The Zhitong Finance App learned that after the US stock market on October 20 (Tuesday) EST, SpaceX (SPCX.US) handed over its second financial report after listing. According to financial reports, SpaceX's third-quarter revenue reached 14.575 billion US dollars, far exceeding analysts' general expectations of 12.72 billion US dollars, an increase of 161% from 5.576 billion US dollars in the same period last year, and an increase of about 87% from 7.814 billion US dollars in the second quarter.
Operating profit reached US$2,791 billion, compared with an operating loss of US$273 million and an operating loss of US$143 million in the second quarter; net profit reached US$4.404 billion, far exceeding analysts' general expectations of US$1.57 billion to US$1.88 billion, a significant improvement over the net loss of US$1,054 million for the same period last year, turning a loss into a profit.
According to US GAAP, SpaceX's diluted earnings per share for the third quarter were $0.35, which also far exceeded analysts' general expectations of $0.15. The loss was $0.36 per share for the same period last year. According to non-US GAAP calculations, SpaceX's adjusted profit before interest, tax, depreciation and amortization (EBITDA) for the third quarter reached US$8.809 billion, up 306% from US$2,169 billion in the same period last year, and about 149% from US$3.538 billion in the second quarter, and far exceeded analysts' general expectations of US$6.59 billion.
The capital expenditure that the market is focusing on reached $19.59 billion in the third quarter, in line with analysts' general expectations of $19.6 billion.
By business, the superintelligence (SI) business became the main driving force behind SpaceX's third-quarter performance growth. The business's third-quarter revenue reached US$7.560 billion, up 830% from US$813 million in the same period last year, and 195% from US$2,561 million in the second quarter, accounting for about 52% of the company's total revenue. Among them, SI solutions and infrastructure revenue reached US$7.022 billion, a significant increase from US$359 million in the same period last year, mainly due to the launch of a new cloud service agreement, the acquisition of artificial intelligence programming company Cursor, and the expansion of the application of the Grok model in the enterprise and consumer markets.
The operating profit of the SI business reached US$431 million in the third quarter, compared with an operating loss of US$1,722 million in the same period last year and an operating loss of US$1,257 million in the second quarter, turning a loss into a profit. The adjusted EBITDA of the business reached US$4.953 billion, compared with a loss of US$428 million in the same period last year.
As another major revenue pillar of SpaceX, the connectivity business continued to grow strongly in the third quarter. The business's third-quarter revenue reached US$5.248 billion, up 69% year on year, up 22% month on month; operating profit reached US$2,367 billion, up 99% year on year; adjusted EBITDA reached US$3.499 billion, up 82% year on year.
Among them, Starlink's consumer business revenue was US$3.121 billion, up 57% year on year, up 26% month on month; corporate and government business revenue reached US$2,127 billion, up 91% year on year and 18% month on month, indicating that demand from enterprise and government customers continues to expand.
In terms of user size, as of September 30, Starlink consumer subscribers reached 13.6 million, an increase of 1.6 million from 12 million at the end of the second quarter, and an increase of about 89% over 7.2 million in the same period last year. Meanwhile, the average monthly revenue of Starlink consumers reached $69 per user, up about 5% from $66 in the second quarter, but still lower than $83 in the same period last year.
The third quarter revenue of the aerospace business reached US$1,767 billion, up 6% year on year and 84% month on month. Among them, launch service revenue was US$1,130 million, up 13% year on year; revenue from launch and development business was US$637 million, down from US$666 million in the same period last year. However, due to the continued increase in R&D investment in the Starship project, the aerospace business still recorded an operating loss of 7 million US dollars in the third quarter, compared with an operating profit of 259 million US dollars in the same period last year. The adjusted EBITDA of the business was US$357 million, down from US$674 million in the same period last year.
SpaceX pointed out that all three of its business segments are showing extremely strong development momentum. The company is trying to use Starships to reduce space transportation costs, use Starlink to expand global communication networks, and use super intelligent computing infrastructure to meet AI computing power requirements, forming a growth system where the three major businesses drive each other.
The company said it is expected to achieve annual recurring revenue (ARR) of 100 billion US dollars by the end of this year, which may reach 500 billion US dollars to 1 trillion US dollars by the end of 2027. Meanwhile, the company expects capital expenditure to rise to 35 billion to 40 billion US dollars in the fourth quarter. The upper end of this guidance range is equivalent to double the capital expenditure in the third quarter.
Regarding the deployment of computing power, SpaceX said it is quite confident that it will deploy an additional 10 gigawatts of computing capacity in 2027. Regarding Starship, the company said it will launch once a week at some stage in the first half of next year, and the launch frequency will reach twice a week by the second half of next year, and may even increase further before the end of next year.
Furthermore, during the conference call, Musk said that SpaceX and Tesla “have set up a committee to discuss the merger,” saying that the merger “has many advantages, but it is not a foregone conclusion and requires shareholder support.”
Here are the details of SpaceX's third-quarter earnings call.
I. Executive speeches
1. Elon Musk - Founder, Chairman, Chief Technology Officer and CEO
For SpaceX, we are currently at a critical turning point. The launch business, communications business, and super intelligence business are all undergoing important changes.
Let's start with Starship (Starship). The 14th flight was a key milestone. The Starship entered orbit for the first time and deployed the Starlink V3 satellite put into actual operation. For our 15th flight, which we expect to take place next month, we'll try to recycle both the booster and the spaceship. If successful, this will be the first time that an orbital launch vehicle has been fully recycled.
This is a major technological milestone. The only way to continue expanding human consciousness beyond Earth is to have a fully reusable orbital launch vehicle. Therefore, I think this will be a far-reaching technological milestone in the history of the development of human civilization and consciousness.
Next, let's take a look at what the simulation process would look like if the 15th flight were successful. If the 15th flight didn't work, we would have achieved our goal in the end. The first few flights aren't a guarantee of success, but I think the next flight has a 50% chance of succeeding. Even without success, we will continue to conduct starship flight tests until the end of this year, and the frequency of test flights will increase next year.
We also announced plans to develop the Louisiana Space Launch Site. I think this will be an important asset for the US and the world at large. At some point in the future, the launch site is likely to have about 20 or more launch towers, capable of carrying millions of tons of payloads into orbit and beyond every year. This is going to be a very significant project.
We expect Starship to be launched once a week at some stage in the first half of next year; by the second half of next year, the launch frequency will reach twice a week, and may even increase further before the end of next year.
As we gradually solve the problem of rapid and complete reuse, the launch frequency of Starships will increase dramatically. We're going to make it completely reusable. Initially, a small amount of maintenance may still be required after each flight, but we will gradually eliminate these maintenance requirements, and finally allow the rocket to be launched again immediately after landing, just like an airplane.
As I said, this is a profound technological milestone. Most people in the world don't realize how important it is right now, but in retrospect, people will understand it.
The Starlink business is also just getting started. The Starlink bandwidth of the 14th flight deployment reached 26 terabits per second, which is equivalent to the effect of 10 Falcon 9 (Falcon 9) launches. In other words, just one launch of the Starship achieved an effect equivalent to 10 Falcon 9 launches, and this time the Starship was not fully loaded; it was only about half loaded.
I expect that in some time to come — and not too far away, I think within about three years — the total bandwidth of Starlink constellations will reach about 100 times the current level. The reason is that we will launch satellites with an order of magnitude higher bandwidth, and the number of Starlink satellites put into orbit will also increase by an order of magnitude. The total number of satellites is likely to reach about 100,000, and the bandwidth of each satellite will exceed 10 times that of the current Starlink system. I think Starlink's revenue will be at least partially proportional to the total bandwidth it provides.
In addition, we will also provide high-bandwidth direct-to-cell (direct-to-cell) capabilities and continue to build terrestrial networks. In this way, wherever space communication services such as Starlink are approved to operate, we can provide a wide range of connectivity services, that is, in any region in the world that is approved to do business.
I think future bandwidth requirements will be much higher than today's. I anticipate that in the next three to five years, bandwidth demand could reach 1,000 times the current level. This is because, on average, each artificial intelligence (AI) or superintelligence (SI) agent will use far more bandwidth than one person, and the number of agents will also far exceed the number of humans. In order to meet these communication requirements, communication bandwidth will eventually need to be increased by several orders of magnitude.
Next, let's talk about SpaceX's super-smart business. It doesn't matter if you call it SpaceX Super Intelligence, SpaceX SI, or whatever.
I think people are beginning to gradually realize that the constraints facing super-smart deployments are more about power, cooling, and infrastructure than the chip itself.
SpaceX's approach is to deploy a team of terrestrial computing infrastructure. The team has previously achieved close to 90% of the global inbound load quality market. The results achieved by this team in the space field are an order of magnitude higher than the performance of all other competitors combined. Now we're applying the same team to the deployment of terrestrial computing infrastructure.
The end result is that we are able to deploy power and cooling infrastructure much faster than any other company, and at some stage, maybe even faster than all others combined. Everyone can take a moment to really appreciate what this means.
I personally spent a lot of time in Memphis recently, often working there four days and nights a week just to get the Colossus II supercomputing cluster in Memphis up and running.
Based on current progress, we expect to deploy around 200,000 new GB300 chips every month remaining this year, and continue to maintain this pace next year. Next year's computing deployment will mainly use Rubin, the Vera Rubin computing platform.
At this stage, I think I am quite confident of deploying an additional 10 gigawatts of computing capacity in 2027. This is an increase on top of the approximately 3 gigawatts expected to be reached by the end of this year; the actual scale is likely to be between 2.5 gigawatts and 3 gigawatts by the end of this year. In other words, I'd be surprised if we had less than 13 gigawatts of computing power deployed by the end of next year.
As I mentioned earlier, infrastructure such as electricity, cooling, and electrical equipment is a limiting factor to a certain extent. To ensure that these factors don't limit the progress of our expansion, we actually plan to add 20 gigawatts of power, cooling, and related infrastructure capacity next year. In this way, we will have far more power and cooling than is needed to start a GPU.
As a result, SpaceX's super-intelligent computing power will exceed 20 gigawatts next year, which is not entirely impossible. At this stage, this possibility may still be relatively small, but it is not zero. In the past, that possibility was zero; now it's not.
We also plan to build the world's most powerful training cluster by the end of the second quarter. The cluster will deploy 440,000 Vera Rubin computing devices to achieve optimal training results. I think this will greatly help Grok grow into the best model possible, or at least make it extremely competitive.
On the Grok Bot side, we've seen a product grow at a rate I've never seen in my life. I've never seen any product grow this fast. We also found that over time, Grok Bot usage per user did not decrease, but rather continued to grow. As we make Grok Bot more useful, we're seeing users using it more, not less, every week.
We will be deploying Grok Bot on a large scale for enterprise customers. This is a very important thing because businesses can effectively use it in teams to work like a virtual employee. We've begun early deployment to some of our trusted customers and expect to be able to roll out Grok Bot widely to enterprise customers in the next few weeks. I think the product that ultimately consumes most of our super-intelligent computing resources is probably Grok Bot.
All in all, this means SpaceX's annual recurring revenue (ARR) could reach $500 billion to $1 trillion by the end of 2027.
2. Gwen Shortwell - President, Chief Operating Officer and Director
Our business execution has once again been outstanding, and we have had excellent results this quarter. Today, I'm going to focus on operational and commercial highlights, as well as my priorities for next quarter. We've shown a slide listing key performance indicators (KPIs), and I'll go through them one by one next.
Let's start with our space business. We remain the world's leading provider of launch services and are responsible for the vast majority of payloads put into orbit.
In the third quarter, we carried out a total of 35 launches, including 21 Starlink launches by Falcon 9, 2 Starship flights, 1 Falcon Heavy (Falcon Heavy) launch, and 12 high-profile customer missions. I'll talk later about the Falcon Heavy Rocket Mission, which sent NASA's Roman Space Telescope into deep space.
We have also carried out 6 national security space launch missions.
Additionally, we have won important new business, including 3 new manned space missions, which will be carried out by Dragon (Dragon), and the mission schedule will continue until 2028; we have also obtained a Starlink Communications contract for NASA's Artemis III (Artemis III) mission.
In addition, we have signed 18 new national security space launch missions, obtained a contract to use a starship to deliver payloads to the lunar surface, and 3 new Starfall missions.
By the way, we just did a Starfall mission today using Falcon 9. Starfall is our space capsule project, which can provide scientific research with an opportunity to enter a microgravity environment in a conventional manner and at a relatively low cost. Related research includes deepening our understanding of diseases such as cancer and diabetes, and developing better treatments for patients around the planet.
In the last week of this quarter, we successfully put Starship into orbit for the 14th flight according to the plan Elon just introduced. Immediately after that, on October 1, we completed a “triple launch” mission: in just 13 hours, we carried out three Falcon missions from three different Falcon rocket launch sites. These include the safe delivery of Crew-13 mission personnel to the International Space Station, a National Reconnaissance Service (NRO) Falcon heavy rocket mission, and a ride-sharing launch mission carrying 130 payloads. This fully demonstrates our ability to perform manned, commercial, and national security missions in a single day, which is amazing.
We are also looking forward to the future development of Starship. Starships will be able to carry maximum payloads and help build the space economy in orbit, other planets, and the Moon. As Elon said, Starship's 14th flight was a major technological milestone. We are still moving forward with the goal of recovering both the first and second stages of the Starship to the launch tower this year, and aim to fly the second stage again at the beginning of next year, while gradually increasing the launch frequency as rapid reuse technology progresses.
We plan to experiment with orbital refills early next year and begin launching the Mobile V2 and Starmind satellites in 2027.
We have had extensive positive and constructive communication with the US Federal Aviation Administration (FAA), and it is anticipated that the license for the 15th flight will allow both the first and second stages to return to the launch site and be captured by the launch tower.
I'm also really looking forward to expanding Starship operations to Louisiana and advancing the second Louisiana land acquisition program. Louisiana's geography, local community, and residents' working spirit are all perfect for the future development of Starship.
Let's talk about Starlink next.
Our performance was once again outstanding this quarter, with consumer, corporate and government businesses all continuing to grow. By the end of the quarter, we had operations in 174 markets, including newly entered Vietnam and Uganda markets.
Looking at user growth, we added 1.6 million new consumer subscribers this quarter — this refers to consumer users, not all users — bringing the total number of global broadband consumer subscribers to just under 14 million. As I mentioned earlier, that number doesn't include hundreds of thousands more business users and airline customers.
We are always encouraged by the positive impact Starlink has had on communities around the world. This includes providing services to schools with hundreds of thousands of students and teachers around the world, as well as supporting emergency relief efforts during natural disasters.
In the third quarter, Starlink's rapid response capabilities supported 25 crisis events, including the earthquake in Venezuela, the earthquake in western Colombia, and Hurricane Lowell in Hawaii. In Venezuela alone, Starlink has donated more than 1,400 sets of equipment to aid agencies, providing critical Starlink broadband connectivity services for more than 250,000 people.
As we mentioned in the previous quarter, there are still significant growth opportunities for corporate and government businesses. Among them, aviation is still our focus area, and the fixed site business is growing at an accelerated pace.
Currently, Starlink has signed an agreement to service approximately 12,000 aircraft, which is equivalent to nearly a quarter of the total number of aircraft available worldwide. We are rapidly expanding our existing customer relationships and developing new ones. We have signed agreements with three of the four largest US airlines, and are continuing to add leading international airline customers. This quarter, Lufthansa, Swiss International Air Lines, Kopa Air, Korean Air, Austrian Airlines, and Asiana Airlines all arranged for the first time to fly an aircraft equipped with Starlink equipment.
To keep up with growing demand, we have quadrupled the production rate of aviation equipment since the beginning of this year, and expect to double again by the end of the year.
Our airline partners say Starlink has become an important reason customers choose to fly with them. Whether it's business travelers who increasingly rely on full-flight connectivity services or regular passengers who want to stay connected anytime, anywhere, they are paying more and more attention to this service. Our data shows that StarLink-supported flights have higher occupancy rates. In fact, our customers also tell us that full in-flight connectivity is rapidly becoming a basic requirement for passengers.
In terms of government business, Starshield (Starshield) continues to provide unique mission-critical capabilities, and we believe there is room for further growth in this business. We still anticipate that in the future, corporate and government revenue will each surpass that of consumer connectivity businesses.
The successful deployment of the V3 broadband satellite put into operation on Starship's 14th flight was an important step in Starlink's development and a key turning point. As Elon said, the increased capabilities of these satellites, as well as the capabilities of the Starship Launch Vehicle itself, will bring additional capacity to various markets and enable us to achieve new sales growth in regions where demand is high.
We anticipate that the size of the communications market will far exceed our traditional understanding of human communication. The spread of automobiles, robots, and superintelligence will all dramatically increase the demand for communication, and this increase is not calculated as a percentage, but may reach several orders of magnitude.
Over the next few years, we expect Starlink to account for a significant share of global internet traffic — the Starlink business itself alone.
We continue to expand our international partnerships in the mobile communications business. In the third quarter, we partnered with new operators to launch services in several countries, including Uganda, the Democratic Republic of the Congo, Kazakhstan, and Costa Rica.
By the end of the quarter, Starlink Mobile (Starlink Mobile) had more than 22 million monthly active users, an increase of 37% over the previous quarter. Recently, our priority remains to continue launching the Mobile V2 satellite constellation next year and integrating the spectrum resources we have purchased to launch the service in early 2028.
Just a few weeks ago, we received approval from the US Federal Communications Commission (FCC) to launch 15,000 Mobile V2 satellites; at the same time, we also announced an agreement to acquire a range of low-frequency spectrum assets. These steps will pave the way for Starlink Mobile to become an important mobile operator in the US.
Next, let's talk about super-smart businesses.
Demand for computing power across the industry continues to grow dramatically. As the scale of Grok and related products continues to expand, we are rapidly improving our computing capabilities to meet our needs and the needs of our cloud service customers. Enterprise customers have given positive feedback on the new models we have recently released, particularly the Grok Bot, which has received strong market adoption once the product was launched.
Grok Bot, our smart assistant, was launched in August. Earlier this month, its paid weekly active users had reached 800,000, more than double the previous month. Business-to-business (B2B) is our fastest growing business category. More and more businesses are using Grok Bot in areas such as sales, marketing, IT, and engineering to enhance their team's ability to work and see it as a team member who can stay online and help handle complex end-to-end workflows. We're also speeding up the release of the Grok model, and expect significant performance improvements in the near future — in fact, we've made some progress, and there will be more in the near future.
At the same time, Cursor's integration work is progressing smoothly, not only enhancing our product capabilities, but also expanding market coverage through large-scale enterprise sales teams. This trend is already showing initial signs. Based on test preparation standards, the number of Grok Enterprise customers, including Cursor, with annual recurring revenue of more than $1 million increased 50% over the previous year.
Last week, we further consolidated our product portfolio and announced the launch of a new package plan that combines Grok, Grok Bot, Cursor, and X Premium in the same package.
Overall, our three business segments are showing extremely strong momentum. The engineering team is expanding the scale of Starship and next-generation satellites to cover broadband, mobile communications, and StarShield businesses; at the same time, the commercial team is expanding the Starlink business to aviation, enterprise, government and global markets.
In the process, we are also using our hardware expertise to expand our super-intelligent business and focus on developing critical computing infrastructure.
3. Brett Johnson - Chief Financial Officer and Company Secretary
We once again achieved strong results in the third quarter of 2026. Revenue increased sharply by 161% year over year to reach US$14.6 billion, mainly due to continued growth in the communications business and a clear inflection point in the superintelligence (SI) business. Adjusted EBITDA once again surpassed revenue, reaching $8.8 billion, up 306% year over year.
Our leading position in the launch business, Starlink's recurring revenue, and the significant expansion of the super intelligent infrastructure business have all contributed to a significant increase in the company's operating leverage. Our balance sheet remains very strong, and our continued focus on capital allocation has enabled the company to accelerate innovation to seize huge revenue opportunities across all business areas.
Next, I'll introduce financial performance by business segment.
Space business
Space business revenue was US$1.8 billion, up 84% month-on-month and 6% year-on-year, mainly benefiting from a favorable customer structure. With strong customer mission reserves for the Falcon Rocket and Dragon spacecraft projects, we can still expect steady growth throughout the year.
In terms of capital allocation, we continue to invest in the Starship project to speed up production of Raptor engines and aircraft, advance Gigabay construction, and advance construction work at Starbase (Starbase) and multiple launch platforms at Cape Canaveral. More importantly, we have identified a site in Louisiana for the next phase of the development of the US space launch business. These investments are critical to unlocking the leapfrog the Starship will bring in terms of cost and carrying capacity.
Communications business
The communications business once again achieved strong revenue growth and profit performance, reaching US$5.2 billion, up 69% year on year and 22% month on month. As Gwen said, our Starlink consumer subscribers reached 13.6 million, and 1.6 million new users were added in the third quarter alone. The growth mainly comes from more mature markets.
This quarter, we began increasing production of V5 kits, which helped ease inventory restrictions. We have strategically managed channel distribution and prioritized the delivery of equipment to retail channels to meet pre-holiday shopping season demand. We expect this to drive a further increase in the number of active users in the fourth quarter.
The average monthly revenue per user (ARPU) of Starlink's consumer business increased 5% month-on-month to $69 per month, mainly benefiting from favorable customer structure changes this quarter.
Corporate and government business revenue increased 91% year over year and 18% month over month. Among them, the US government's StarShield project continues to grow, and the aviation business also maintains a strong momentum.
The aviation business continues to be an important growth opportunity. Many of the world's largest airlines are still in the early stages of installation, and we can already see that the service activation process of major partners is expected to accelerate in 2027.
The gross profit of the telecommunications business doubled year-on-year, and the gross margin reached 55%, including a one-time income of about 3 percentage points, mainly from tariff refunds this quarter. However, even after excluding these one-time benefits, the adjusted EBITDA margin increased by about 2 percentage points year over year.
Looking forward to the future, the satellite development roadmap will be a key factor driving up the capacity and capabilities of communications services. From the end of this year to the whole of 2027, we anticipate large-scale deployment of V3 broadband satellites. As the payload scale is gradually expanded with each launch of the Starship, its downstream capacity will exceed 20 times that of the current satellite deployed by the Falcon 9 launch.
At the same time, starting in 2027, we will use Starship to accelerate the deployment of Mobile V2 satellites, expand direct mobile phone services using the recently expanded wireless spectrum portfolio, and become an important mobile operator in the US, as Gwen just mentioned. In the three types of communication applications: Starlink Broadband, Starlink Mobile, and StarShield, the increase in Starship's launch frequency is expected to bring leaps and bounds in total capacity, coverage, and service quality.
Super smart business
Superintelligence (SI) was once again the most prominent driver of growth this quarter. The business segment's revenue increased sharply to 7.6 billion US dollars, an increase of 5 billion US dollars over the previous year, and an increase of 6.7 billion US dollars over the previous year, mainly driven by the expansion of the cloud service agreement business.
As the newly deployed computing power began to be monetized, we significantly improved the profitability of this business segment and brought strong operating leverage to the company. By the end of the third quarter, our computing power nameplate had reached 1.7 gigawatts, and more data halls were put into operation in early October.
In the super-smart solutions and infrastructure business, revenue was $7 billion, of which $608 million came from Cursor, which completed its acquisition around the middle of the quarter. Even without Cursor, revenue from the super-smart solutions and infrastructure business increased 192% month-on-month, mainly due to the launch of a new cloud service agreement.
Currently, many customers are looking for computing resources, and we are actively expanding our supply to meet internal and external needs. We expect to gain more customers through new deals in the future, and will continue to evaluate the best use of computing resources in order to prioritize demand with the highest profit margin between new and existing cloud service agreements and inference services for our own super-smart products.
What we saw in the market was amazing. At current pricing, the payback period for newly deployed computing resources is less than a year. The current pricing is slightly above the upper limit in the $30-50 per watt range we discussed last quarter, further reflecting the appeal of terrestrial super-intelligent computing investments. We will continue to be demand-driven, carefully determine the speed of deployment, and rigorously and proactively plan capital expenditure and semiconductor supply procurement based on immediate customer needs.
The current super intelligent computing customer contract also provides us with flexibility: we can not only adjust our computing capacity to meet the needs of our own model training, but also monetize revenue through super intelligent cloud service agreements at favorable prices when market supply continues to tighten.
Ad revenue increased 19% year over year, thanks to our continued momentum to drive the new ad technology platform launched in the previous quarter, that is, the second quarter.
The adjusted EBITDA of the Super Smart Business segment was US$5 billion, an increase of US$3.8 billion over the previous year, and an increase of US$5.4 billion over the previous year, mainly driven by high-margin cloud service agreements. The operating profit for this business segment was $431 million. However, if the $1.2 billion equity incentive fee associated with the acquisition of Cursor is excluded, the sector's operating profit reached 1.6 billion US dollars in the third quarter alone.
Let's talk about equity incentive fees in more detail. When we bought Cursor, its equity was worth $60 billion, of which $23 billion was classified as being used to retain key employees and replace unpaid equity incentives. Unrealized equity rewards, including resetting vesting arrangements for key executives. These equity rewards will gradually be recognized as equity incentive expenses over the next three years or so.
The company's total capital expenditure for the third quarter was 19.6 billion US dollars, mainly to support the expansion of superintelligent computing infrastructure, the construction of starship infrastructure, and satellite production.
As of September 30, our total backlog of orders reached $63 billion at the end of the quarter, of which $17.5 billion has been recognized as deferred revenue. Approximately $42 billion of this backlog of orders is expected to be recognized as revenue within the next year.
Looking ahead, demand from the three major business segments remains strong, especially the cloud services business. We are still on schedule to advance our previously announced goal of achieving $100 billion in annual recurring revenue (ARR) by the end of this year. The ARR here is calculated using the annualized monthly income for December.
Memphis and Southaven's super-smart infrastructure continues to be in operation, allowing us to clearly see a path to further accelerate construction throughout the rest of the year. Therefore, we expect that by the end of October alone, the computing power nameplate will be close to 2.2 gigawatts; as Elon just emphasized, we also have a clear certainty that the computing power nameplate will increase to 2.5 gigawatts to 3 gigawatts by the end of this quarter.
As more computing power will be put into operation this quarter, we currently expect capital expenditure of $35 billion to $40 billion for the fourth quarter.
We also expect that the company's overall gross margin in the fourth quarter will be closer to the level of the second quarter, as there were some one-time earnings and favorable customer structure factors in the third quarter, and these factors have been explained to you before.
Given our increased visibility into computing service contract reserves over the next six months, we have an opportunity to provide further forward-looking guidance on how we expect business growth to accelerate in the first quarter of 2027. We currently expect the company's total revenue to reach $33 billion to $37 billion in the first quarter of 2027.
Finally, I would like to emphasize that we are at a critical stage in the transformation of the company's financial structure.
We are actively investing in three areas where we believe will bring the highest long-term returns: super-intelligent computing capabilities, Starship R&D and production, and next-generation Starlink broadband and mobile satellite constellations. These investments are driven by market demand, and the immediate returns in the markets they are in have clear, specific visibility.
We are building future infrastructure in a disciplined manner, focusing on capital efficiency, strong return on invested capital, and maintaining a sound investment-grade balance sheet, while continuing to seize immediate market opportunities fast enough. A diverse profit base, a focus on high vertical integration, and strong market demand for superintelligence have all strengthened our confidence in the future development path.
II. Q&A session
1. Morgan Stanley analyst Adam Jonas
I have a two part question about Terafab. You want to make chips, and you want to manufacture chips on a large scale in the US or in a safe and secure location. First, can you confirm whether SpaceX has ordered extreme ultraviolet lithography (EUV) equipment? If already ordered, when do you expect delivery?
The second question is, if you start from first principles and re-examine the semiconductor fab supply chain. I know you've visited companies such as TSMC, Asmack, AMD, and Fanlin Group, and I'm sure you'll agree that these are all excellent companies. If you work your own way, what aspects of these companies' processes and supply chains do you think would be most likely to be improved?
Musk
We are currently still exploring chip manufacturing. However, I can tell you our internal expectations. Of course, these expectations have a large margin of error because we don't know if we can actually achieve our goals.
But what we're trying to do internally is: next year, around May or June, so that our first test chip is capable of running properly. It's just an SRAM-like chip, which is basically used to test the entire production line. The work will be carried out at a research fab in Austin.
Our goal is to achieve mass production at Terafab's main base, Gibbons Creek (Gibbons Creek) around the third quarter of 2028. These schedules are extremely aggressive. To achieve this goal, we are very grateful to Asmat and all other suppliers for speeding up the delivery of the equipment.
Arguably, among all possible outcomes, we at least think it is possible to achieve mass production in the third or fourth quarter of 2028. This is clearly an extremely tight and extremely aggressive timeline for chip manufacturing.
I think we are indeed likely to make some major breakthroughs in chip manufacturing, but I don't think I should discuss these things on the performance conference call. Let's just not show off all the undercards. However, I am optimistic that we can significantly speed up chip production, and that Terafab will eventually achieve the goal its name suggests: to produce super-smart chips equivalent to 1 terawatt of computing power every year.
2. Doug Anmuth, analyst at J.P. Morgan
Elon, can you talk more about how you're addressing power, cooling, and infrastructure constraints? What makes you confident of achieving 13 gigawatts of computing power by the end of 2027? Also, Brett, can you help us understand how you can consider the corresponding funding requirements as computing power expands to this scale?
Musk
Yes. We expect to deploy around 3 gigawatts, or very close to 3 gigawatts, in the next few months alone. So, you only need to look at the speed at which we are currently actually deploying and launching computing power to understand this.
In my opinion, in the information you've heard from other hyperscale cloud service providers, few companies will detail how many GPUs they actually launch each month. And we'll tell you this data, and I'll give you as accurate information as possible.
Now, deploying and supplying the electricity, all the electrical equipment, cooling systems, batteries, networks, and other supporting facilities needed to start 20,000 GB300 chips every month is extremely crazy in itself. But that's really what we're doing right now. Obviously, the pace will pick up even more next year. We're not slowing down. As to how we get all this electricity, I don't want to reveal too much.
Arguably, we do have a plan that aims to increase power and cooling capacity to 20 gigawatts by the end of next year. However, the specific plan is inappropriate for public disclosure, as this is highly proprietary information. If I told you about it, other businesses would try to do the same.
Also, I'm not saying we're bound to have 20 gigawatts of electricity. What I want to say is that if all goes well, our plan is to add 20 gigawatts of electricity capacity next year. In other words, the total size reached approximately 23 gigawatts.
Of course, I don't think everything will go smoothly. That's why I say I'm confident that I'll add 10 gigawatts next year, and I have a high level of confidence in that. As for adding 20 gigawatts, it is difficult, but not impossible.
I've publicly mentioned some of the things we're doing before. For example, we are casting blades and guide vanes for turbines. SpaceX already has very advanced foundry capabilities; this is the technology we have developed for rocket engines. I think we probably have the most advanced casting capabilities on Earth because rocket engines require more castings than any other product you can imagine.
So it's actually much easier for this excellent team with world-leading foundry technology to make some blades and guide vanes for ground turbines compared to the work they have done in the past. When we spoke to the turbine supplier, they told us that the delivery schedule for turbine blades and guide vanes was 12 to 18 months behind other turbine components. In other words, they can deliver all other parts of the turbine 12 to 18 months in advance, but not deliver the blades and guide vanes until later.
As a result, the casting process poses an extremely difficult bottleneck. Moreover, judging from the current situation, this bottleneck seems to be getting worse. This means we can put electricity into use 12 to 18 months earlier than other companies. This is an advantage that is hard to replicate. Unless other companies are also doing advanced casting or building advanced foundry plants, they won't be able to compete with SpaceX in this regard.
We're also looking for many different sources of electricity, not just gas turbines. Obviously, we want to use solar, wind, and every other source of electricity we can get.
We believe that there are also plenty of opportunities for deductive computation. For example, if it is possible to manufacture modular inference computing devices at the factory level, integrate power and cooling systems in the factory, and then distribute these modular inference computing clusters all over the country, or even all over the world, it may create great opportunities.
That's all I can say about expanding power and cooling capacity. But as I said, at this stage, I have high confidence in adding 10 gigawatts of computing power next year, which will bring the total scale to 13 gigawatts. As for eventually achieving a scale of more than 20 gigawatts, the possibilities are rapidly rising.
Brett
Doug, the second part of your question concerns funding. I think we do have some advantages over other businesses: we have an investment-grade credit rating and a very solid balance sheet. We are currently communicating with multiple sources of debt capital to finance this strong growth.
Another very important factor is that the payback period is very short. As a result, these additional capabilities will soon begin to self-fund. In other words, even the additional computing power at the end of this quarter will allow the relevant fund recovery mechanism to continue to work in the first half of next year.
First, we can rely on our own operating cash flow; second, since we have an investment-grade credit rating and a robust balance sheet, there are multiple funding sources to choose from in the debt market. Additionally, based on our current experience, we have the opportunity to work with participants across the entire value chain, including suppliers and customers.
Of course, running SpaceX is not easy, so financing will naturally also be a challenge we must face. However, we are very confident that we can proceed with this plan in a prudent and orderly manner and continue to maintain our investment-grade credit rating throughout the process.
Musk
Of course, if you look at our historical performance in terms of capital efficiency, whether it's Starships or all other projects, it's hard to find a company in the world that uses capital more efficiently than SpaceX.
3. Deutsche Bank analyst Xin Yu
Recently, there have been many reports about a possible merger between SpaceX and Tesla, and the market is looking forward to it. I know you may have some limitations on this issue, and you may not be able to reveal too much information. But if we take a step back, how do you assess the pros and cons of a potential merger between the two companies? What other hurdles need to be addressed before moving forward with this matter?
Musk
This question really makes me feel like dancing in a minefield, so I have to be careful about wording it. What I want to say is that at the request of shareholders, both SpaceX and Tesla have set up special committees to discuss this idea. These special committees are composed of independent directors.
I cannot actually participate in the Special Committee's discussions due to my own conflict of interest. So it's kind of like a weird kabuki play: the two groups are discussing with each other, and I can't participate in it.
I don't think we should take it for granted that SpaceX's merger with Tesla is a foregone conclusion. I know some people might think that this merger is bound to happen, but that judgment is overstated.
Do I think the merger is a good idea? Yes, I think so. But at the end of the day, shareholders must back the deal. For Tesla in particular, we need 50% of the voting support of all shares, not just 50% of the shares that have already participated in the voting. In other words, it must be supported by 50% of the total number of shares held by all shareholders. That's a pretty high threshold.
I can tell you that currently the special committees of the two companies have not agreed on the terms of any deal. Obviously, there is no guarantee that the two sides will eventually reach an agreement. However, I do think there is great value in merging the two companies because we are near the threshold of the super-smart singularity era.
Through collaboration, we have been able to reap significant benefits in various fields, including super-intelligent software development, chip design, Terafab projects, and of course robotics, and the use of the Starmind satellite constellation to put payloads into orbit.
So I think the merger would be a good idea, but the final decision is in the hands of the two companies' special committees. They must come up with a set of proposals before submitting them to shareholders' voting.
4. Evercore analyst Kutgun Maral
I'd like to ask about Starlink's mobile business. It has come a long way in a short time, including a recent low-frequency spectrum deal. Even so, we're still uncertain to what extent your spectrum mix can compete with existing US mobile carriers, even though it has improved significantly and is likely to expand in the future.
Another unknown factor for outside observers is how much coverage and capacity the satellite network layer can actually provide to bridge the gap with peer terrestrial networks. So what role do you expect satellite networks and terrestrial networks to play, respectively? How should we understand the traffic distribution between the two, and how will you provide a 5G-like user experience in the US?
More broadly, how much of your communications business is focused on traditional retail wireless communications, and how much of it is likely to come from large-scale physical AI opportunities?
Shotwell
Elon can add a few details, but I'd like to start by pointing out that the vast majority of mobile communication is actually done via Wi-Fi.
I think the ratio is around 85% to 93%. Also keep in mind that Starlink Broadband itself also takes on a large number of Wi-Fi connectivity tasks. As for the division of labor among different spectrums, we expect the S-band spectrum to take on most of the communication tasks.
At the same time, we obviously need a low frequency spectrum, or the 800 MHz band, which is mainly used to penetrate buildings. In this way, the signal can be received even when the user is not connected to Wi-Fi.
I'll talk about that first, and if I find myself saying something inaccurate later, I'll correct it. But I think the S-band will handle most of the traffic, while the low frequency spectrum will take a smaller share, but it is still a critical part of the overall network.
Additionally, we also talked about some smaller radios during the last quarter's results conference call. They may not be sufficient to replace complete communication towers, but they can be installed in homes, corporate sites, and communication towers where Starlink equipment has already been deployed. Therefore, we will not only use the S-band spectrum mentioned earlier, but we will also use Starlink's existing installation sites and other communication tower site resources.
Musk
Yes. I think we should probably host a longer presentation detailing the overall plan of Starlink Communications business and why we think it will eventually become a business with annual revenue of hundreds of billions of dollars. You don't currently know some of the information we have internally.
I can say that we have indeed seen a clear path to hundreds of billions of dollars in communications business revenue each year. However, I think we should explain this in more detail when we have more time, rather than being limited to a few minutes during the performance conference call.
5. Goldman Sachs analyst Eric Sheridan
I'd like to go back to the computing power expansion plan you introduced at the beginning of the conference call. As you consider how to allocate these resources between external cloud service market opportunities and internal computing needs, and especially considering that Cursor and Grok Bot scale requires the use of internal computing power, how have your views on how to achieve the right balance between the two changed? From the perspective of commercial monetization, how can we understand the benefits this distribution method may bring to the company in the next 12 to 18 months?
Musk
The distribution ratio between the computing power we provide to third parties and the computing power used internally may change from month to month. Broadly speaking, our goal is five or five: half of the computing power is for internal use and the other half is leased to other customers. However, this ratio may be adjusted to 64 degrees or change in the opposite direction. The exact ratio will be dynamic. Obviously, our approach is to rent out computing power at actual spot market prices rather than sign these long-term contracts.
We believe that the value of computational power corresponding to a unit of power will increase rather than decrease over time. I think the reason behind this is actually quite obvious: as AI software continues to improve and models become more powerful, each watt of electricity can generate more and more effective intelligence. The value of computing resources is directly proportional to the effective intelligence that can be generated per watt of electricity.
That's why you'll see that the current rental price of even old computing devices like the H100 is probably higher than when they first launched. I think this trend is likely to continue. I think there will be a shortage of super-intelligent computing power, and this shortage will continue for a long time.
Therefore, I think the smart approach is to sign a short-term contract, essentially betting that the effective intelligence generated per watt of electricity will rapidly improve. And the facts have proven that this increase is actually happening. This is the current state of affairs.
Additionally, we can observe price changes when short-term contracts are renewed: is the renewal price per watt of computing power rising or falling? What we're seeing is that renewal prices are rising, not falling.
Frankly speaking, all hyperscale cloud service providers and dozens of smaller companies are constantly pressing us to rent out more computing power. Existing computing resources can even be leased out for several rounds if we agree to everyone's requests.
This is the reality we are facing right now. That's why I posted that video, modeled after the “Cheese Shop” skit by the British comedy group Monty Python (Monty Python). In that skit, a person walks into a cheese shop and wants to buy cheese. As a result, there is actually no cheese in the store. So he kept asking, “What about this kind of cheese? What about that kind of cheese?” But there really isn't any cheese in the store.
In fact, we have also tried to transfer some inference computation tasks to other cloud service providers in order to use more computational power for model training. But the result was just like that cheese shop skit. We'll ask, “Do you have GB300?” The other party said no. “What about GB200?” Still not. “What about TPU? What about Trainium?” Then we asked, “OK, what about the H100?” The other person might say, “We probably have some more somewhere overseas.” But after inquiring, they came back and told us, “No, we haven't either.” This is just ridiculous.
So I don't think we're going to face a shortage of demand. Moreover, the payback period for computing resources next year is likely to be only six months. It's not an impossible situation. Of course, I'm not saying it's bound to happen; I'm just saying that the possibility does exist.
In addition, SpaceX executives also answered questions submitted directly to the public Q&A platform by individual investors.
Question 1: When will the increase in capacity brought about by the V3 satellite translate into higher network speed and lower latency that users can clearly feel? In regions where demand is strong, how many V3 satellites will need to be deployed to improve the network experience significantly?
Shotwell
I think it will take a few hundred V3 satellites to be deployed before users can really feel the impact. Because the number of satellites is still very small, users may temporarily feel an improvement in the network experience, but after 10 minutes, this improvement may disappear. So we need a few hundred satellites. I think we will be able to achieve this goal by mid-next year.
Q2: Can you please update on the progress of the Louisiana Interstellar Base? What is the project timeline? How confident are you currently about carrying out the first Starship launch from this base as early as 2029?
Musk
In fact, our internal goal is to launch our first launch from Louisiana within two years, which is to be achieved by 2028. However, I think it is very likely that the first launch will be achieved in 2029. I think we'll be able to launch large-scale launches from Louisiana by 2029. I'm very confident about that.