Economic Observer reporter Zheng Chenye

On the early morning of October 21st Beijing time, SpaceX (SPCX. O) held its 2026 third quarter performance meeting. I think there will be a shortage of super intelligent computing power, and this shortage will last for a long time, "said Elon Musk, the CEO of the company, at the meeting

Super intelligence is SpaceX's term for its AI business, which mainly includes renting out computing power and AI products such as Grok. Musk also stated at the meeting that he would be surprised if SpaceX's computing power put into use by the end of 2027 is less than 13 gigawatts. As of the end of September 2026, this scale is 1.7 gigawatts (calculated based on the capacitance of the data center). That is to say, in order to achieve this expectation, SpaceX's computing power needs to expand to more than 7 times that of the end of September in just over a year.

SpaceX's 2026 third quarter performance data shows a revenue of $14.575 billion, a year-on-year increase of 161%, higher than market expectations; Among them, the revenue from the super intelligent business was 7.56 billion US dollars, surpassing the connectivity business mainly based on Starlink. As a result, on October 21st Eastern Time, SpaceX's stock price closed up 7.79%.

In Musk's view, the demand for computing power will be greater than the supply in the long run, and currently the constraints on supply are more on electricity, cooling, and infrastructure, with chips coming in second.

Demand Exceeds Supply

SpaceX's computing power business is currently in short supply, and external customers want to rent more, so their own products also need to be used more.

Musk said at the earnings conference, "Every large cloud service provider and dozens of small companies are constantly looking for us to rent more computing power." The company's CFO, Bret Johnson, also stated that many customers are looking for SpaceX's computing power, and the company is actively expanding its supply to meet internal and external demand.

In the past year, SpaceX's computing power scale has increased from 0.6 gigawatts to 1.7 gigawatts, and the revenue from its super intelligence business has jumped accordingly. According to Johnson's introduction, in the third quarter of 2026, the revenue of this business increased by approximately $5 billion month on month and $6.7 billion year-on-year, mainly due to the gradual increase in cloud service agreements.

The computing power used by SpaceX's own products is also tight. Musk stated that in order to free up more computing power to train the model, the company has attempted to outsource some of the inference tasks (i.e., the computation when the model responds to user requests) to other cloud service providers. Because compared to training, reasoning tasks are easier to split and run on other platforms. He also believes that the agent assistant Grok Bot may ultimately use up most of the company's super intelligent computing power.

Grok Bot was launched in August, and Musk said it is the fastest-growing product he has ever seen. It is currently being used by some customers and will be widely deployed to enterprise customers within a few weeks. According to Gwen Shotwell, the President and Chief Operating Officer of the company, Grok Bot's paid weekly active users had reached 800000 earlier in October, more than doubling in a month, with enterprise customers being the fastest-growing category.

Musk estimates that with the expansion of computing power and the demand for Grok Bot, SpaceX's annualized recurring revenue (calculated as annual revenue at the end of the year) will be between $500 billion and $1 trillion by the end of 2027. Johnson mentioned at the performance meeting that the target for this figure by the end of this year is $100 billion.

We believe that the value of each watt of computing power will increase over time and not decrease, "Musk said at the earnings conference." As AI software continues to improve and models become better, more 'useful intelligence' can be produced per watt of computing power, making computing power more valuable. That's why even old models like the H100 (Nvidia's AI chip launched in 2022) now have higher rents than when they were first launched.

So I think the smart approach is to sign short-term contracts and bet on the rapid growth of useful intelligence per watt, "Musk said. According to him, SpaceX rents computing power at prices close to those in the spot market and does not sign long-term contracts; From the situation of renewal, the price per watt of computing power is higher than before. If computing power is indeed becoming increasingly valuable, signing a long-term contract is equivalent to locking the price at the current level.

According to publicly released documents by SpaceX, the clients renting its computing power include AI companies Anthropic and Google. Although both agreements have a term until 2029, after the initial stage, both parties can terminate them with 90 days' notice.

Johnson also stated at the performance meeting that the current price of computing power is slightly higher than the upper limit of the $30 to $50 per watt range mentioned by the company at the 2026 Q2 performance meeting; At this price, the newly deployed computing power can recoup costs in less than a year, and the operating cash flow can support expansion. Musk believes that the payback period for computing power investments in 2027 may be shortened to six months, although this is not "necessarily going to happen".

Power Bottleneck

Musk stated at the earnings conference that more and more people are realizing that deploying super intelligence is more constrained by electricity, cooling systems, and infrastructure than by chips. The computing power scale of AI data centers is calculated based on electricity capacity, and for every additional gigawatt of computing power, corresponding power generation and cooling facilities must keep up.

According to Musk's introduction, SpaceX plans to put into use a training cluster consisting of 440000 Nvidia Vera Rubin chips by the end of the second quarter of 2027. In addition, he is also promoting the establishment of a self built chip factory, but acknowledges that the company is still "exploring" in this area. The timetable he provided is to produce the first functional test chip using Austin's research wafer fab in May or June 2027, and to achieve mass production of the planned chip factory Terafab around the second half of 2028.

In terms of electricity, due to the fact that it often takes several years to connect to the grid, many AI data centers choose to use gas turbines to build their own power sources, making gas turbines in high demand. According to data from gas turbine manufacturer GE Vernova (GEV. N), as of the end of the second quarter of 2026, its on hand orders for gas power generation equipment and capacity reservation agreements (customers pay to lock in future production schedules) totaled 116 gigawatts, while according to the plan, its annual gas turbine capacity will only increase to 20 gigawatts in the third quarter of 2026. The CEO of the company, Scott Strazick, stated at the July earnings conference that the focus of negotiations with clients is already on deliveries for 2032 and beyond.

Musk also stated that gas turbine suppliers told SpaceX that the delivery of blades and guide vanes is 12 to 18 months later than other components of the entire machine, and casting is a "very difficult bottleneck" among them. The blades rotate at high speed with the rotor, while the guide vanes remain stationary and are responsible for guiding the flow of gas. Both directly withstand the erosion of high-temperature gas and require precision casting with high-temperature alloys. There are not many manufacturers that can provide stable supply.

American aviation parts manufacturer HWM. N is one of the main suppliers of this type of blade. John Plant, the executive chairman and CEO of the company, stated at the August earnings conference that H ä agen Aviation holds over 50% of the global market share for gas turbine blades used in power generation, and the company's newly built factory in Japan is already fully booked.

According to Plant, the new production capacity will not be gradually released until 2028 to 2030, and the company also has to book production capacity from some machine tool suppliers. Patrick Winterlich, the Chief Financial Officer of Haomeite Airlines, attributed the growth of the gas turbine business to the increasing demand for power generation, especially the demand for gas power generation in data centers.

Yes, Musk's solution is still for SpaceX to cast the blades and guide vanes themselves. After he announced this plan at the end of August, the stock price of Haomeite Airlines fell by more than 7% on the following first trading day. At the aforementioned performance meeting, Musk explained that SpaceX's casting capabilities for rocket engines can be used to produce these components, "so we can connect the electricity 12 to 18 months earlier than anyone else. He also stated that in addition to gas turbines, SpaceX hopes to use solar, wind, and all available electricity.

Johnson stated that by the end of October, SpaceX's computing power is expected to approach 2.2 gigawatts. According to this calculation, approximately 0.5 gigawatts will be added in October alone, compared to approximately 1.1 gigawatts added in the previous year. Musk also stated that it is "crazy" to build power, cooling facilities, electrical equipment, batteries, and networks to support the large number of new chips that are put into operation every month, but the pace of deployment in 2027 will accelerate.

Musk's plan is that by the end of 2026, SpaceX's computing power will reach 2.5 to 3 gigawatts. In 2027, SpaceX aims to add 20 gigawatts of electricity, cooling, and infrastructure. His view on this is: "Difficult, but not impossible." At least, he is "very confident" in adding 10 gigawatts of computing power by 2027.

Of course, the related capital investment is also increasing. In the third quarter of 2026, SpaceX's capital expenditure was $19.593 billion, and Johnson expects it to reach $35 billion to $40 billion in the fourth quarter of 2026. In addition, the specific deployment of computing power is demand-oriented and pushed forward according to the rhythm of the contract. Funds can not only come from operating cash flow, but also from investment grade credit and multiple debt fund pools.

Whether these plans can be fulfilled as scheduled, Musk himself has also left room for improvement. He admitted at the performance meeting that he doesn't expect everything to go smoothly.

(Author Zheng Chenye)