SpaceX is targeting at least 13 gigawatts (GW) of artificial intelligence computing capacity by the end of 2027, marking an aggressive expansion of its AI infrastructure. During the company’s third-quarter 2026 earnings call, CEO Elon Musk said SpaceX expects to add approximately 10 GW of compute capacity next year, on top of the 2.5 GW to 3 GW it expects to have operating by the end of 2026.
The plan involves expanding the company’s data-center infrastructure, deploying large numbers of Nvidia AI processors and building additional power and cooling capacity. SpaceX is also preparing a major training cluster based on Nvidia’s Vera Rubin architecture to support the development of Grok, the AI model associated with Musk’s xAI business. However, the 13 GW target remains a forward-looking goal, and the company faces significant challenges involving electricity, cooling equipment, chip supply and capital spending.
Key Takeaways
- 13 GW target: SpaceX aims to have at least 13 GW of AI compute capacity online by the end of 2027.
- 10 GW expansion: Musk expects to add roughly 10 GW of computing capacity during 2027.
- Large-scale deployment: The company plans to continue installing approximately 200,000 Nvidia GB300 processors per month, with next year’s expansion increasingly using Vera Rubin hardware.
- Major training cluster: SpaceX is targeting a 440,000-unit Vera Rubin cluster by the end of the second quarter of 2027.
- Power infrastructure: The company plans to pursue up to 20 GW of additional power and cooling infrastructure in 2027 to avoid constraining its computing expansion.
- Key risks: Electricity availability, equipment delivery, construction, financing and the pace of chip deployment could affect the timeline.
What Is SpaceX Planning for AI Compute Capacity?
SpaceX’s AI infrastructure ambitions extend beyond operating a rocket and satellite business. The company is building large computing facilities to support AI model training, inference and other workloads while also considering opportunities to sell computing capacity to external customers.
According to the company’s third-quarter earnings call, SpaceX had approximately 1.7 GW of compute capacity at the end of September 2026. Management expects that figure to increase to roughly 2.5 GW to 3 GW by the end of the year.
Musk said he expects the company to add approximately 10 GW of computing capacity during 2027. He added that he would be surprised if SpaceX ended the year with less than 13 GW online.
The target represents a substantial increase in the company’s computing footprint in a relatively short period. It also reflects SpaceX’s strategy of combining its existing infrastructure capabilities with large-scale investment in AI computing.
| Milestone | SpaceX’s stated plan or reported position |
|---|---|
| September 30, 2026 | Approximately 1.7 GW of compute capacity |
| End of 2026 | Approximately 2.5–3 GW expected |
| During 2027 | Approximately 10 GW of additional compute |
| End of 2027 | At least 13 GW targeted |
| End of Q2 2027 | Target for a 440,000-unit Vera Rubin training cluster |
These figures are company targets and management estimates, not guarantees that all planned capacity will be operational on schedule.
How Nvidia Chips Fit Into the Expansion
AI compute capacity depends on more than the number of processors a company purchases. Facilities also need electrical equipment, cooling systems, networking infrastructure, storage and the buildings required to house the hardware.
SpaceX is planning to continue deploying Nvidia GB300 systems at a reported pace of approximately 200,000 units per month, with the expansion next year expected to rely primarily on Nvidia’s Vera Rubin architecture.
Musk also outlined plans for a 440,000-unit Vera Rubin training cluster targeted for deployment by the end of the second quarter of 2027. The cluster is intended to provide substantial computing resources for developing and training Grok.
Nvidia’s Vera Rubin platform is designed for demanding AI workloads, including model training and inference. A deployment at this scale would give SpaceX access to substantial processing capacity, provided the supporting infrastructure and equipment arrive on schedule.
The scale of the planned deployment also illustrates how competition in AI increasingly depends on access to hardware and the ability to bring complete computing systems online. Buying processors alone is not sufficient if the company cannot supply electricity, cool the equipment and connect the systems efficiently.
Why SpaceX Is Targeting 20 GW of Power and Cooling Infrastructure
One of the most significant constraints in the expansion is the infrastructure required to operate the computing equipment.
Musk said SpaceX plans to pursue approximately 20 GW of additional power and cooling infrastructure during 2027. The objective is to ensure that electricity supply, cooling equipment and related systems do not prevent the company from activating its planned processors.
The distinction between compute capacity and power infrastructure is important. A company may have enough chips to support a particular computing capacity but still be unable to operate them because its facilities lack sufficient electricity or cooling.
SpaceX is attempting to create a buffer by planning infrastructure beyond the level required for its initial compute target. Musk acknowledged that not all projects would necessarily be completed on time and did not provide a detailed explanation of the entire power-supply strategy.
The company’s approach is particularly ambitious because large data centers require substantial electrical equipment and specialized cooling systems. These components can involve long lead times, complex construction and regulatory requirements.
SpaceX’s ability to bring its proposed computing capacity online will therefore depend on more than processor deliveries. Its infrastructure expansion must progress at a comparable pace.
Colossus and the Role of SpaceX’s AI Data Centers
SpaceX has been investing heavily in large computing facilities, including the Colossus cluster in Memphis, Tennessee, and the Colossus II expansion.
Musk said he has been spending substantial time in Memphis overseeing efforts to bring Colossus II online. These facilities are central to the company’s effort to scale computing capacity for AI training and related workloads.
The company’s infrastructure strategy also reflects the growing relationship between AI model development and large-scale computing resources. Training advanced models requires access to extensive processor capacity, while serving models to users requires additional computing resources as demand grows.
By expanding its own facilities, SpaceX can potentially support internal AI development while creating capacity that may be offered to third parties.
However, operating large data centers is capital-intensive. The company must balance the speed of construction with the cost of acquiring hardware, securing electricity and maintaining the facilities.
What the Expansion Could Mean for Grok
A major purpose of SpaceX’s computing investment is to strengthen the infrastructure available for Grok, the AI model developed by xAI.
Larger training clusters can help developers run more extensive experiments, train models using greater computing resources and support the development of future model versions. Additional inference capacity can also help serve growing numbers of users and enterprise customers.
Musk said the planned 440,000-unit Vera Rubin cluster would help make Grok more competitive. That is an expression of management’s expectations, rather than independent evidence that the model will outperform competing systems.
The relationship between computing capacity and model quality is not automatic. Performance also depends on training methods, data quality, model architecture, software efficiency and the ability to use hardware effectively.
Nevertheless, access to large computing resources can be a competitive advantage, particularly when companies are racing to develop increasingly capable AI systems.
The Financial Challenge Behind a 13 GW Target
Building AI infrastructure at this scale requires substantial capital investment. SpaceX must fund processors, data-center construction, electricity systems, cooling equipment, networking and supporting infrastructure.
A report from Yahoo Finance citing an analyst assessment put the potential total spending associated with reaching the 13 GW target at approximately $350 billion. This is an analyst estimate, not a confirmed company-wide budget for the expansion.
The company’s third-quarter earnings discussion also highlighted the pace of planned spending. Reports said SpaceX expected fourth-quarter 2026 capital expenditure of approximately $35 billion to $40 billion, reflecting the intensity of its infrastructure buildout.
The actual cost of reaching 13 GW will depend on the type of computing hardware deployed, construction costs, the price and availability of power, equipment lead times and how much capacity SpaceX leases rather than owns.
For investors and the wider AI industry, the financial question is whether the additional computing capacity can generate enough revenue or strategic value to justify the investment.
The Bigger Picture
SpaceX’s 13 GW target illustrates how AI competition is becoming a race to build infrastructure as well as develop better models. Access to advanced processors matters, but so do electricity, cooling, construction expertise and the ability to deploy large computing clusters quickly.
The expansion could give SpaceX greater control over the infrastructure supporting Grok and potentially create opportunities to sell AI computing services to outside customers. At the same time, the scale of the plan introduces considerable execution and financing risks. The target should therefore be viewed as an ambitious company objective rather than a guaranteed outcome.
FAQs
1. What is SpaceX’s AI compute target for 2027?
SpaceX aims to have at least 13 GW of AI compute capacity online by the end of 2027. CEO Elon Musk said the company expects to add roughly 10 GW during the year.
2. How much AI compute capacity does SpaceX currently have?
SpaceX reported approximately 1.7 GW of compute capacity at the end of September 2026. Management expects to reach approximately 2.5 GW to 3 GW by the end of 2026.
3. What Nvidia hardware will SpaceX use?
SpaceX plans to continue deploying Nvidia GB300 systems and increasingly use Nvidia’s Vera Rubin architecture for its 2027 expansion. It is targeting a 440,000-unit Vera Rubin training cluster by the end of the second quarter of 2027.
4. Why does SpaceX need 20 GW of power and cooling infrastructure?
The company wants to ensure that electricity supply and cooling equipment do not become bottlenecks as it activates more processors. The 20 GW figure is a target for additional power and cooling infrastructure, not a claim that 20 GW of compute will necessarily be operational.
5. Will the expansion make Grok the world’s best AI model?
More computing resources could help SpaceX train and serve more capable models, but they do not guarantee superior performance. Model architecture, training data, software, research and evaluation also determine how well an AI system performs.
Looking Ahead
The next major milestones will be the completion of Colossus II, the continued deployment of Nvidia hardware and progress toward the 440,000-unit Vera Rubin training cluster. SpaceX will also need to demonstrate that its power, cooling and electrical infrastructure can keep pace with its computing ambitions throughout 2027.
The broader AI industry will be watching whether SpaceX can convert its infrastructure spending into competitive models and a sustainable computing business. If the company reaches its targets, it could become a more significant provider of AI computing capacity. If infrastructure or financing constraints slow the rollout, the timeline and economics of the expansion could change considerably.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
