Written by | Zhang Linyu

Edited by | Zhang Nan

Designed by | Zhen Youmei

At 4:30 PM Central Time on July 22, Tesla's Q2 earnings call began, with little mention of its traditional automotive business.

During the Q&A session, analysts no longer asked when the low-cost model colloquially known as Model 2 would go into mass production. Tesla has never officially used this name, nor did they inquire about the progress of the next-generation model.

Elon Musk and his core executives shared information with the audience regarding FSD (Full Self-Driving, Tesla's fully autonomous driving software, which still requires driver supervision at all times in the current version), Robotaxi, Optimus robot, and chip factory.

This forms a certain contrast with Tesla's current revenue structure.

Revenue Increase Without Profit Increase

In Q2 2026, Tesla delivered a total of 480,126 vehicles globally, a 25% year-over-year increase and a 34% quarter-over-quarter increase, marking the company's best Q2 delivery record, second only to the historical record of 497,099 vehicles set in Q3 2025.

Among them, 467,762 Model 3 and Model Y vehicles were delivered, accounting for 97.4% of total deliveries; other models including Model S, Model X, Cybertruck, and Semi trucks totaled 12,364 deliveries, accounting for only 2.6%.

Production for the same period was 451,758 units. Deliveries exceeded production by 28,368 units, and global vehicle inventory days dropped from 27 days in the previous quarter to 15 days.

Tesla did not disclose absolute sales by region for Q2, but Chief Financial Officer Vaibhav Taneja revealed that compared to Q1, deliveries grew 60% in the Americas, 27% in Asia-Pacific, and 12% in EMEA. He also stated that Tesla ended Q2 with the largest order backlog since 2023.

The shareholder letter listed the markets where Q2 delivery records were set: South Korea, Australia, Colombia, Japan, Taiwan, Thailand, Portugal, Philippines, Chile, Slovenia, and Lithuania. Taneja additionally noted that Model Y set sales records in the Netherlands, Australia, and New Zealand.

Sales grew, but profits did not increase correspondingly.

In Q2, Tesla's total revenue was $28.236 billion, up 26% year-over-year, marking the first time revenue over the past 12 months exceeded $100 billion. Among them, automotive business revenue was $20.516 billion, up 23% year-over-year, accounting for 72.7% of total revenue; energy generation and storage business revenue was $3.139 billion, up 13%; service and other business revenue was $4.581 billion, up 50%, becoming the fastest-growing business.

Tesla's gross profit for the quarter was $4.751 billion, up 23% year-over-year; GAAP gross margin was 16.8%, down 41 basis points from 17.2% in the same period last year.

Operating profit was only $398 million, down 57% year-over-year; operating margin dropped from 4.1% to 1.4% compared to the same period last year.

In other words, Tesla's revenue increased by $5.74 billion year-over-year, but operating profit decreased by $525 million.

One source of pressure on profits came from expenses. In Q2, Tesla's R&D expenses reached $2.371 billion, up 49% year-over-year; sales, general, and administrative expenses reached $1.982 billion, up 45%. The two expenses totaled $4.353 billion, almost eating up all gross profit.

The shareholder letter attributed expense growth to three areas: R&D projects such as AI, equity incentives (including 2025 CEO compensation plan), and sales and admin expenses. Equity incentive expenses were $1.151 billion pre-tax this quarter, compared to $635 million in the same period last year.

Automotive regulatory credit revenue dropped from $439 million last year to $146 million, down 67% year-over-year; it was $380 million last quarter. The $7,500 U.S. federal EV tax credit expired on September 30, 2025; after federal law amendments, fines for automakers failing to meet fuel economy standards were zeroed out, removing the motivation to buy Tesla credits.

Automotive business GAAP gross margin was 16.9%; automotive gross margin excluding regulatory credit revenue was 16.3%, higher than 15.0% in the same period last year, but significantly lower than 19.2% in Q1 this year.

Taneja explained that Q1 automotive gross margin was helped by approximately $230 million in reserve reversals for warranties and tariff reductions; these benefits did not repeat in Q2; excluding these impacts, automotive gross margin excluding credits was roughly flat with the previous quarter.

In the attribution of revenue and profit in the shareholder letter, 'decline in average selling price per vehicle' is listed as a negative factor in two places.

The energy business also saw 'revenue increase without profit increase'.

In Q2, Tesla's energy storage deployment reached 13.5 GWh, up 41% year-over-year and 53% quarter-over-quarter, the second-highest quarter in history. However, due to approximately $240 million in warranty expenses caused by supplier cell issues, the energy business gross margin dropped from 39.5% in Q1 to 20.4%.

Service and other businesses became one of the few clear profit growth points.

This business's gross profit grew from $166 million in the same period last year to $648 million, an increase of $302 million quarter-over-quarter; gross margin rose from 5.5% last year and 9.2% last quarter to 14.1%, a historic high.

Tesla's GAAP net profit for the quarter was $1.114 billion, down 5% year-over-year; diluted EPS was $0.32, down 3% year-over-year.

This $1.114 billion includes an unrealized gain from a SpaceX equity investment, $1.005 billion pre-tax and $763 million after-tax. Tesla bought this SpaceX equity for $2.002 billion in Q1 2026, realizing a paper gain of $1.005 billion one quarter later.

Taneja said this approximately $1 billion gain was offset by approximately $300 million in foreign exchange losses and about $100 million in Bitcoin losses. On the income statement, the total 'Other income (expense), net' for the quarter was $590 million.

Excluding equity incentives, SpaceX paper gains, digital asset fluctuations, and special tax items, Tesla's non-GAAP net profit was $1.153 billion, down 17% year-over-year; non-GAAP diluted EPS was $0.33, down 18% year-over-year.

In Q2, Tesla's operating cash flow was $4.697 billion, up 85% year-over-year; capital expenditures reached $5.789 billion, up 142% year-over-year, resulting in $1.092 billion of negative free cash flow.

The company expects capital expenditures for the full year 2026 to exceed $25 billion, and capex will continue to grow in the next two to three years.

As of the end of June, Tesla held $43.524 billion in cash, cash equivalents, and short-term investments, down $1.219 billion from Q1.

Automotive remains Tesla's largest revenue source and the foundation supporting cash flow and capital expenditures, but the earnings call is no longer about cars.

Elon Musk Only Discussed Three Things

After the earnings call began, Elon Musk sounded emotionally down. He later explained to analysts that he was sick that day, 'If I sound a bit off, it's because I'm sick today and feeling unwell.'

He briefly mentioned three things.

The first thing was Cybercab. He announced that Cybercab production has started at the Texas Gigafactory. As a two-seater model designed specifically for Robotaxi with no steering wheel or pedals, Cybercab is seen as the core carrier for Tesla shifting from selling cars to operating an autonomous driving fleet. Tesla has provided Cybercab ride experiences for employees within the Texas factory campus since July.

However, Cybercab will not scale up rapidly in the short term.

Since it uses a completely new chassis, Tesla must first let test vehicles equipped with steering wheels, accelerator pedals, and brake pedals accumulate sufficient driving data to complete the autonomous driving system calibration for the Cybercab chassis.

The second thing was the Optimus robot.

Musk again called Optimus 'the largest product ever', but he also admitted this will be the hardest product for Tesla to date to achieve mass production.

Cars at least have mature supply chains for tires, glass, mirrors, and body parts, but almost all components of Optimus need to be redesigned and supply systems established.

Tesla has dismantled the Model S and Model X production lines at the Fremont factory and started installing the first-generation Optimus production line; the Optimus building at the Texas factory is also under construction.

The shareholder letter said the first batch of Optimus will be deployed within Tesla's Optimus Academy for collecting training data and developing features.

Musk stated that Optimus production ramp will follow a typical S-curve, but due to many new components, the initial flat phase of the curve will be very long.

Musk hopes the third-generation Optimus will eventually reach an annual production of 1 million units, and the fourth generation 10 million units. However, he immediately emphasized these are ideal targets, and achieving true mass production is 'surprisingly difficult'.

The third thing was the chip factory.

This is a massive chip manufacturing project in cooperation between Tesla and SpaceX. Terafab not only produces logic chips but also covers storage chips, advanced packaging, testing, and photomask manufacturing.

Musk did not disclose the site location at the earnings call, only stating there will soon be a separate launch event because this project 'deserves its own day' and should not be squeezed into an earnings call.

He envisioned placing photomask manufacturing, logic chips, storage chips, packaging, and testing in the same building to shorten chip design and validation cycles.

Terafab's main task is not to produce traditional chips for cars, but to solve the AI computing power needed for future large-scale Optimus mass production.

Musk believes that without its own chip manufacturing capability, Optimus will eventually be constrained by AI chip supply.

There was another chip statement in the shareholder letter: Tesla's semiconductor wafer fab in Austin is advancing construction and equipment procurement; the project is still in early stages, aiming to build its own chip manufacturing capability to ensure long-term supply of logic and storage chips. The full shareholder letter does not mention Terafab.

June 4, 2026, Hsinchu, Taiwan, TSMC 2026 Annual Shareholders' Meeting.

A corporate shareholder representative asked TSMC Chairman and CEO C.C. Wei how he viewed Musk building his own wafer fab. Wei said: 'My only comment is to 'bless him'.' He also said TSMC has never lacked opponents; the only way to face them is to continue to strive and win.

Cybercab, Optimus, and Terafab form the main axis of Musk's entire narrative at this earnings call. The traditional automotive business only exists as a cash source and AI implementation platform.

The Limitation on Production Is Not Demand, but Battery Packs

The shareholder letter stated in two places that battery pack capacity is the main limiting factor for recent vehicle production increases. Taneja added at the meeting that production growth will be limited by supply chains, not just batteries, but also electronic components.

In the same quarter, Tesla faced the largest order backlog since 2023.

As of the end of Q2, Tesla's paid FSD subscriptions reached 1.48 million, up 56% year-over-year; global paid users approached 1.5 million, with 55% one-time purchase and 45% subscription. Over 55% of new cars in North America had FSD subscriptions activated at delivery.

The shareholder letter stated Tesla has pushed FSD v14 lite to vehicles equipped with AI3 hardware in the U.S. and South Korea, by distilling driving behaviors from the v14 series on AI4 into the camera and compute configuration of AI3.

Hardware 3 and Hardware 4 correspond to AI3 and AI4. Musk said in the Q&A session that upgrading vehicles below Hardware 4 'will eventually be economically reasonable', but Tesla public documents have no hardware upgrade plans.

The shareholder letter listed nine Robotaxi metro areas: California's SF Bay Area with safety drivers; Texas Austin, Dallas, Houston and Florida Miami, Orlando, Tampa in unsupervised operational ramp-up phase; Arizona Phoenix, Nevada Las Vegas in preparation.

Tesla claimed Robotaxi is online in 7 major metro areas, cumulatively driving over 380,000 miles safely with no accident records.

Analyst Q&A Transcript

Below is the transcript of the Q&A session from Tesla's Q2 2026 Earnings Call, compiled by this publication from the conference call recording, with omissions.

Earnings Call Executive List

Host: Travis Axelrod (Travis Axelrod, Head of Investor Relations at Tesla)

Musk (Elon Musk, Tesla Co-founder and CEO)

Vaibhav Taneja (Vaibhav Taneja, Tesla CFO)

Ashok Elluswamy

Karn Budhiraj

Lars Moravy (Lars Moravy, Vice President of Vehicle Engineering at Tesla)

Brandon Ehrhart (Brandon Ehrhart, Tesla General Counsel and Corporate Secretary)

Q1: Currently there is no mature Optimus supply chain, so Tesla has to do much of the work internally. In the process of advancing this, are there suppliers willing to co-invest with you to establish local manufacturing capability in the U.S., helping Tesla achieve scale faster? From an investment cycle perspective, having partners advance together could improve capital efficiency. How do you consider this?

Musk: Our suppliers are performing exceptionally well. To support Optimus, Robotaxi, and other projects, they have invested and are continuing to invest huge funds.

Especially Samsung Electronics and TSMC, both building wafer fabs. TSMC is building in Arizona, Samsung is building in Texas, investing tens of billions of dollars to produce AI computing chips for Optimus and Robotaxi.

Panasonic has also invested billions of dollars to expand battery production.

Please excuse me, I'm feeling a bit unwell today, slightly sick. If my voice sounds off, it's because I'm indeed feeling unwell today.

Our partners are all excellent. I want to thank them for their support, their investments, and their hard work.

Karn Budhiraj: I want to add a point on this basis. A significant portion of Samsung's wafer fab capacity will be used for future projects. This is a massive investment involving billions of dollars.

We also see investment of similar scale in the storage chip sector. Additionally, investment has emerged in some new dedicated parts areas, such as metal injection molding parts, flexible printed circuits, and various non-traditional technologies. These technologies are mostly designed for robots, differing from our traditional automotive supply chains in the past.

In some areas, if we cannot find suitable partners, we never hesitate to move production internally. We have a very strong manufacturing engineering team and design team that can help us scale up production of these products.

Musk: I also want to thank Micron Technology (Micron Technology, U.S. storage chip manufacturer) for allocating storage chip capacity for us. They must make some very difficult decisions on storage chip capacity allocation. Given the crazy high price of storage chips today, we are very grateful that Micron is willing to reserve space for Tesla in the coming years and allocate considerable capacity to us under reasonable terms.

Q2: Regulations in U.S. states seem to be changing, possibly involving sensor configuration requirements. For Robotaxi deployment, what approach do you want regulators to take? To avoid excessive regulation while ensuring Robotaxi lands correctly, what regulatory environment do you hope to see?

Lars Moravy: When discussing regulations, especially the U.S. regulatory environment, significant progress has been made. The federal government has taken very proactive measures around the 'Federal Motor Vehicle Safety Standards' (FMVSS), beginning to accept and promote specifically designed autonomous vehicles. We are very grateful for the support from the U.S. National Highway Traffic Safety Administration (NHTSA) in this regard.

As for individual states, some recent movements in New Jersey are obviously somewhat disappointing.

But as Ashok said in his opening remarks, what we really care about is the vehicle's own performance. Ultimately, it is product performance that drives regulators and the public to accept this technology.

I believe the best way to regulate is to set a goal or task for companies and innovators, then let us find solutions ourselves. If some regulations define the solution before clearly stating the problem, we certainly won't welcome that practice.

Tesla has always proven itself with actual performance in the past, and will continue to do so in the future. This will also become the foundation for our continued expansion.

Q3: Next, what milestones should investors focus on? After reaching which milestones will you accelerate Robotaxi fleet expansion or market deployment? Is it further safety validation, such as mileage per accident, or other metrics? Additionally, will you consider partnering with third parties like ride-hailing platforms to increase vehicle utilization, or continue to maintain full vertical integration?

Musk: We expect Robotaxi to adopt a vertical integration model, like Tesla's other businesses. I don't think Robotaxi will face a problem of insufficient demand. Its economics will be extremely attractive, with many people wanting to use this service, and demand will exceed our service capacity. In this case, vertical integration is obviously the choice.

The real issue is just what we call the progression of reliability 'multiple 9s'. That is, to scale up, how many 9s of reliability does the system actually need to reach?

Ideally, we want reliability to reach 99.999999%. I believe the only limitation to Robotaxi growth is moving towards more 9s of reliability.

Q4: Outsiders have been discussing the possibility of SpaceX and Tesla merging in some way. The two companies have already carried out extensive cooperation. Will merging the two companies eventually produce synergies? As time passes, is this reasonable?

Musk: As everyone can see from the extensive cooperation carried out by the two companies in many fields, the overlap between their businesses is increasing, especially the Terafab project. This will be a true massive project.

But obviously, we cannot discuss matters like company mergers on an earnings conference call. Such matters must be carried out according to appropriate procedures.

Brandon Ehrhart: We have always benefited from our relationship with SpaceX. SpaceX is an excellent partner, and we have conducted many transactions beneficial to both parties with them.

Earlier this year, we deepened the relationship between the two sides through an investment and a framework agreement. This will enable us to continue working with SpaceX to advance the projects Elon just mentioned, such as Terafab and Digital Optimus.

Musk: Additionally, there are many other collaborations. Grok (the large model of Musk's AI company xAI) has now been integrated into the car, and Grok is also helping drive Digital Optimus. Starlink (SpaceX's satellite internet service) is being integrated into Cybercab. In the future, Starlink will also be integrated into all our automotive products, at least in markets where Starlink is already operating.

Because in Robotaxi application scenarios, the network must cover all places.

Even in Silicon Valley, there are many places with very poor mobile network signals, sometimes even no signal at all. When I drive to work, the first 10 to 15 minutes I cannot actually make calls because the mobile network connection is too poor.

We cannot let Robotaxi enter these 'Bermuda Triangles' lacking mobile network connections and get stuck there. Starlink can provide network connections anywhere, so it is very important. This avoids Robotaxi losing connection.

If people are sitting in the car, they might want to handle work efficiently, or they might hope to get an entertainment experience. With Starlink, people can watch 4K live sports events in the car, and the cost per GB of data is very low. This is actually difficult to achieve through mobile communication networks.

Vaibhav Taneja: You can understand this from the Cybercab ride experience. In a world where you don't need to focus on driving, all the time you sit in the car can be used for other things, whether attending conference calls, watching movies, or doing other activities.

Therefore, if you observe the design of Cybercab, you will see it is equipped with a very large screen.

We have started providing Cybercab ride experiences within the Austin factory. In the not-so-distant future, consumers will also be able to experience it. Once you truly experience it, you will understand why network connections become so important.

Q5: Currently Robotaxi is constantly adding operating cities, but according to media reports, the number of vehicles seems to still be only dozens, not hundreds. Why not concentrate on expanding the scale of Austin or one or two cities first? To achieve higher deployment volume in a major city, what else do you need to solve?

Ashok Elluswamy: The reason we chose to expand to different cities instead of focusing on only one city is that we want to ensure the tech stack has strong generality.

It is indeed a general system. We just want to prove to ourselves and the outside world that this system can run in many different cities, and entering each city does not require investing too much extra work. This is exactly what we see internally.

As Elon just mentioned, the growth rate is actually rising exponentially, it's just that it's currently in the early stage of the exponential curve, so the outside world finds it hard to understand this growth.

As for whether to focus on mileage or vehicle count, since vehicles in the Robotaxi fleet will basically run continuously, while human drivers might only use vehicles a few hours a day, there is a huge difference between the two.

These vehicles will operate continuously for the vast majority of the time. This means that even if the number of vehicles is small, a large amount of mileage can be generated. Therefore, we focus more on unsupervised driving mileage rather than simply the number of vehicles.

We hope to operate a very efficient fleet. All work that improves operational efficiency will further increase the mileage each vehicle in the fleet can contribute.

Vaibhav Taneja: We are trying to expand this business, while also ensuring to solve various problems that may exist. These problems involve not only software but also operations, and we are also working on handling them. Therefore, we hope to first expand the geographic coverage, and before truly large-scale deployment, solve these problems through a smaller, controllable fleet.

Musk: Regarding Cybercab, one point needs special explanation. Because Cybercab uses a completely new vehicle chassis, before mass deployment, we must accumulate driving data specifically for Cybercab.

Model 3, Model Y, and other models already have millions of cars driving on the road, we have a lot of data, but Cybercab does not yet have such data foundation.

We must first let Cybercab equipped with devices such as steering wheels, accelerator pedals, and brake pedals accumulate mileage, thereby completing system calibration for the Cybercab chassis.

When we establish sufficient confidence in this, everyone will see a significant increase in the number of Cybercabs in various cities.

Lars Moravy: Another reason to point out is that total cost of ownership and transportation regulatory environments are not the same between different cities and different states. Currently, there is no unified federal regulatory framework.

The reason we expand city by city is to ensure we can meet these requirements separately and respond to the specific requirements of each city as much as possible.

We must enter every city to meet local requirements. We will continue to do this, and expand the scale further after meeting all conditions.

Q6: The U.S. National Highway Traffic Safety Administration is pushing to cancel pedals and steering wheels. Will these become factors limiting Cybercab production expansion? At the federal level, are there any other problems that need to be solved to truly release your ability to expand Cybercab scale?

Lars Moravy: The short answer is, no. We maintain a very good relationship with the U.S. National Highway Traffic Safety Administration, especially with Administrator Jonathan Morrison.

I think they are just catering to public demand and the trend known worldwide to be coming. They are trying to stay ahead of the trend to ensure appropriate measures are taken.

In the past few years, we have always honestly introduced our plans and ongoing work to them.

I don't want to say the two sides are completely in step, but I feel we have a partner there, and both sides are advancing relevant work together.

[Editor's Note: Morrison was confirmed by the U.S. Senate to serve as NHTSA Administrator on September 18, 2025. He served as NHTSA General Counsel from 2017 to 2021. Before returning to NHTSA in 2025, he was responsible for legal, regulatory, government affairs, and policy in Apple's Special Projects Group (i.e., Apple's car project)]

Q7: Starlink has been integrated into Cybercab. Can Cybercab in turn become a remote hotspot for Starlink mobile services? Also, when do you expect to start developing autonomous driving for Tesla Semi? Considering the shortage of truck drivers, autonomous driving trucks also seem to be a market of large scale.

Musk: As you mentioned, there is indeed a very serious shortage of truck drivers now. Not enough people are willing to drive trucks, and trucks are crucial for U.S. freight transportation. Autonomous Semi will be very important for solving the truck driver shortage problem and also helps improve safety.

For Tesla Semi still driven by truck drivers, autonomous driving functions will also significantly improve driving safety and comfort.

Since the number of Tesla Semis is currently still very small, even by the end of this year, its proportion in Tesla's total vehicles will still be very low, therefore, it is reasonable to prioritize autonomous driving R&D resources for high-volume models.

We will prioritize solving the autonomous driving problems of Model 3, Model Y, and Cybercab, truly making these models achieve universal, unsupervised autonomous driving.

We expect Tesla Semi's autonomous driving function might be realized by the end of this year or early next year. I just don't want the Semi project to distract our attention from continuously improving safety reliability and moving towards more 9s on Model 3, Model Y, and Cybercab.

Therefore, in the next approximately six months, the priority of the autonomous Semi project will be slightly pushed back. However, it will definitely be achievable next year and able to catch up with Tesla Semi's process of large-scale production increase.

As for communication, we might treat the Starlink terminals in Tesla cars as mobile communication base stations, or some kind of network connection relay towers, providing network relay to mobile phones on the ground and anyone who wants to use Wi-Fi.

Obviously, fixed Starlink terminals can also provide similar services. [Editor's Note: Shareholder letter states Tesla Semi will start production at a new factory in Nevada this year]

Q8: Optimus will use microprocessors, microcontrollers, actuators, and other semiconductors. Tesla can purchase from the market, design them themselves and hand them over to third-party wafer fabs for manufacturing, or design and produce them themselves. For these components, which approach do you prefer?

Musk: Optimus uses a large number of highly specialized power electronic devices and circuit boards. All of these are designed by Tesla, but the manufacturing work is completed by suppliers.

Fourth-generation Optimus will be produced in Austin. Compared to third-generation products, the supply system of fourth-generation Optimus will achieve a higher degree of vertical integration.

Our goal is for the production volume of fourth-generation Optimus to be an order of magnitude higher than third-generation products. Ideally, the annual production of third-generation Optimus will be 1 million units, and fourth-generation products will reach 10 million units per year.

Of course, all relevant cautionary statements need to be added here, because scaling up production is an extremely difficult task.

Fourth-generation Optimus will adopt a higher degree of vertical integration. Therefore, at that time we may complete a large amount of printed circuit board related work internally.

Q9: Does Tesla still plan to upgrade Hardware 3 to Hardware 4 in order to run V15 and future versions of FSD? Will existing vehicles still undergo hardware upgrades?

Musk: I think upgrading all vehicles below Hardware 4 will eventually be economically reasonable, basically all cars already equipped with cameras.

If there is not even a camera system, the required modifications might be too many. For any vehicle already designed around the camera system, upgrading at a certain point in time will be economically reasonable. I think we hope to upgrade these vehicles directly to the next-generation AI compute board.

Specifically, there might be two options. We have an upgraded AI4 chip, with a certain degree of improvement compared to AI4, which might go into production around mid-next year. Then there is AI5, we hope AI5 can also achieve mass production around mid-next year. AI5 will initially be applied to Optimus.

Currently, we have made very good progress in AI5. Tesla's chip team is performing excellently, advancing AI5 R&D at an unbelievable speed.

I am very excited about the design of Tesla's AI6 chip. I think it will become the world's best edge computing chip. If someone owns a better chip, I really want to meet him and shake hands, because the AI6 chip is really too excellent.

Progress in chips is very smooth. Thank you again to TSMC, Samsung, and Micron for their support.

Q10: In the current round of capital expenditure cycle, to what extent is the investment speed limited by efficiency? If investment continues to increase, will it lead to a decrease in capital efficiency? To what extent does the capital expenditure speed determine Tesla's progress in lifting supply constraints?

Musk: The requirement I put to the Tesla team is that as long as it doesn't cause excessive waste, we should carry out capital expenditure as quickly as possible.

We are not pursuing extreme high efficiency of capital investment, because that would slow down project progress. We need to strike a balance between capital efficiency and time.

If the project can be completed earlier, then slightly lower capital efficiency is acceptable, because from the company's perspective, this will actually bring higher net present value.

Overall, I am quite satisfied with the current progress. We are carrying out construction and capacity expansion of astonishing scale in multiple completely different fields simultaneously.

I think, previously no company has advanced so much work on such a large scale at the same time. Perhaps when Henry Ford expanded Model T production, it was close to this situation in relative scale; or during World War II, enterprises once tried to massively expand weapon production.

I think, this might be the fastest industrial-scale expansion in the United States since World War II.

Vaibhav Taneja: If you observe our capital expenditure, you will find that all expenditures are used to build assets that can be put into production. We are expanding various factories, including Optimus factories, Cybercab factories, Lithium Iron Phosphate Battery factories that started production earlier this year, Semi factories, and even semiconductor wafer factories.

In addition, we also plan to build large-scale solar manufacturing capacity in the U.S., raising the scale of solar manufacturing in the U.S. by an order of magnitude. This point is very important, the importance of solar manufacturing is underestimated.

None of this work is easy. To build a factory, you must start from zero. Due to too many construction projects underway, Tesla eventually became the general contractor for almost all construction projects.

We are expanding at a very fast speed. This means we need to advance many things simultaneously. Therefore, we can only move forward at the fastest speed humans can reach to ensure these projects can truly operate in the real world.

Musk: I think our capital expenditure efficiency is good beyond conventional scales.

Q11: In the foreseeable future, to what extent will the energy storage business be limited by supply? Of the current energy storage projects, how much is used for public utility peak shaving and valley filling, and how much is used to solve power quality problems caused by drastic fluctuations in electricity consumption at data centers? What is the state of demand in the data center and utility sectors respectively?

Musk: This is not just peak shaving and valley filling, it also includes grid balancing. Battery packs are not only activated for short periods. For balancing the power generated by wind power and solar power, batteries are an excellent tool.

In the future, the vast majority of global energy will be produced through a combination of solar and batteries. All Starlink satellites are also powered by solar panels and batteries. You can completely imagine the Earth as a huge satellite.

The energy the sun can provide far exceeds any other energy source. We believe that power constraints will become a major problem facing AI. In fact, it is already a major problem.

Just starting AI computers requires consuming a large amount of power. The power demand for AI computing is very high, and even super-large cloud computing companies find it difficult to find enough power for AI computing and smoothly enable relevant equipment.

In addition, power fluctuations must be suppressed. Especially during AI model training, power demand will change drastically in a very short time.

During a training process, power consumption might drop 70% within 100 milliseconds. This requires very fast-responding advanced power electronic equipment to suppress large power changes, especially changes during AI model training.

This is why SpaceX bought a large number of Megapacks (Tesla's large-scale energy storage product) for data centers. They are mainly used to suppress power fluctuations appearing during training. Batteries can also help data centers obtain more power from the grid.

If you tell utility companies that during the few hours or days with the worst power usage conditions in the year, they don't need to supply power to you because batteries can undertake the power supply task, then they will be easier to provide power access to you.

In fact, the best way to increase the total output of energy in the United States might be to use batteries on a large scale. The United States has about 1.2 to 1.3 terawatts of power generation capacity, but the average power consumption is only about 0.5 terawatts. That is to say, the power generation capacity in the United States is about 2.5 times the average power consumption.

This means that by just using batteries, the energy output of the United States could potentially double. Therefore, we believe that the demand for Megapacks in the future market will be very high.

撰文 | 張霖郁

編輯 | 張 南

設計 | 甄尤美

美國中部時間 7 月 22 日下午 4:30，特斯拉第二季度財報會開始，幾乎冇涉及傳統汽車業務。

問答環節，分析師們不再追問外界俗稱 Model 2 嘅廉價車型何時量產兌現，呢個名特斯拉從未正式用過，亦不再過問新一代車型嘅進展。

馬斯克本人以及佢嘅核心高層同大家分享嘅，係 FSD（Full Self-Driving，特斯拉嘅完全自動駕駛軟件，當前版本仍要求駕駛員全程監督）、Robotaxi、機器人 Optimus 以及芯片工廠。

呢樣同特斯拉當下嘅收入結構形成某種反差。

增收不增利

2026 年第二季度，特斯拉喺全球共交付 480126 輛汽車，同比增長 25%，環比增長 34%，創下公司史上最好嘅第二季度交付成績，僅次於 2025 年第三季度創下嘅 497099 輛歷史記錄。

其中，Model 3 同 Model Y 交付 467762 輛，佔總交付量嘅 97.4%；包括 Model S、Model X、Cybertruck 同 Semi 卡車在內嘅其他車型合計交付 12364 輛，只佔 2.6%。

同期產量 451758 輛。交付比產量多出 28368 輛，全球整車庫存天數從上一季度嘅 27 日降到 15 日。

特斯拉冇公佈第二季度各地區嘅絕對銷量，但首席財務官瓦伊巴夫·塔內賈（Vaibhav Taneja）透露，與第一季度相比，美洲地區交付量增長 60%，亞太地區增長 27%，歐洲、中東同非洲地區增長 12%。佢仲講，特斯拉退出第二季度時，持有 2023 年以來最大嘅在手訂單積壓。

股東信列出的第二季度交付創紀錄市場係韓國、澳大利亞、哥倫比亞、日本、中國台灣、泰國、葡萄牙、菲律賓、智利、斯洛文尼亞同立陶宛。塔內賈另外提到，Model Y 喺荷蘭、澳大利亞同新西蘭創下銷售紀錄。

銷量增長，利潤卻冇同步嚟。

第二季度，特斯拉總營收為 282.36 億美元，同比增長 26%，過去十二個月營收首次突破 1000 億美元。其中，汽車業務收入 205.16 億美元，同比增長 23%，佔公司總收入嘅 72.7%；能源發電與儲能業務收入 31.39 億美元，同比增長 13%；服務及其他業務收入 45.81 億美元，同比增長 50%，成為增長最快嘅業務。

特斯拉當季毛利潤為 47.51 億美元，同比增長 23%；GAAP 毛利率為 16.8%，同去年同期嘅 17.2% 下降 41 個基點。

營業利潤只有 3.98 億美元，同比下降 57%；營業利潤率從去年同期嘅 4.1% 降至 1.4%。

換句話講，特斯拉營收同比增加 57.40 億美元，營業利潤卻減少咗 5.25 億美元。

利潤受到嘅一重壓力來自費用。第二季度，特斯拉研發費用達到 23.71 億美元，同比增長 49%；銷售、一般同管理費用達到 19.82 億美元，同比增長 45%。兩項費用合計 43.53 億美元，幾乎食晒全部毛利潤。

股東信把費用增長歸為三處：AI 等研發項目、股權激勵（含 2025 年 CEO 薪酬方案）、銷售同管理費用。當季股權激勵費用稅前 11.51 億美元，去年同期為 6.35 億美元。

汽車監管積分收入從去年同期嘅 4.39 億美元降至 1.46 億美元，同比下降 67%，上一季度仲有 3.80 億美元。7500 美元嘅美國聯邦電動車稅收抵免已於 2025 年 9 月 30 日到期；聯邦法律修改後，車企未達標燃油經濟性標準嘅罰款被歸零，購買特斯拉積分嘅動機隨之消失。

汽車業務 GAAP 毛利率為 16.9%，扣除監管積分收入後嘅汽車毛利率為 16.3%，高於去年同期嘅 15.0%，但明顯低於今年第一季度嘅 19.2%。

塔內賈解釋，第一季度嘅汽車毛利率曾受到約 2.3 億美元質保準備金回撥同關稅減免幫助，呢啲利好冇喺第二季度重複出現；剔除呢啲影響，扣除積分嘅汽車毛利率與上一季度基本持平。

股東信喺營收同利潤嘅歸因入面，兩處把「單車平均售價下降」列為負面因素。

能源業務亦出現咗「增收不增利」。

第二季度，特斯拉儲能部署量達到 13.5GWh，同比增長 41%，環比增長 53%，為歷史第二高季度，但由於供應商電芯問題帶來約 2.4 億美元質保費用，能源業務毛利率從第一季度嘅 39.5% 下降至 20.4%。

服務及其他業務成為少數明確嘅利潤增長點。

該業務毛利潤由去年同期嘅 1.66 億美元增長至 6.48 億美元，環比增加 3.02 億美元；毛利率由去年同期嘅 5.5%、上一季度嘅 9.2% 升至 14.1%，創歷史新高。

特斯拉當季 GAAP 淨利潤為 11.14 億美元，同比下降 5%；攤薄後每股盈餘為 0.32 美元，同比下降 3%。

呢 11.14 億美元入面包含一筆 SpaceX 股權投資未實現收益，稅前 10.05 億美元，稅後 7.63 億美元。特斯拉喺 2026 年第一季度以 20.02 億美元買入呢筆 SpaceX 股權，一個季度後浮盈 10.05 億美元。

塔內賈講，呢筆約 10 億美元嘅收益被約 3 億美元匯兌損失同約 1 億美元比特幣損失抵銷。利潤表上，當季「其他收益（費用）淨額」合計 5.90 億美元。

剔除股權激勵、SpaceX 浮盈、數字資產波動同特殊稅務項目後，特斯拉非 GAAP 淨利潤為 11.53 億美元，同比下降 17%；非 GAAP 攤薄後每股盈餘為 0.33 美元，同比下降 18%。

第二季度，特斯拉經營現金流為 46.97 億美元，同比增長 85%；資本開支達到 57.89 億美元，同比增長 142%，由此產生 10.92 億美元負自由現金流。

公司預計，2026 年全年資本開支將超過 250 億美元，未來兩三年資本開支仲會繼續增長。

截至 6 月底，特斯拉持有 435.24 億美元現金、現金等價物及短期投資，較第一季度減少 12.19 億美元。

汽車依然係特斯拉最大嘅收入來源，亦係支撐現金流同資本開支嘅基本盤，但財報會已唔係一場關於汽車嘅財報會。

馬斯克只講咗三件事

財報會開始後，馬斯克嘅聲音聽起嚟情緒唔佳，佢其後向分析師解釋，自己當日病咗，「如果聽起嚟有點唔對，係因為我今日生病，有些唔舒服。」

佢簡單講咗三件事。

第一件事係 Cybercab。佢宣佈，Cybercab 已喺德克萨斯超級工廠開始生產。作為一款冇方向盤同踏板、專門為 Robotaxi 設計嘅兩座車型，Cybercab 視為特斯拉從賣車轉向運營自動駕駛車隊嘅核心載體。特斯拉 7 月起已喺得州工廠園區內為員工提供 Cybercab 乘坐體驗。

不過，Cybercab 短期內唔會迅速放量。

由於佢採用全新底盤，特斯拉必須先讓配備方向盤、加速踏板同制動踏板嘅測試車積累足夠多嘅行駛數據，完成針對 Cybercab 底盤嘅自動駕駛系統校準。

第二件事係機器人 Optimus。

馬斯克再次把 Optimus 稱呼為「可能係有史以來最大嘅產品」，但佢同時承認，呢將係特斯拉迄今最難實現規模化生產嘅產品。

汽車至少擁有成熟嘅輪胎、玻璃、後視鏡同車身零部件供應鏈，Optimus 卻幾乎所有部件都需要重新設計同建立供應體系。

特斯拉已經拆除弗里蒙特工廠嘅 Model S 同 Model X 生產線，開始安裝第一代 Optimus 生產線，得州工廠嘅 Optimus 廠房亦喺建設緊。

股東信講，首批 Optimus 將投入特斯拉內部嘅 Optimus Academy，用於採集訓練數據同開發功能。

馬斯克稱，Optimus 生產爬坡將遵循常見嘅 S 形曲線，但由於大量零部件係全新嘅，曲線最初嘅平緩階段會非常漫長。

馬斯克希望第三代 Optimus 最終達到年產 100 萬台，第四代達到年產 1000 萬台。不過，佢隨即強調，呢啲只係理想目標，真正實現規模化生產「難度高得驚」。

第三件事係芯片工廠。

呢係特斯拉同 SpaceX 合作嘅巨型芯片製造項目。Terafab 不僅生產邏輯芯片，仲要覆蓋存儲芯片、先進封裝、測試同光刻掩膜製造。

馬斯克冇喺財報會上公佈選址，只講好快會有一場單獨嘅發布活動，因為呢一項目「值得擁有自己嘅一天」，而不應該被塞入財報會。

佢設想將光刻掩膜製造、邏輯芯片、存儲芯片、封裝同測試放喺同一座建築物中，以縮短芯片設計同驗證週期。Terafab 嘅主要任務並唔係畀汽車生產傳統芯片，而係解決未來 Optimus 大規模量產所需要嘅 AI 計算能力。

馬斯克認為，如果冇自己嘅芯片製造能力，Optimus 最終會受到 AI 芯片供應限制。

股東信入面另有一處芯片表述：特斯拉喺奧斯汀嘅半導體晶圓廠正喺推進建設同設備採購，項目仍處早期，目的係自建芯片製造能力，保障邏輯芯片同存儲芯片嘅長期供應。股東信全文冇出現 Terafab。

2026 年 6 月 4 日，中國台灣新竹，台積電 2026 年度股東常會。

一位法人機構股東代表問台積電董事長兼行政總裁魏哲家點睇馬斯克自建晶圓廠。魏哲家講：「我唯一嘅說法就係『祝福佢』。」佢同時講，台積電一直唔缺對手，唯一能面對嘅方式就係持續努力持續贏。

Cybercab、Optimus 同 Terafab，構成咗呢場財報會上馬斯克全部敘事嘅主軸。傳統汽車業務只係作為現金來源同 AI 落地平台存在。

限制產量嘅唔係需求，係電池包

股東信兩處寫明，電池包產能係近期整車產量提升嘅主要限制因素。塔內賈喺會上補充，產量增長將受供應鏈限制，唔只係電池，仲包括電子元器件。

同一個季度，特斯拉面臨 2023 年以來最大嘅在手訂單積壓。

截至第二季度尾，特斯拉付費 FSD 訂閱數 148 萬份，同比增長 56%；全球付費用戶接近 150 萬，其中 55% 為一次性買斷，45% 為訂閱。北美超過 55% 嘅新車喺交付時已開通 FSD 訂閱。

股東信稱，特斯拉已同美國同韓國搭載 AI3 硬件嘅車輛推送 FSD v14 lite，做法係把 AI4 上 v14 系列嘅駕駛行為蒸餾進 AI3 嘅攝像頭同算力配置。

Hardware 3、Hardware 4 對應嘅正係 AI3、AI4。馬斯克喺問答環節講，升級低於 Hardware 4 嘅車輛「最終會擁有經濟合理性」，特斯拉公開文件入面冇硬件升級計劃。

股東信列出九個 Robotaxi 都會區：加州舊金山灣區配備安全員；得克萨斯州奧斯汀、達拉斯、休斯頓同佛羅里達州邁阿密、奧蘭多、坦帕處於無監督運營爬坡階段；亞利桑那州菲尼克斯、內華達州拉斯維加斯籌備緊。

特斯拉稱 Robotaxi 已喺 7 個主要都會區上線，累計安全行駛超過 38 萬英里，無事故記錄。

分析師問答實錄

以下係特斯拉 2026 年第二季度財報會嘅回答環節實錄，本刊據電話會錄音編譯，有刪節。

財報會現場高層名單

主持人：特拉維斯·阿克塞爾羅德（Travis Axelrod，特斯拉投資者關係主管）

馬斯克（Elon Musk，特斯拉聯合創始人、行政總裁）

瓦伊巴夫·塔內賈（Vaibhav Taneja，特斯拉首席財務官）

阿肖克·埃盧斯瓦米（Ashok Elluswamy）

卡恩·布迪拉傑（Karn Budhiraj）

拉爾斯·莫拉維（Lars Moravy，特斯拉車輛工程副總裁）

布蘭登·爾哈特（Brandon Ehrhart，特斯拉總法律顧問兼公司秘書）

Q1：目前實際上並唔存在一條成熟嘅 Optimus 供應鏈，因此特斯拉不得不將好多工作轉為內部完成。喺推進呢項工作嘅過程中，有無供應商願意同你哋共同投資，喺美國建立本土製造能力，從而幫助特斯拉更快實現規模化？從投資週期睇，讓合作夥伴共同推進可能會提高資本效率。你哋點樣考慮呢個問題？

馬斯克：我哋嘅供應商表現非常非常出色。為咗支持 Optimus、Robotaxi 以及其他項目，佢哋已經投入，正喺投入巨額資金。

尤其係三星電子同台積電，佢哋都喺建設晶圓廠。台積電正喺亞利桑那州建廠，三星則喺得州建廠，並投入數百億美元，為 Optimus 同 Robotaxi 生產 AI 計算芯片。

松下亦投入咗數十億美元，用於擴大電池生產。

唔好意思，我今日身體有些唔舒服，有一點生病。如果我嘅聲音聽起嚟唔太啱，係因為我今日確實有些唔舒服。

我哋嘅合作夥伴都非常出色。我要多謝佢哋給予我哋嘅支持，多謝佢哋所做嘅投資同付出嘅辛勤勞動。

卡恩·布迪拉傑：我想喺呢個基礎上補充一啲。三星嘅晶圓廠將有相當大一部分產能用於未來項目。呢係一筆規模巨大嘅投資，涉及數十億美元。

我哋仲看到，存儲芯片領域亦喺進行同等規模嘅投資。此外，一啲新嘅專用零部件領域亦出現咗投資，例如金屬注射成型零部件、柔性印刷電路，同各種非傳統技術。呢啲技術更多係為機器人設計嘅，同我哋過去嘅傳統汽車供應鏈唔同。

喺某些領域，如果我哋搵唔到合適嘅合作夥伴，我哋從來唔會猶豫是否將其轉為內部生產。我哋擁有一支能力非常強嘅製造工程團隊同設計團隊，可以幫助我哋擴大呢啲產品嘅生產規模。

馬斯克：我仲要感謝美光科技（Micron Technology，美國存儲芯片製造商）為我哋分配存儲芯片產能。佢哋必須喺存儲芯片產能分配方面做出一些非常艱難嘅決定。考慮到如今存儲芯片價格高得瘋狂，我哋非常感謝美光願意喺未來幾年為特斯拉留出空間，並合理條件向我哋分配相當可觀嘅產能。

Q2：美國各州嘅監管規定似乎正喺發生變化，其中可能涉及傳感器配置要求。對於 Robotaxi 部署，你哋希望監管機構採取咩做法？為咗避免過度監管，同時確保 Robotaxi 以正確方式落地，你哋希望睇到咩樣嘅監管環境？

拉爾斯·莫拉維：當我哋討論監管，尤其係美國嘅監管環境時，已經取得咗好大進展。聯邦政府圍繞《聯邦機動車安全標準》（FMVSS）採取咗一啲非常積極嘅舉措，開始接受同推動專門設計嘅自動駕駛汽車。我哋非常感謝美國國家公路交通安全管理局（NHTSA）喺呢方面給予嘅支持。

至於各州嘅情況，新澤西州最近出現嘅某啲動向顯然有些令人失望。

但正如阿肖克喺開場發言中所講，我哋真正關注嘅係車輛本身嘅表現。最終，正係產品表現推動監管機構同公眾接受呢項技術。

我認為，最好嘅監管方式係同企業同創新者設定一個目標或任務，然後讓我哋自己搵解決方案。如果某啲監管規定已經未明確問題係咩，就預先規定咗解決方案，我哋確實唔會歡迎呢種做法。

特斯拉過去一直、未來亦會繼續用實際表現證明自己。呢亦會成為我哋未來繼續擴張嘅基礎。

Q3：接下來，投資者應該關注邊啲里程碑？達到邊啲里程碑後，你哋會加快 Robotaxi 車隊擴張或市場部署？係進一步嘅安全驗證，例如每發生一次意外對應嘅行駛里程，定係其他指標？另外，你哋會否考慮同網約車平台等第三方合作，以提高車輛利用率？定係繼續保持完全垂直整合？

馬斯克：我哋預計 Robotaxi 將採用垂直整合模式，好似特斯拉嘅其他業務一樣。我唔認為 Robotaxi 會面對需求不足嘅問題。佢嘅經濟性將極具吸引力，希望使用呢項服務嘅人會非常多，需求將超過我哋嘅服務能力。喺呢種情況下，顯然可以選擇垂直整合。

真正嘅問題只係我哋所謂嘅可靠性「多個 9 嘅進階」。即係話，為咗擴大規模，系統究竟需要達到幾多個 9 嘅可靠性？

理想情況下，我哋希望可靠性達到 99.999999%。我認為，Robotaxi 增長真正受到嘅唯一限制，就係不斷向更多個 9 嘅可靠性邁進。

Q4：外界一直喺討論 SpaceX 同特斯拉以某種方式合併嘅可能性。兩間公司目前已經進行咗大量合作。最終將兩間公司合併會否產生協同效應？隨著時間推移，咁樣做係唔係合理？

馬斯克：正如大家從兩間公司喺眾多領域開展嘅大量合作中看到那樣，雙方業務之間嘅重合度越來越高，尤其係 Terafab 項目。呢將係一個真正嘅巨型項目。

但好明顯，我哋唔能喺財報電話會上討論公司合併之類嘅事情。呢類事項必須按照適當程序進行。

布蘭登·爾哈特：我哋一直從同 SpaceX 嘅關係中受益。SpaceX 係一位非常出色嘅合作夥伴，我哋同佢哋開展咗許多對雙方都有利嘅交易。

今年早些時候，我哋透過一筆投資同一份框架協議进一步深化咗雙方關係。呢將使我哋能夠繼續同 SpaceX 合作推進埃隆剛才提到嘅項目，例如 Terafab 同數字 Optimus。

馬斯克：此外仲有好多其他合作。而家車入面已經接入咗 Grok（馬斯克旗下人工智能公司 xAI 嘅大模型），Grok 亦喺幫助驅動數字 Optimus。Starlink（SpaceX 嘅衛星互聯網服務）正喺集成到 Cybercab 中。未來，Starlink 仲將集成到我哋所有嘅汽車產品中，至少喺 Starlink 已經運營嘅市場會咁樣做。

因為喺 Robotaxi 嘅應用場景中，網絡必須覆蓋所有地方。

即使係喺矽谷，亦有許多地方嘅流動網絡信號非常差，有時甚至完全冇信號。我開車上班時，最初 10 分鐘到 15 分鐘實際上無法打電話，因為流動網絡連接實在太差。

我哋唔能讓 Robotaxi 進入呢啲缺少流動網絡連接嘅「百慕大三角」，然後被困喺嗰度。Starlink 能夠喺任何地方提供網絡連接，因此好重要。咁樣可以避免 Robotaxi 失聯。

如果人坐喺車入面，可能會想高效處理工作，亦可能希望獲得娛樂體驗。有咗 Starlink，人可以喺車內睇 4K 直播體育賽事，而且每 GB 數據嘅成本非常低。呢透過流動網絡實際上好難實現。

瓦伊巴夫·塔內賈：可以從 Cybercab 嘅乘坐體驗理解呢一點。喺一個唔需要專注於駕駛嘅世界入面，你坐喺車入面嘅全部時間都可以用於其他野，無論係參加電話會議、睇電影，定係從事其他活動。

因此，如果觀察 Cybercab 嘅設計，就會睇到佢配備咗一塊好大嘅屏幕。

我哋已經開始喺奧斯汀工廠內提供 Cybercab 乘坐體驗。喺唔遠嘅未來，消費者亦將能夠體驗到。等你真正體驗過之後，就會明白點解網絡連接會變得如此重要。

Q5：而家 Robotaxi 一直喺增加運營城市，但根據媒體報道，車輛數量似乎仍然只有幾十架，而唔係幾百架。點解唔先集中擴大奧斯汀或一兩座城市嘅規模？要想喺一座主要城市實現更高投放量，你哋仲要解決邊種問題？

阿肖克·埃盧斯瓦米：我哋之所以選擇向不同城市擴張，而唔係只集中發展一座城市，係因為我哋希望確保技術棧具備好強嘅通用性。

佢確實係一套通用系統。我哋只係希望向自己同外界證明，呢套系統能夠喺許多不同嘅城市運行，而且進入每座城市唔需要投入太多額外工作。正係我哋喺內部睇到嘅情況。

正如埃隆剛才提到嘅，增長率實際上已經呈現指數級上升，只係而家仲處於指數曲線嘅早期階段，因此外界好難理解呢種增長。

至於應該關注行駛里程定係車輛數量，由於 Robotaxi 車隊中嘅車輛基本上會持續行駛，而人類駕駛員可能每日只使用車輛幾個小時，所以兩者存在好大差異。

呢啲車輛喺絕大多數時間都會持續運營。這代表，即使車輛數量唔多，亦能產生大量行駛里程。因此，我哋更關注無人監督行駛里程，而唔係單純關注車輛數量。

我哋希望運營一支效率非常高嘅車隊。所有提高運營效率嘅工作，都會進一步提升車隊中每架車能夠貢獻嘅行駛里程。

瓦伊巴夫·塔內賈：我哋正喺嘗試擴大呢項業務，同时要確保解決其中可能存在嘅各種問題。呢啲問題唔僅涉及軟件，仲涉及運營，我哋亦努力處理。因此，我哋希望先擴大地理覆蓋範圍，喺真正大規模部署之前，透過規模較細、可控嘅車隊解決呢啲問題。

馬斯克：關於 Cybercab，有一點需要特別說明。因為 Cybercab 採用咗全新嘅車輛底盤，喺大量投放之前，我哋必須積累專門針對 Cybercab 嘅行駛數據。

Model 3、Model Y 以及其他車型已經有數百萬輛汽車喺道路上行駛，我哋擁有大量數據，但 Cybercab 仲未具備咗咁樣嘅數據基礎。

我哋必須先讓配備方向盤、加速踏板同制動踏板等裝置嘅 Cybercab 累計行駛里程，從而對 Cybercab 底盤完成系統校準。

當我哋對此建立足夠信心後，大家將會睇到，各城市中嘅 Cybercab 數量出現大幅增長。

拉爾斯·莫拉維：另一個需要指出嘅原因係，總擁有成本以及交通運輸監管環境，喺不同城市、不同州之間都唔同。而家並唔存在統一嘅聯邦監管框架。

我哋之所以逐個城市擴張，係為咗確保能夠分別滿足呢啲要求，並盡一切可能回應每座城市提出嘅具體要求。

我哋必須進入每一座城市，才能滿足當地要求。我哋會繼續咁樣做，滿足所有條件後再進一步擴大規模。

Q6：美國國家公路交通安全管理局正喺推動取消踏板同方向盤。呢啲會否成為限制 Cybercab 擴產嘅因素？喺聯邦層面，仲有無其他問題需要解決，才能真正釋放我哋擴大 Cybercab 規模嘅能力？

拉爾斯·莫拉維：簡單嘅回答係，冇。我哋同美國國家公路交通安全管理局，特別是同喬納森·莫里森（Jonathan Morrison）局長，保持著非常好嘅關係。

我認為，佢哋只係喺順應公眾需求，以及全世界都知道即將到嚟嘅趨勢。佢哋試圖走喺趨勢前面，確保制定適當嘅措施。

過去幾年入面，我哋一直坦白地向佢哋介紹我哋嘅計劃同正喺推進嘅工作。

我唔想講雙方完全步調一致，但我感覺我哋喺嗰度擁有一位合作夥伴，雙方正喺共同推進相關工作。

〔編者注：莫里森 2025 年 9 月 18 日經美國參議院確認出任 NHTSA 局長。佢曾於 2017 年至 2021 年任 NHTSA 總法律顧問，2025 年重返 NHTSA 之前，喺蘋果特別項目組（Special Projects Group，即蘋果造車項目）負責法務、監管、政府事務同政策〕

Q7：Starlink 已經集成到 Cybercab 中。Cybercab 能否倒轉成為 Starlink 移動服務嘅遠程熱點？另外，你哋預計幾時開始為 Tesla Semi 開發自動駕駛？考慮到卡車司機短缺，自動駕駛卡車似乎亦係一個好大嘅市場。

馬斯克：正如你提到嘅，而家確實存在非常嚴重嘅卡車司機短缺問題。願意駕駛卡車嘅人唔夠多，而卡車對於美國嘅貨物運輸至關重要。自動駕駛 Semi 對於解決卡車司機短缺問題將非常重要，亦有助於提高安全性。

對於仍然由卡車司機駕駛嘅 Tesla Semi 來說，自動駕駛功能亦會顯著提高駕駛安全性同舒適性。

由於 Tesla Semi 而家嘅數量仍然好少，即使到今年年底，佢喺特斯拉全部車輛中嘅佔比仍然會非常低，因此，將自動駕駛研發資源優先投入高銷量車型係合理嘅。

我哋會優先解決 Model 3、Model Y 同 Cybercab 嘅自動駕駛問題，真正使呢啲車型實現通用嘅、無人監督嘅自動駕駛。

我哋預計，Tesla Semi 嘅自動駕駛功能可能喺今年年底或明年初實現。我只係唔希望 Semi 項目分散我哋喺 Model 3、Model Y 同 Cybercab 上不斷提高安全可靠性、向更多個 9 邁進嘅注意力。

因此，喺未來大約六個月入面，自動駕駛 Semi 項目嘅優先級會稍微靠後。不過，佢明年肯定可以實現，並能夠趕上 Tesla Semi 大規模提高產量嘅進程。

至於通訊，我哋或許可以把特斯拉汽車中嘅 Starlink 終端當作流動網絡基站，或者某種網絡連接中繼塔，為地面上海機同任何希望使用 Wi-Fi 嘅人提供網絡中繼。

好明顯，固定式 Starlink 終端都可以提供類似服務。〔編者注：股東信稱，Tesla Semi 今年將喺內華達州新工廠投產〕

Q8：Optimus 將使用微處理器、微控制器、執行器同其他半導體。特斯拉可以從市場採購，可以自己設計後畀第三方晶圓廠製造，也可以自己設計同生產。對於呢啲零部件，你哋更傾向於採取點種方式？

馬斯克：Optimus 使用咗大量高度專用嘅電力電子設備同電路板。呢啲全部由特斯拉設計，但製造工作由供應商完成。

第四代 Optimus 將喺奧斯汀生產。與第三代產品相比，第四代 Optimus 嘅供應體系將實現更高程度嘅垂直整合。

我哋嘅目標係，第四代 Optimus 嘅產量比第三代產品高一個數量級。理想情況下，第三代 Optimus 嘅年產量為 100 萬台，第四代產品達到每年 1000 萬台。

當然，呢度需要附加所有相關嘅謹慎說明，因為擴大生產規模係一件極其困難嘅事情。

第四代 Optimus 將採用更高程度嘅垂直整合。因此，到時我哋可能會喺內部完成大量印刷電路板相關工作。

Q9：特斯拉是否仍計劃把 Hardware 3 升級至 Hardware 4，以便運行 V15 及未來版本嘅 FSD？現有車輛是否還會進行硬件升級？

馬斯克：我認為，升級所有低於 Hardware 4 嘅車輛最終會擁有經濟合理性，基本上係所有已經配備攝像頭嘅汽車。

如果連攝像頭系統都冇，所需要嘅改動可能太多。對於任何已經圍繞攝像頭系統設計嘅車輛，喺某個時間點進行升級，喺經濟上都係合理嘅。我認為，我哋希望直接將呢啲車輛升級到下一代 AI 計算板。

具體而言，可能有兩種選擇。我哋有一款升級版 AI4 芯片，相比 AI4 有一定幅度嘅提升，可能會喺明年年尾左右投入生產。然後係 AI5，我哋希望 AI5 亦能喺明年年尾左右實現量產。AI5 最初將應用於 Optimus。

而家，我哋喺 AI5 方面取得咗非常好嘅進展。特斯拉芯片團隊嘅工作非常非常出色，正以難以置信嘅速度推進 AI5 研發。

我對特斯拉 AI6 芯片嘅設計感到非常興奮。我認為，佢將成為全世界最優秀嘅邊緣計算芯片。如果有咗更優秀嘅芯片，我好想見佢，並與佢握手，因為 AI6 芯片實在太出色。

芯片方面嘅進展非常順利。再次感謝台積電、三星同美光給予我哋嘅支持。

Q10：喺目前呢輪資本開支週期中，投資速度喺多大程度上受到效率限制？如果繼續增加投資，會否導致資本效率下降？資本開支速度喺多大程度上決定咗特斯拉解除供應限制嘅進度？

馬斯克：我向特斯拉團隊提出嘅要求係，只要唔會造成過度浪費，我哋就應該盡可能快進行資本開支。

我哋唔係要追求資本投入嘅極致高效率，因為嗰會拖慢項目進度。我哋需要喺資本效率同時間之間取得平衡。

如果能夠讓項目更早完成，那麼資本效率略低係可以接受嘅，因為從公司嘅角度睇，呢實際會帶來更高嘅淨現值。

總體而言，我對而家嘅進展相當滿意。我哋正喺多個完全不同嘅領域，同時進行規模驚人嘅建設同產能擴張。

我認為，此前從來冇有一間公司以如此大嘅規模同時推進咗這麼多工作。或許當年亨利·福特擴大 Model T 生產時，喺相對規模上接近呢種情況；或者喺第二次世界大戰期間，企業曾經嘗試大規模擴充武器生產。

我認為，呢可能係第二次世界大戰以來，美國速度最快嘅一次工業規模擴張。

瓦伊巴夫·塔內賈：如果觀察我哋嘅資本開支，就會發現，所有支出都用於建設能夠投入生產嘅資產。我哋正喺擴大各種工廠，包括 Optimus 工廠、Cybercab 工廠、今年早些時候已經投產嘅磷酸鋰鐵電池工廠、Semi 工廠，甚至仲包括半導體晶圓廠。

此外，我哋仲計劃喺美國建設大規模太陽能製造能力，把美國嘅太陽能製造規模提高一個數量級。呢一點好重要，太陽能製造嘅重要性被低估咗。

呢啲工作都唔容易。要建設一座工廠，必須從頭開始。由於正喺進行嘅建設項目實在太多，特斯拉最終成為幾乎所有建設項目嘅總承包商。

我哋正喺以非常快嘅速度擴大規模。這代表我哋需要同時推進好多野。因此，我哋只能以人類所能達到嘅最快速度前進，確保呢啲項目能夠喺現實世界真正運轉起嚟。

馬斯克：我認為，我哋嘅資本開支效率高得好超出常規尺度。

Q11：喺可預見嘅未來，儲能業務喺多大程度上會受到供應限制？而家嘅儲能項目入面，有幾多用於公用事業削峰填谷，有幾多用於解決數據中心因用電量劇烈波動帶來嘅電能質量問題？數據中心同公用事業領域嘅需求分別處於點咩狀態？

馬斯克：呢唔僅僅係削峰填谷，仲包括電網平衡。電池包唔只喺短時間內啟用。對於平衡風力發電同太陽能發電產生嘅電力，電池係一種非常出色嘅工具。

未來，全球絕大多數能源都將透過太陽能同電池嘅組合生產。所有 Starlink 衛星亦係透過太陽能電池板同電池供電。你完全可以把地球想像成一顆巨大嘅衛星。

太陽所能提供嘅能源遠遠超過其他任何能源。我哋認為，電力限制將成為 AI 面對嘅一項重大問題。事實上，佢而家已經係一個重大問題。

僅僅係啟動 AI 電腦，就需要消耗大量電力。AI 算力嘅電力需求非常高，就算係超大型雲端計算公司，都好難為 AI 算力搵到足夠嘅電力並順利啟用相關設備。

此外，仲必須平抑電力波動。尤其係喺 AI 模型訓練過程入面，電力需求會喺極短時間內發生劇烈變化。

喺一次訓練過程中，耗電量可能會喺 100 毫秒內下降 70%。呢就非常需要反應迅速嘅先進電力電子設備，以平抑大幅度嘅電力變化，尤其係 AI 模型訓練過程中嘅變化。

呢就係 Space 為什麼喺数据中心購買大量 Megapack（特斯拉大型儲能產品）嘅原因。佢哋主要用於平抑訓練過程中出現嘅電力波動。電池仲可以幫助數據中心從電網獲得更多電力。

如果你話話公用事業公司，喺一年中用電狀況最糟糕嘅幾個小時或幾日入面，唔需要佢哋向你供電，因為電池可以承擔供電任務，那麼佢哋就容易向你提供電力接入。

事實上，提高美國能源產出總量嘅最佳方式，可能就係大規模使用電池。美國大約擁有 1.2 太瓦至 1.3 太瓦嘅發電能力，但平均用電功率只有大約 0.5 太瓦。即係話，美國嘅發電能力約為平均用電量嘅 2.5 倍。

這代表，僅僅係透過使用電池，美國嘅能源輸出量就有可能增加一倍。因此，我哋認為，未來市場對 Megapack 嘅需求將會非常高。

7月9日，小鹏集团董事长何小鹏亲自完成Robotaxi内测首单，标志该项目自2025年11月官宣起，仅8个月便跑通了常态化道路测试（2026年1月）、量产车下线（5月18日）及全链路内测。

在此之前，小鹏曾召开Robotaxi业务首次全员会议，宣布该业务正式进入员工内测阶段。明确Robotaxi将作为集团物理AI与“机器人汽车”战略的核心载体。

无独有偶，在今年的北京国际车展上，吉利汽车正式发布了中国第一款完全自己从头打造的Robotaxi无人出租车原型车——Eva Cab。这台车完全基于自动驾驶运营场景设计，集成了吉利的整车制造能力、千里浩瀚G-ASD L4智驾系统，以及曹操出行十年的运营数据与服务洞察。吉利宣告，已全面进入全域AI 2.0时代。

新势力的高调入局、传统车企的全力押注——种种迹象表明，在国内，Robotaxi赛道正从"百度领跑"的单极格局，走向多方角力的群雄竞逐。

而这个时间节点，距离震动业界的萝卜快跑武汉"趴窝"事件，仅仅过去了三个多月。

2026年3月31日晚8时，在武汉城市三环线、太子湖大桥、白沙洲大桥等主要交通干道，近百辆萝卜快跑无人驾驶出租车，几乎同时减速，然后直挺挺地停驻在车流之中。

这场突如其来的集体“瘫痪”，持续了近两小时，导致部分路段出现严重拥堵，也引发了公众对无人驾驶技术可靠性的广泛质疑。

更戏剧性的一点是，就在前一天，百度刚刚宣布，萝卜快跑将在迪拜启动全无人驾驶商业化运营。

这并非萝卜快跑首次出现类似问题。2024年7月，亦有车辆在晚高峰时段异常停车——但此次事件的特殊性在于并非单车故障，而是区域范围内运营车辆的集体响应。

萝卜快跑是百度在智能驾驶与出行服务领域的核心业务，也是百度将AI技术商业化的重要方向。据官方消息，萝卜快跑由百度四层全栈AI架构（云基础设施、深度学习框架、大模型、应用程序）提供支撑。

截至2026年6月，萝卜快跑布局已扩展至全球‌27座城市‌，其中，自2025年2月起，萝卜快跑在中国内地所有运营城市，包括北京、上海、深圳、武汉、成都、重庆、海口、三亚等，实现100%的全无人驾驶运营（即车内无安全员），即L4级别，并在多个城市获得收费许可。

这标志着自动驾驶技术从“有人监督”向“完全无人”的关键突破。

国际市场方面，其已先后进入迪拜、阿布扎比、伦敦、圣加仑、首尔等城市。6月16日，萝卜快跑又正式获得瑞士L4级自动驾驶出租车运营许可。这是中国自动驾驶技术首次在欧洲取得规模化商业运营资质。

根据公开数据，萝卜快跑累计向公众提供的无人驾驶出行服务订单已超2200万单。

其技术水平也得到了业界的广泛认可——自动驾驶行业分析机构Guidehouse Insights在2025年第四季度发布的研究报告中，将百度Apollo列为全球自动驾驶领域的两大领导者之一，另一位则是谷歌旗下的Waymo。

但也正是因此，才让此次事件变得尤其令人关注。

根据武汉交警后续发布的公告，初步判断此次事件为“系统故障”所致。

而根据《财闻》公众号采访同济大学汽车与能源学院教授、中国汽车工程研究院智能网联首席专家朱西产的报道，朱教授的分析称：“3月31日晚上的故障，很可能就是通讯网络出了问题。车和云之间的联系断了，系统触发安全机制，网络故障区域内的车辆集体停车。这不是‘车坏了’，而是‘网断了’。”

也就是说，与车辆误识别障碍物、错误预测行人轨迹等“感知决策类事故”性质不同，此次更多是通信与系统设计层面的问题。

但这恰恰是自动驾驶一直以来最令人担忧的部分。

当前多数自动驾驶系统采用“车-云协同”模式，其中，云端负责高精地图实时更新、交通全局调度、复杂场景计算支援。一旦云端这个“大脑”与车辆的“神经”联结中断，汽车就会失去指挥。

正常情况下，一旦遇到类似情况，人工可以直接进行接管，从而保证驾驶的连续性与安全性；但在无人驾驶的场景下，车辆就只能遵循已有的设定执行——此次事件中，萝卜快跑就直接停在了主干道上，所幸根据后续通告，此次事件未造成人员受伤。

但这也暴露出萝卜快跑对此类事件的应急预案准备不够充分。

对此，朱西产表示，合理的做法是，触发安全机制后，应该是把车驶离城市快速路，开到路边停车，并提醒乘客下车，到安全区域后再联系监管平台。

但我们能因此全盘否定自动驾驶技术吗？显然不能。此次事件中车辆依据预设安全逻辑主动停车而非失控，恰是安全设计的体现。朱西产亦指出，示范区运营的本质就是在真实场景中暴露问题、推动成熟。

毫无争议的一点是，自动驾驶是人工智能最具社会价值与经济意义的应用领域之一，也是数字经济发展与产业升级的重要引擎。不能因局部性、阶段性问题而否定技术探索的整体方向。

事实上，我国在自动驾驶测试评价体系、技术标准、认证监管等方面已在持续推进——从封闭场地到开放道路，从单车智能到车路协同，相关安全框架与测评方法正在不断完善。

数据显示，在武汉等部分城市，萝卜快跑已实现核心区域的全覆盖，用户满意度高达4.9分（满分5分），这反映出自动驾驶服务在真实场景中的接受度仍在持续提升。

而在海外，Waymo和特斯拉的实践已验证了Robotaxi规模化运营的可行性——Waymo在2万辆车队规模即可实现盈亏平衡，特斯拉无人出租车在100万辆规模下每英里成本可降至0.2美元。

从技术路径看，Waymo的高冗余传感器路线和特斯拉的视觉大模型路线正在相互借鉴融合，L4公司通过持续优化降低成本，车企则通过大模型提升安全性，最终将共同迈向完全无人驾驶——值得注意的是，特斯拉V14.2版本实测显示其"危险接管里程"已提升15倍，展现出技术迭代的加速态势。

总体而言，无人驾驶已进入商业化前夜，未来三年将决定各企业在新时代的竞争地位，而像萝卜快跑这样的先行者，其经验与教训，都将成为推动整个行业向前发展的重要基石。

从全球范围看，自动驾驶出租车（Robotaxi）已成为科技企业和传统车企竞相布局的赛道。而武汉3·31事件之后的三个多月里，行业非但未因此降温，反而迎来了更为密集的商业化推进。

在中国市场，小鹏的快速推进最具代表性。从2025年11月官宣到2026年7月完成内测首单，仅用8个月便走完了从立项到全链路内测的全过程。

吉利Eva Cab则走了一条不同于百度基于量产车改装、小鹏从自有车型切入的路径——Eva Cab是一款从零开始为无人驾驶运营场景设计的原生Robotaxi。这标志着主机厂正从"提供车辆"向"定义出行空间"转型。

萝卜快跑也在事件后持续推进。截至2026年7月，萝卜快跑在中国香港的测试项目已运营超过一年，63辆持牌自动驾驶车辆累计行驶24万公里，测试范围已扩展至九龙东和南区。在阿布扎比，其已正式启动面向公众的全无人驾驶商业化运营，并计划2026年部署数百台无人车。

小马智行、文远知行等头部玩家也在各自深耕的城市扩大运营规模，中国Robotaxi市场正从"百度独大"走向"多方竞合"的新阶段。

此外，中国企业在政府支持、路网协同和高复杂场景应对方面积累了独特经验。特别是中国城市道路的复杂情况，包括大量电动车、行人和非机动车混行的场景的经验，有利于无人驾驶技术的数据积累。

总的来看，无人驾驶路径带来的变革正在重塑整个汽车与出行产业。

一方面，产业合作已成主流，车企与科技公司形成“制造+技术”共生关系。比如丰田投资小马智行，就是典型的车企提供制造与车辆平台，科技公司贡献自动驾驶技术，期望形成优势互补。

另一方面，车辆形态也被重新定义。为共享出行而生的定制化Robotaxi开始出现，如百度的Apollo颐驰06以及Eva Cab就取消了方向盘甚至是驾驶座，以图重构车内空间与用户体验。

对普通用户与社会而言，Robotaxi带来的最大影响是出行成本的大幅降低：规模化后，其价格可能比当前网约车低30%-50%。而公众担忧的"司机被迫下岗"问题，从现阶段看，纯驾驶岗位的确可能减少，但运维、远程监控、客服等新岗位则在同步涌现。

但无论从何种角度看，安全与信任仍是最关键的核心。尽管理论上无人驾驶能减少人为事故，但3·31事件表明，在技术完全成熟前，任何一次系统性故障都可能动摇公众的信心。

3·31事件如同一面镜子，将长期照出L4自动驾驶商业化必须直面的系统性挑战——通信网络的可靠性、应急预案的完备性、监管框架的适应性；也映照出行业已从实验室走向真实道路、从示范运营迈向规模化商业落地的不可逆趋势。

而小鹏、吉利等新玩家的加速入局，则给出了另一个维度的信号：这个赛道不会因为一次挫折而停滞，反而在阵痛中迎来了更丰富的参与者、更多元的路径和更激烈的竞争。

7月2日，广汽集团发布6月产销公告。数据显示，上半年，全集团汽车销量77.31万辆，同比增长2.35%，其中节能与新能源车销量占比达62.82%。自主品牌出口量保持翻倍增速，上半年累计达12.15万辆，同比增132%，已接近去年全年水平。

今年以来，随着“番禺行动”系列改革积极效果传导至市场端，广汽集团自主品牌销量加速攀升。上半年，广汽自主品牌累计销量34.60万辆，同比增长35.69%。今年7月，广汽集团累计产销量将突破3000万辆里程碑，标志着规模化发展正式迈入新阶段。

自主与合资齐头并进

新能源、高端化成果加速兑现

今年以来，广汽集团在新能源和高端化赛道全面提速。

昊铂埃安BU积极推进改革及品牌焕新，上半年累计销量达18.16万辆，同比增长67.08%，C端用户占比超80%。面向大众市场的埃安持续推出多款热销车型，其中埃安i60月销持续破万，跻身A级双动力SUV销量前列；埃安N60自4月底上市以来，已交付超万辆。与此同时，锚定新豪华赛道的昊铂持续发力，20万级豪华运动SUV的昊铂S600已于6月12日上市。

埃安N60

昊铂S600

广汽传祺上半年销量达16.44万辆，同比增长12.36%。MPV家族累计销量突破88万辆，6月同比劲增36%。广汽传祺显著加快产品升级速度，传祺M6 MAX、向往E8 PHEV及向往M8 PHEV L已于上半年相继上市，并计划于今年第三季度推出首款硬派越野SUV传祺“越7”。渠道与服务同步升级，5月，广汽传祺与中升集团战略合作落地，首批12家全新授权经销门店在核心城市同步揭牌，预计全年将落地超30家。

传祺向往M8 PHEV L

今年6月，广汽集团向高端化突破取得重要进展，与华为乾崑联合打造全新高端品牌——启境，首款车型启境GT7于6月26日在杭州正式上市，24小时大定订单突破5200辆，市场反响热烈，新车将于7月初开启全国交付。启境GT7搭载华为六大汽车智能化解决方案，拥有最强全栈技术，将首批搭载华为乾崑智驾ADS 5。根据规划，启境还将于下半年推出新一代智能阔五座SUV启境GX7，双车矩阵战略正式落地，品牌发展全面驶入快车道。

启境GT7

合资板块中，广汽丰田上半年销量35.6万辆，同比增长3.29%，并于3月、4月、5月连续三个月居合资车企销量榜首。本土化开发车型持续发力，铂智3X上市以来累计终端销量已突破11万辆，刷新了合资新能源车型达成该纪录的速度，连续10个月领跑合资纯电市场，成为上半年合资纯电销冠；另一款本土开发车型铂智7上市3个月交付已破万。全球战略车型表现稳健，赛那与凯美瑞上半年于各自细分市场均居销量前列。此外，近期广汽丰田第1000万辆量产车下线，晋升千万级车企行列。

广汽丰田铂智3X

广汽本田6月销量环比增长55.65%，第二季度连续2个月实现双位数环比正增长。雅阁在《2025年度中国汽车保值率报告》中蝉联合资中型车保值率冠军。新皓影于5月20日焕新上市，作为本田首款搭载与华为云联合打造智能云车机的车型，精准回应家庭用户全场景出行需求。面向2027年，广汽本田将推出3款全新产品，覆盖燃油、混动及新能源三大领域，展现出强劲的体系韧性与转型活力。

借势2026年新能源汽车下乡，广汽8款车型入选活动名录，渠道加速下沉，年内计划落成1000家县域授权门店，截至6月底，已完成600家网点认定。产品向上、渠道向下，广汽正以体系实力打开增量新空间。

广汽本田皓影

上半年出口同比132%

全球五大区跨越式增长

随着体系化出海战略持续推进，海外市场已成为广汽集团爆发力最强的板块。1-6月，广汽自主品牌出口量达12.15万辆，同比大增132%。

广汽“体系化出海”加速全球布局

上半年，广汽在海外市场销量集中爆发，包括美洲、亚太、中东、非洲及欧洲在内的五大区域同步实现高增。其中，在美洲的墨西哥，埃安UT和ES分别位居B/C级纯电市场前三；在玻利维亚，传祺向往S7、埃安UT拿下年度插混、纯电车型销量冠军，广汽接连多月蝉联当地中国品牌乘用车市场销冠；在巴西和哥伦比亚，广汽6月销量环比分别大增1129%、804%，乌拉圭则同比增长254%。

在亚太区域，广汽市占率在多地刷新纪录。其中，在中国香港1-5月电动车市占率突破11%；泰国市场6月销量环比劲增207%；新加坡市场6月环比增长77%。非洲及中东市场也在稳步突围。继1-2月中东区域销量同比增长282%后，5月影速拿下黎巴嫩B级SUV销冠，埃塞俄比亚6月销量环比大增510%。

在备受关注的欧洲市场，广汽已建立从本地化生产到核心市场终端销售的全链路运营体系，上半年进入意、英、西等国，拿下6月希腊中国品牌纯电销量第二，同时埃安UT已于奥地利KD工厂投产。

具身智能、飞行汽车、Robotaxi

前沿科技版图加快商业化突破

广汽集团以前瞻布局加快培育新质生产力，上半年在低空经济、具身智能和自动驾驶三大赛道均实现商业化突破。

6月25日，广汽孵化的具身智能机器人企业慧仑科技与广汽动力BU签署战略协议，将联合打造万台人形机器人标准化产线，并率先在广汽动力BU工厂内部署机器人产品，加速具身智能在工业场景落地。同时，慧仑科技自研的第四代具身智能人形机器人GoMate Mini已具备全天候、全场景智慧物业运维能力，上半年在多个社区及产业园区实现常态化运营与小规模商业化落地，已获近千万元人民币订单，规划2027年实现规模化量产。

广汽第四代具身机器人GoMate Mini在广州地铁站内投入服务

在飞行汽车方面，高域科技广州黄埔工厂已于5月底落成，并下线首台飞行汽车GOVY AirCab。目前，该机型已获近2000架意向订单，适航取证稳步推进，预计年底完成认证交付。

广汽高域多旋翼飞行汽车GOVY AirCab

在Robotaxi领域，广汽与滴滴自动驾驶联合打造的Robotaxi R2已于今年1月正式交付，并先后获广州和北京两地智能网联汽车道路测试牌照，正稳步推进复杂场景下的L4级自动驾驶验证工作。旗下如祺出行“Robotaxi+”战略正推进自动驾驶规模化运营，运营的Robotaxi车辆超300辆，未来计划组建超万辆车队，覆盖100个核心城市。

下半年，广汽集团将推出硬派越野SUV传祺“越7”、面向年轻市场的埃安智能轿跑、新一代智能阔五座SUV启境GX7等重磅车型，产品阵容再升级，并将以更强劲的产品与体系实力，为用户创造超越期待的移动出行价值。

【导语：2026年，Robotaxi商业化落地步伐正在加速。日前，自动驾驶科技公司、整车企业、出行平台集中发布新的Robotaxi发展计划。小马智行拓展全球车队规模；文远智行携手联想开启五年规划；小鹏集团聚焦物理AI。以上种种，无一不是围绕Robotaxi技术研发、成本控制、牌照资源、市场份额展开的全方位竞争。】

作者：蔡言

在一众无人驾驶玩家中，吉利旗下的曹操出行发布了中国首款原生Robotaxi，采用了完全不同于其他企业的技术路线。那么，当2026年Robotaxi行业跨过技术验证期，转向规模化和商业化竞争，曹操出行这块“原生量产”的拼图，能否悄然重构Robotaxi的中国路径？

何为原生Robotaxi ？

原生Robotaxi，即车辆从立项、研发、落地都是围绕“智能无人驾驶”开发的，车辆电子电气架构、底盘控制、座舱布局、智驾算法优先满足无人化、智能化、安全性与空间体验。

曹操出行Robotaxi很可能会改变Robotaxi竞争格局，主要源于“新模式”对“旧模式”的冲击。过去很长一段时间里，无论是全球无人驾驶标杆Waymo，还是国内早期玩家， Robotaxi大多采用“量产车+后装套件”模式，即“在民用量产车的基础上，增加一堆硬件，从而满足无人驾驶条件”。

这套“后装模式”，之所以行得通，除了硬件水平外，主要原因在于：一方面早期Robotaxi赛道资本不计成本地大撒币，试验状态下能够支持企业高成本运行；另一方面则是Waymo、萝卜已经用这套模式跑通了基础商业闭环，这些企业们最先考虑的事情，往往是拓展运营规模，借助规模化摊薄单车成本。

实际上，无论是Waymo还是国内头部无人驾驶企业，即便行业不断刷新算法能力和路测数据，但始终未能跨越规模化运营盈利的门槛。导致“后装模式”盈利困难的痛点在于，一方面为改装成本高、可靠性差、能耗高、车辆总拥有成本（TCO）居高不下；另一方面则是技术与运营脱节，自动驾驶科技公司、整车企业、出行平台作为Robotaxi赛道的三大势力，每种企业都有独特的优势，同时也会出现科技公司缺车、车企缺运营、平台缺技术对应的难题。

因此，尽管2024年之后，Robotaxi玩家都在加快商业化，努力实现自身造血，但最亮眼的战报，往往局限于“片区盈利，亏损收窄，车队扩张，整车成本下降，牌照增加”。

曹操出行原生Robotaxi的模式优越性在于，“智能定制车+智能驾驶技术+运营平台”才是Robotaxi商业闭环的正确打开方式。或者说未来的Robotaxi商业化模式，与其科技公司、整车企业、出行平台各自为战，不如多方合作发挥各自优势。

曹操出行Robotaxi有什么优势？

吉利和曹操联合打造Robotaxi——Eva Cab，最值得关注的地方，并非产品层面，而是商业化模式，曹操出行作为吉利的出行平台，同时也是吉利Robotaxi最重要的商业化载体，曹操出行Robotaxi不必从零开始搭建模式，直接拥有“智能定制车+智能驾驶技术+运营平台”全方位能力。

在定制车和智能驾驶技术层面，Eva Cab成为了吉利在整车和AI技术体系的能力体现。

Eva Cab在设计之初便采用转向、制动、电源、计算平台多重冗余配置，并且配备传感器自清洁、线控双冗余等系统。在座舱空间方面，由于Eva Cab取消了方向盘和副驾，车内乘员空间使用空间大幅提升，车门则采用对开式电滑门布局，让车辆所有设计都围绕乘员舒适性体验服务。

为了适应中国城市高密度、高复杂度的运营环境，Eva Cab搭载全球首个“量子AI电子电气架构EEA 4.0”，为AI时代重新设计整车电气系统。量子加密技术覆盖蓝牙钥匙、远程车控、OTA升级、数据隐私等全链路。

在核心算力方面，Eva Cab在英伟达SuperChip+Thur U+高通骁龙8797三块旗舰芯片的堆料下，车端算力超过3000TOPS，这也让WAM世界行为模型的参数量提升7倍、推理帧率提升3倍，决策响应效率是人类驾驶员的3倍之多。

在感知硬件配置上，Eva Cab原型车配备了43颗感知元件，最值得关注的是，全球首个2160线数字化激光雷达上车，重点是吉利称其为“可量产千线级”超长距数字化激光雷达，每秒2592万点、最远600米探测距离，主要满足Robotaxi对于“冗余、稳定、应对复杂路况”的多维度需求。

软件层面，Eva Cab以“千里浩瀚G-ASD L4”量产方案为技术底座，融合曹操出行十年出行积累的调度算法、亿级用户基础和超19亿次海量真实出行数据，让算法不再单纯依赖仿真，而是扎根真实的道路场景，持续反哺Robotaxi自动驾驶体验。

将角度落到更加宏大的“生态”叙事，此次曹操出行和吉利深度打造的，并不局限于单独一辆车，而是一套可以持续升级的系统。

吉利把这套布局描述为“地面+低空+低轨”的天地一体化出行解决方案，Robotaxi通过接入低轨卫星、AI数字底盘、云端大脑以及曹操出行“绿色智能通行岛”，打造技术端到运营端的落地链条。

回归最现实的商业化落地角度，无论是整车、智能技术还是运营、可持续生态层面，吉利曹操围绕Robotaxi所有的动作，值得关注的最核心原因——成本控制。

吉利负责整车供应链成本控制，曹操出行负责运营经验，能够无缝迁移至Robotaxi场景，成为Robotaxi大规模铺开的经济基石。

原生Robotaxi让整车设计寿命达到普通私家车2~3倍，关键部件寿命远超普通乘用车，保养周期大幅延长，从源头降低车辆TCO（拥车成本）。运营体系的复用，无论在营销层面，还是数据反哺层面，都能大幅降低费用成本。

根据弗若斯特沙利文数据，曹操出行定制车相比民用纯电车，平均TCO降低约36.4%，目前传统改装Robotaxi的每公里运营成本普遍在2.5元至4元之间。

说到底，曹操出行这次联手吉利深度定制原生Robotaxi，一开始就瞄准了“车辆即服务”的终局，而非科技公司的炫技式叙事。如此一来，硬件回归最本质的服务载体，软件成为无人驾驶的驱动数据，将技术持有、运营、维保到退役，全部纳入一条精密运转的资本回报系统。

Robotaxi商业化提速

按曹操出行Robotaxi“三步走”战略，目前已经从初期的技术验证和小规模测试运营，来到了探索无人化运营阶段，曹操出行已经在杭州、苏州等城市试运营Robotaxi超过一年。

今年将逐步开拓更多城市，计划在中国香港及其他国际市场探索Robotaxi业务，2027年吉利量产面向曹操出行的定制版Eva Cab，到2030年，累计投放10万辆原生定制的Robotaxi全面铺向全球各地。

据灼鼎咨询报告预计，2030年中国Robotaxi市场规模有望达到515亿元。国内和海外玩家都很有默契的在今年，加快Robotaxi商业化速度。Waymo计划2026年在美国及国际市场20个城市扩大测试并推出新的商业服务，目标为2026年底前实现美国市场每周超过100万次付费Robotaxi出行。小马智行宣布2026年计划在全球20余城部署超过3000辆Robotaxi；文远知行与联想携手计划未来五年内全球范围内部署20万辆Robotaxi。

此外，3月小鹏汽车宣布成立Robotaxi独立业务部，计划在年内推出3款全栈自研车型，并于下半年开启载客示范运营。特斯拉于4月正式投产Cybercab，加之5月初，特斯拉监管版全自动驾驶车队累计行驶总里程突破100亿英里，这均意味着特斯拉Robotaxi业务加速落地。

点评

2026年，“商业化、规模化、体系化”作为Robotaxi行业的关键词，竞争模式已经从算法的单点突破，进入到了“制造+运营+算法”的全产业体系对决，未来拥有硬核技术实力+精细化成本管理+全域运营体系的玩家，才是真正的市场主流选手。

（本文系《禾颜阅车》原创，未经授权，不得转载）

编译 | 莫 莉

编辑 | 黄大路

设计 | 甄尤美

来源 | Reuters，Financial Times等

Uber这次打算真金白银押注Robotaxi了。

4月15日，路透社援引英国《金融时报》报道称，Uber已承诺投入超过100亿美元，用于采购数千辆自动驾驶汽车，并入股相关技术开发商。

此举标志着Uber告别轻资产的零工经济（gig economy）商业模式，以应对自动驾驶出租车带来的行业冲击。

按测算，约有25亿美元会流向股权投资，超过75亿美元会用于车队采购，而且相关协议大多都附带部署里程碑条件。

针对这一情况，Uber未立即回应媒体的置评请求。

不过，把Uber过去两个月连续公开的动作放在一起看，这并不是一条孤立消息，更像是公司已经启动的一轮战略转向。

从今年2月起，Uber一边对外强调自动驾驶是核心战略优先项，一边开始把钱投向整车、股权和配套基础设施。

其先后披露了对Rivian、Lucid、Waabi等伙伴的资金承诺，也宣布建设自动驾驶充电枢纽、推出面向合作方的商业化工具包，并把Robotaxi的扩张时间表写得更具体。

按Uber和合作方近期公开的信息，到2026年底，平台计划促成Robotaxi行程覆盖全球最多15座城市；到2028年，搭载英伟达方案的无人车队目标是在至少28座城市落地。

从轻资产转型

这次之所以引发关注，关键不只在100亿美元本身，而在于涉及到Uber过去几年一直强调的边界。

2020年，Uber出售自研自动驾驶业务，转向更谨慎的合作路线，以降低现金消耗，把重心拉回出行和配送主业。现在，Uber并没有重回自研自动驾驶的老路，但也不再只是做撮合订单的平台，而是开始提前为车辆、股权和后续运营投入真金白银。

2月4日Uber财报公布后，路透社评价称，公司正在为这种资本密集、仍处早期阶段的自动驾驶战略背书。

Uber当时表示，会向车辆合作伙伴投入资本，以确保早期供应并加快部署，同时还会与银行和私募股权机构合作，为大部分自动驾驶车队融资。

Uber首席执行官达拉·科斯罗萨西（Dara Khosrowshahi）对外表示，公司愿意先把资本拿出来，先锁定后续供给。

这意味着，Uber仍想做平台，但它也在调整做法：如果想在Robotaxi时代继续掌握用户入口，只靠撮合订单已经不够了。

Uber判断，Robotaxi不是简单替代现有出行需求，而是会通过增加运力、改善可靠性和压低价格，把整个出行市场再做大。

其公开目标是，到2026年底在全球最多15座城市促成Robotaxi出行，并准备把业务扩展到西班牙马德里、香港、休斯敦和瑞士苏黎世。

达拉·科斯罗萨西还表示，在Uber平台上运营的自动驾驶车辆，利用率和接驾速度都优于独立运营的Robotaxi服务，这也是它判断大平台能做出更好经济账的原因。

随后，Uber把这套合作思路进一步落到具体业务上，正式推出Uber Autonomous Solutions。从公司自己的定义看，这不是一个单独产品，而是一整套给自动驾驶合作方用的商业化能力，分成基础设施、用户体验和车队运营三块。

Uber提到，它能提供的不只是需求入口，还包括数据采集、地图、监管支持、融资、车内交互、远程协助、现场运维和保险。

达拉·科斯罗萨西表示，自动驾驶的创新进展很快，但真正有意义的商业化会慢得多，Uber要做的，是把自己过去十多年在按键叫车、履约、客服和运营上的积累，对外输出给合作伙伴。

这条消息之所以值得汽车行业重视，在于Uber的角色正在变化。它不再只是等合作方把车开上路后再接入平台，而是开始向前一步，介入车辆采购、资金投入和运营准备。

路透社还特别提到，经历多年跳票之后，近几个月市场对无人驾驶出租车的兴趣重新回升，而人工智能能力进步以及科技公司、平台和车企之间的新合作，被认为有机会更快处理复杂路况，也有机会摊薄高成本。

Uber也正是在这一轮行业回暖中，重新调整自己在Robotaxi产业链里的位置。

向车和股权砸钱

如果把Uber近几个月披露的合作一项项拆开看，这笔钱具体会投向哪里，其实已经越来越清楚。

最重的一笔，是3月19日公布的Rivian合作。

据报道，Uber将向Rivian最多投资12.5亿美元，首笔先投3亿美元，剩余资金到2031年前分阶段支付，前提是Rivian达到自动驾驶里程碑。

作为交换，Uber将从2028年起部署1万辆全自动Rivian R2 SUV作为Robotaxi，且在2030年还拥有再买4万辆的选择权。

对Rivian来说，这笔钱既是研发自动驾驶路线图所需的资金，也带来一个大体量商业客户。对Uber来说，它提前锁住的是一批能规模化交付的整车供给。

另一笔钱，是Lucid与Nuro这条线。今年1月5日，Lucid、Nuro和Uber在拉斯维加斯发布了一款接近量产状态的Robotaxi，为当年晚些时候的商业落地做准备。

4月14日，Lucid进一步宣布，Uber已把采购承诺从此前水平提高到至少3.5万辆Lucid车辆，覆盖Lucid Gravity和未来的中型平台车型，同时再追加2亿美元投资，使其对Lucid的总投资升至5亿美元。

Lucid还披露，下一代Robotaxi服务计划在2026年晚些时候于旧金山湾区启动，Nuro主导的道路测试已在2025年12月启动，Lucid在今年2月完成了全部测试车辆交付。

在这笔Lucid合作里，Uber看中的不只是Lucid的车型本身，也看重这套车在采购、制造和后续运营上的成本空间。

Lucid提到，未来的中型平台起售价计划压到5万美元以下。达拉·科斯罗萨西表示，Lucid未来的中型平台，让Uber看到了更清晰的单车经济性路径。

这意味着，Uber现在挑伙伴，不只是看谁的自动驾驶能跑，还看谁的整车平台能把采购、制造和运营成本压下来。

Lucid时任临时首席执行官Marc Winterhoff也在公告中提到，Lucid中型平台要解决的是成本效率、制造简化和面向下一代出行网络的用户体验。

再往前看，Uber对软件公司的下注也早已开始。

自动驾驶公司Waabi在1月完成10亿美元融资并与Uber达成合作，达拉·科斯罗萨西当时表示，Waabi未来生产的首批2.5万辆乘用车，将独家部署在Uber平台。

Waabi的公告称，这笔合作包含来自Uber的额外里程碑式资本支持，目标是在未来部署2.5万辆或更多搭载Waabi Driver的Robotaxi。

从已披露的合作看，Uber的资金已经同时投向整车企业和自动驾驶技术公司，而不是只押注其中一端。多项合作也都设置了量产、验证或部署节点。

平台不只管接单

从最近公开的信息看，Uber现在补的不只是车源和合作伙伴，还包括Robotaxi落地后每天怎么跑、怎么充电、怎么运营。

目前最成熟的案例，还是Waymo和Motional。

Waymo官网和Uber官网都显示，奥斯汀和亚特兰大的乘客已经可以只通过UberApp匹配Waymo车辆，车型为纯电Jaguar I-PACE。

乘客发起UberX、Uber Comfort等订单后，有机会被系统匹配到Waymo，不需要额外下载另一个应用。

另一边，Uber与Motional在3月13日宣布，拉斯维加斯用户已经可以在指定区域通过UberApp叫到Motional IONIQ 5 Robotaxi，涵盖拉斯维加斯大道沿线的多个上车点。

Uber强调，用户如果被匹配到自动驾驶车辆，既可以接受，也可以改回普通订单。

国际化扩张也在同步推进。今年2月10日，Uber与百度将在迪拜上线自动驾驶打车服务，首批覆盖迪拜朱美拉部分区域，乘客可以在UberApp里选择Autonomous选项，或在UberX、Uber Comfort订单中被匹配到Apollo Go车辆。

到了3月31日，Uber又宣布与WeRide在迪拜启动全无人、可收费的Robotaxi运营。此外，其与WeRide的合作已支撑阿布扎比、迪拜和利雅得的商业部署。

除中东市场外，欧洲、美国西海岸和日本的新项目也在陆续推进。

4月8日，Volkswagen集团旗下MOIA America宣布，已在洛杉矶启动自动驾驶ID. Buzz上路验证，测试阶段车队会扩至100多辆，计划在2026年晚些时候通过Uber平台向乘客开放，并在后续扩至数千辆。

今年三月中旬，Uber与英伟达宣布，搭载英伟达软件方案的L4级的Robotaxi将于2027年上半年先在洛杉矶和旧金山上线，到2028年扩至全球28座城市。

此外，日产、Uber和Wayve在3月12日宣布，三方计划在2026年晚些时候于东京启动Robotaxi试点，首阶段仍会配备受训安全员，并通过日本本地持牌出租车伙伴运营。

这解释了为什么Uber在2月18日又单独宣布，要拿出超过1亿美元建设自动驾驶充电枢纽。

据了解，这笔钱将先投向湾区、洛杉矶和达拉斯，用于建设自动驾驶车队的日常充电站和中途补能点。同时，Uber还会和EVgo、Electra、Ionity等运营商签“利用率保障协议”，在多个城市新增数百个充电桩。

对于车企和自动驾驶技术公司来说，车辆能不能上路只是第一步，车队能不能稳定充电、持续周转、把运营效率跑出来，是另一道更现实的门槛。Uber现在也在补这部分能力。

放在一起看，Uber这笔100亿美元承诺并不是一句空话。

过去两个月，公司已经围绕整车采购、股权投资、运营落地和充电网络连续出手。按路透社的说法，Uber正把自己定位为多个Robotaxi运营商的合作平台。对它来说，接下来要做的，不只是把更多自动驾驶车辆接入平台，而是更早介入车辆、资金和运营环节。