Musk: Grok 4.6 to Launch Next Week With SpaceX Data, Exclusive Nvidia Computing Power

TradingKey - On August 4, Eastern Time, SpaceX ( SPCX) held its first earnings conference call since listing. SpaceX CEO Elon Musk revealed the latest progress in its AI business during the call: first, model iteration is accelerating, with Grok 4.6 to be released next week, followed by Grok 4.7; second, its computing power is fully bound to Nvidia ( NVDA ), and extends into space.

About a month ago, SpaceX AI just released Grok 4.5. Jointly trained by xAI and Cursor, this model focuses on code generation while also handling professional scenario tasks such as finance and law. Gwynne Shotwell, President and Chief Operating Officer of SpaceX, revealed during the call that token consumption of the model tripled after the release of Grok 4.5.

The more critical information lies in the training data. Musk explicitly stated during the call: "We will integrate the entire dataset of SpaceX—basically all the data SpaceX has ever generated—into the training of Grok 5, which will be launched at the end of the year." According to the plan, Grok 5, which incorporates this data, is expected to be launched by the end of this year.

On the computing power front, Musk confirmed during the call that SpaceX will exclusively adopt Nvidia's AI computing architecture, building AI services entirely based on the Vera Rubin system. He flatly stated that Nvidia's Vera Rubin is "currently the best AI computer on the market."

In terms of computing power scale, the target SpaceX presented this time is more aggressive than before: online computing power is expected to exceed 2 gigawatts by the end of this year and approach 10 gigawatts by the end of next year. Musk stated that the current focus of resource investment will be on power and cooling equipment rather than the GPUs themselves, and it is expected that about 15 gigawatts of online computing power can be achieved by the end of 2027.

This computing power will not only be deployed on the ground but will also extend into space. SpaceX has already co-developed the computing power payload for the Starmind AI 1 satellite with Nvidia, with each 'Starmind' satellite carrying Nvidia Rubin GPUs and Vera CPUs to provide data center-grade space computing capabilities. The first batch of Starmind satellites equipped with this system is scheduled to begin launching next year, extending AI computing power to low Earth orbit.

[Source: X]