Elon Musk: Company has a possibility of achieving 20GW superintelligent computing power by the end of 2027
During SpaceX's Q3 earnings call, CEO Elon Musk stated that there is a non-zero probability that the company will achieve 20GW of superintelligent computing power online by the end of 2027, with the current nameplate capacity at 1.7GW as of September 30. Musk expressed high confidence in adding approximately 10GW of capacity by 2027, which would bring the total close to 13GW, while the goal is to have an additional 20GW of power and cooling infrastructure online by the end of the year; the capacity target for the end of 2026 is 2.5 to 3GW. The company is deploying at a rate of about 200,000 Nvidia GB300 GPUs per month, will later shift to the Vera Rubin architecture, and plans to launch a large training cluster of 440,000 GPUs by mid-2027. Q3 total revenue reached $14.6 billion, a 161% year-on-year increase, with the superintelligent segment contributing $7.56 billion, an 830% increase, becoming the largest business segment for the first time; the adjusted EBITDA for this segment was $5 billion, with capital expenditures of $19.6 billion, mainly for AI infrastructure expansion. Musk noted that the computing team and the rocket team are the same group of people, with the latter achieving a 96% success rate in sending payloads to space, comparing the computing team's efforts to the New York Yankees playing in a minor league. He has been frequently stationed in Memphis recently, about four days a week. The company has signed multiple cloud service agreements, including collaborations with Anthropic and Google Cloud, and in September signed a $13.3 billion annual computing power leasing agreement with an undisclosed client; current annualized computing power leasing revenue is about $41.1 billion, with a goal to exceed $100 billion ARR by year-end. Musk expects that by the end of 2027, SpaceX's annual recurring revenue will fall between $500 billion and $1 trillion; based on a value of $30 to $50 per watt per year, 10GW of computing power could support revenue of about $300 billion to $500 billion, while 20GW would correspond to a higher ceiling, creating a dual pull of event-driven capital expenditures and customer prepayments. The funding flow mainly comes from internal cash reserves (about $90 billion) and customer prepayments, benefiting Nvidia chip suppliers and power cooling equipment vendors, while the pressured parties include the power grid and land resource bottlenecks, as well as potential project delivery risks. Source: Public information
ABAB AI Insight
Elon Musk's companies have always been driven by extreme goals. SpaceX has transitioned from multiple failures with Falcon 1 to achieving reusable rockets and dominating global launch quality. xAI has rapidly launched the Grok series and integrated Cursor from scratch. Historical paths show a tendency to compress time windows through vertical integration and high capital expenditures, but there have also been instances of delayed targets and execution deviations. The current capital path involves directly transplanting the rocket team's systems engineering capabilities to the computing cluster, deploying 200,000 GPUs monthly and pre-building 15-20GW of power and cooling, treating computing power as a rentable infrastructure for external clients like Anthropic and Google, while also serving internal model training. The motivation is to utilize existing cash and customer prepayments to achieve capital recovery in less than a year, converting the scale advantages of the rocket business into pricing and delivery speed advantages for AI computing power. This is similar to early AWS opening e-commerce infrastructure to external clients or Tesla extending battery capacity to energy storage and vehicles. SpaceX is transitioning from a launch and connectivity service provider to a computing infrastructure controller, moving from vertical integration to platform output, with current computing power growth far outpacing the historical ramp-up curve of traditional cloud vendors. Essentially, this represents a restructuring of the industry chain and a transfer of pricing power: by building its own power cooling and chip deployment rhythm, SpaceX aims to gain control over the timing and cost of AI computing supply, with a mechanism of high capital expenditures locking in customers upfront, and then forming barriers through rapid delivery, forcing competitors to passively follow in power and cluster delivery rather than relying solely on technological substitution or regulatory drives. ABAB News · Cognitive Laws 1. Computing power scale equals time compressor; speed determines pricing power. 2. Infrastructure is built before being sold; customer prepayments are the best leverage. 3. The same team migrating across fields; system capabilities are more valuable than single technologies.