At a SpaceX employee meeting, Musk estimated that the power capacity of xAI’s data centers will increase approximately sevenfold by the end of 2027, reaching 10 GW, and claimed that xAI has already built the world’s most powerful AI training cluster, with plans to expand the cluster to ten times its current size by the end of next year. If each watt is valued at $30 to $50, the 10 GW of computing power coming online by the end of next year could generate annual revenues of $300 billion to $500 billion for xAI. SpaceX AI currently operates two data centers—in Memphis and Southaven—with a combined rated power consumption of 1.4 GW, and plans to scale this to 10 GW within three years. This means that SpaceX, once known primarily for rockets, is transitioning into an AI infrastructure company; however, a 10 GW power demand will also pose significant challenges to the grid and energy supply.

Author and source: AIBase

$30–50 per watt, aiming for annual revenue of $500 billion

At a SpaceX employee meeting, Musk estimated that the power capacity of SpaceXAI (formerly xAI)’s data centers will increase by approximately sevenfold by the end of 2027, reaching 10 GW. This means that the company renowned for rockets is rapidly expanding its AI computing infrastructure at an almost exponential pace. Musk further claimed that SpaceXAI has already built the “world’s most powerful AI training cluster” and plans to scale it up to roughly ten times its current size by the end of next year.

It’s important to clarify that GW (gigawatt) is a unit of power, with 1 GW equaling one billion watts. Data centers use it to describe power supply capacity or rated power consumption, but power capacity does not equate to actual computing power. Electricity is consumed collectively by systems such as cooling, lighting, power distribution, networking, storage, and servers; therefore, 10 GW represents more of an upper limit on infrastructure energy consumption than pure computational output.

Musk crunched the numbers on this power gamble: if each watt is worth approximately $30 to $50, and 10 GW of AI computing capacity comes online by the end of next year, xAI could generate annual revenues of $300 billion to $500 billion. At current exchange rates, this range equals approximately RMB 2.03 trillion to 3.38 trillion—nearly equivalent to the GDP of a mid-sized nation.

Supporting this goal is the current infrastructure: SpaceXAI’s two data centers in Memphis and South Haven have a combined rated power consumption of 1.4 GW, planned to increase to 10 GW by the end of 2027—that is, a sevenfold expansion in approximately one and a half years. This aligns with Musk’s earlier prediction that “in five years, 99% of SpaceX’s value will come from AI”—as rocketry is overtaken by the flood of computational power, SpaceX has effectively transformed into an AI infrastructure company. However, a 10 GW power demand will impose immense strain on the grid and generate a massive carbon footprint; the battle between computational expansion and energy constraints may prove more decisive to the success of this gamble than the models themselves.

","author":"ME Selection","type":"Repost"}