producing right-hand drive Model Y in the Berlin factory and delivering in the UK. Nevertheless, we still export Model 3 from Shanghai to Europe and are evaluating other alternative solutions. We are confident in obtaining better tax rates, but this takes time. Meanwhile, the Berlin factory will increase exports to Taiwan and the UK. We will continuously adjust and adapt according to the situation.

Q: Regarding the Robotaxi strategy, a key factor for deployment is regulatory approval. What specific regulations should we focus on? Will the strategy be different? You have taken a nationwide borderless approach in FSD, so will Robotaxi still be subject to state legislation restrictions?

A: Our solution is universal, does not require high-definition maps, can work anywhere, unlike companies like Waymo that are still restricted by regional limitations. Once it is proven that unsupervised fully autonomous driving is significantly safer than human driving, regulatory agencies should support deployment. Therefore, we do not believe that regulatory approval will be a limiting factor. However, autonomous driving capabilities outside North America are far behind, and with the release of version 12.5, possibly after version 12.6, we will request approvals in Europe, China, and other countries, expecting approval by the end of the yearRegarding regulatory approval, all vehicles in the United States are governed by FMVSS standards, which are consistent across the rules of the 50 states. Therefore, creating a universal solution allows us to deploy quickly within the 50 states. Of course, as a transportation company or taxi deployment, different state and local regulations may apply. But in terms of getting vehicles on the road, this is a federal standard.

From a technical perspective, an end-to-end network basically does not consider location. You can obtain data from different countries, and it performs equally well in different places, just like humans can drive in other countries, we can design software in the same way.

Q: Regarding regulatory issues, General Motors canceled vehicles without steering wheels and pedals due to regulatory uncertainty. Tesla's Robotaxi also lacks a steering wheel and pedals. Will deploying these vehicles face different regulatory issues?

A: Obviously, General Motors' failure was due to their technology not meeting standards, not because of regulatory agencies misleading the public. Waymo operates well in these markets.

Q: You mentioned that after lowering the price, the adoption rate of FSD significantly increased. Can you quantify what this means specifically?

A: Yes, the adoption rate has significantly increased. Although we started from a lower base, the results are encouraging. The key is that customers need to experience FSD firsthand because its effects cannot be described in words. We ensure that every time a vehicle is delivered, we demonstrate how the system works to the customer. For example, I commute over 20 miles every day, almost entirely on autopilot, and the latest version can even track eye movements, with minimal alerts as long as you keep your eyes on the road. We are improving recognition accuracy so that even if you wear glasses, you can use it in the future. Many times, I chat with people who live in the center of New York or Boston, and they hardly drive. I tell them, you can try driving with FSD. If you haven't tried it yourself, you have no idea what the actual situation is.

Q: Regarding the deployment of Robotaxi, once you have vehicles that can replace taxis, is there a clear deployment plan? Is this a plan that Tesla will complete on its own? Or will it cooperate with local or global partners?

A: This will be the Tesla network. Users simply need to open the Tesla app to summon a car, and we will send a car to pick you up. Our fleet will soon reach 7 million vehicles, exceeding 10 million in the United States, and then 20 million. Vehicles can operate 24/7, unlike human drivers. Therefore, with just a software update, the fleet can expand immediately. This is a fleet owned by customers, who can choose to have their cars join the fleet to earn income or retrieve and use them at any time.

We will also produce some vehicles owned by Tesla that will directly join the fleet, which will be more like Uber. But this will be the Tesla network. Each Tesla vehicle will be restricted by terms of the agreement, only for use within the Tesla network and not for third-party autonomous driving services.

**Q: Is this scale gradually increasing? For example, starting with a few cars in a city and then increasing the number of vehicles over time? Or does it need to reach a certain critical mass to provide high-quality services that compete with existing services like Uber?**

A: I may not have made it clear, but basically the entire Tesla fleet will participate. Some people may not want their cars to be used to earn income, but I think most people would be willing. This is an immediate expansion.

Q: Can you talk about the pricing strategy for large fixed storage systems and the level of market saturation in specific regions?

A: We are working with a large number of participants in the market, signing contracts with a fairly long lead time to start deliveries. We have good pricing leverage. Competition from Chinese OEMs is very intense, just like in the automotive market. So we stay in close contact with customers to ensure competitiveness where needed to secure contracts and sales of energy. We have a very strong quarter for signings and continue to build up backlog orders for 2025 and 2026. We are very satisfied with our market position. Despite intense competition, we offer a fully integrated product, including power electronics and field control systems, which gives us a strong competitive edge.

The uniqueness of Megapack lies in its inclusion of the entire software stack, giving us a huge advantage over competitors. Customers sometimes try to piece together solutions, only to find that our solution is better, and then they come back to us. We have not seen saturation in the global market, just some saturation in specific markets. With the increasing demand for the power grid, new markets are constantly emerging, creating more opportunities globally. These GPUs are very powerful in AI computing. The volume of new orders we receive in areas like data center backup is increasing significantly.

Q: There have been some news about your equipment suppliers, can you talk about the progress of the qualification certification of potential incremental suppliers in the 4680 process technology steps?

A: You may be referring to the litigation issue we have with a supplier. I don't think this will affect our ability to roll out the 4680 battery. We have a very strong intellectual property position in this technology. Most of the equipment we use is internally designed, and some are also internally manufactured. So we can transfer our intellectual property to other manufacturers if needed. Therefore, this is not currently a problem.

People underestimate the demand for grid energy storage. In fact, the demand is severely underestimated. So, the actual energy output, such as the U.S. power grid, if power plants could run in a stable state, the total energy output would be at least 2-3 times the current energy production, because there is a huge gap. There is a significant difference between the peak and off-peak energy or power output. To keep the grid from blackouts, you must be able to support the load of the worst day of the year, which means there will be a lot of excess generating capacity at other times, but no way to store that energy. Once you add battery packs, you can make power plants run in a stable state. Stable state means that the cumulative energy output of any power plant in a year is at least twice the current level, sometimes even three times.

Q: If Trump wins, the IRA (Inflation Reduction Act) may be cut. You have said that Tesla does not rely on subsidies. But if subsidies are canceled, will it affect Tesla's profitability?A: There may be some impact, but this is devastating for our competitors and has a minor impact on us. In the long run, it may be beneficial for Tesla. Compared to the issue of autonomous driving, these are minor issues. I suggest that those who do not believe in the success of autonomous driving should not invest in Tesla stock. If you believe that Tesla will solve the issue of autonomous driving, then you should invest in Tesla.

When we consider the business, we have always assumed that the business can grow healthily regardless of subsidies. We hope the business can operate without subsidies, which has always been our consideration. Compared to our competitors, we have a significant advantage in this regard, as reflected in the data of the past few years. So, the fundamentals of the business cannot be ignored. In addition, with autonomous driving, these issues become insignificant.

Risk Disclosure and Disclaimer of this article: Dolphin Investment Research Disclaimer and General Disclosure