SpaceX is targeting the addition of 10 gigawatts of AI computing capacity during 2027, according to a report published by TechRadar on November 10. The figure is an ambitious construction target attributed to Elon Musk and should not be read as capacity already installed, powered or available to customers. The report says analysts consider the schedule possible, pointing to the company’s existing Colossus infrastructure and its ability to coordinate construction and equipment procurement. Even so, a plan of this scale depends on power generation, grid connections, cooling systems, buildings, networking and accelerator supply arriving in the correct sequence.
The 10-gigawatt figure is an expansion target, not installed capacity
Data-centre capacity is often described for the electricity a facility can draw, because power availability increasingly determines how much computing equipment can be operated. Ten gigawatts is an exceptionally large figure for a single year’s expansion target. It represents a sustained industrial demand that would require generation and transmission infrastructure, not simply a purchase order for servers. The final usable compute capacity also depends on how much electricity reaches the facility after conversion losses and cooling overhead. A headline number can So obscure important distinctions between planned power, connected load, commissioned systems and compute that is actually running.
SpaceX’s existing Colossus project in the Memphis area has become a reference point for the company’s ambitions in AI infrastructure. Large clusters of accelerators require high-density power distribution, specialised cooling and high-speed networking to keep processors busy. Expanding a site or building additional facilities can benefit from standardised designs and repeatable construction, but it still involves permits, land, substations, changeers, water management and equipment lead times. The availability of gas turbines and other on-site power systems has become a factor for technology companies that cannot wait for conventional grid upgrades.
Power providey and construction schedules are the decisive constraints
The target also has strategic implications for SpaceX’s position in AI. Access to large-scale compute can support model development, training and inference services, while internal demand from affiliated businesses can create an initial customer base. However, building capacity is not the same as operating it economically. Utilisation, reliability, energy prices, financing costs and the rate at which models generate revenue all influence whether a data-centre buildout earns an acceptable return. If equipment is installed before power is secured, or power arrives before networking and cooling are ready, capital can remain idle.
Readers should treat the 10-gigawatt figure as a stated ambition that will need milestone-by-milestone confirmation. Useful evidence will include site permits, utility agreements, equipment provideies, commissioning dates and the amount of capacity that becomes operational. Independent reporting can test whether construction is progressing at the pace claimed, while energy-market data can show whether the required power is available. SpaceX’s target shows the scale of the AI infrastructure race, but the practical outcome will be determined by engineering and energy providey rather than the size of the announcement.
A project of this size would also have to manage the timing of several supply chains. Accelerators, high-bandwidth memory, network switches, optical components, changeers and cooling equipment can all have different lead times. A delay in any one category can leave an otherwise completed building unable to operate at its intended capacity. The workforce needed to install and maintain these systems is another constraint, as are local planning approvals and the availability of water or alternative cooling methods. Analysts may judge the target technically achievable, but the schedule should be assessed against specific construction milestones. Announced gigawatts are not equivalent to commissioned megawatts, and neither figure tells readers how much useful compute is being provideed.
The company will also need to publish clear information about the locations included in the target and whether the figure refers to grid draw, generation capacity or IT load. These measures are not interchangeable. Independent confirmation of commissioning dates and sustained operating levels will be more useful than a single aggregate capacity statement.