SpaceX is scaling its computational power to an unprecedented level, aiming to quintuple its capacity by the end of 2027. According to Elon Musk, this expansion will push power consumption beyond two gigawatts, with Nvidia serving as the exclusive chip provider through the Vera Rubin platform.

Add AlexTech.ai asPreferred Source on Google

The Road to Two Million GPUs

The current Colossus clusters utilize approximately 540,000 GPUs, including H100 and GB series. To meet the new targets, projections suggest a requirement of well over one million units; if the expansion relies solely on Rubin architecture, that number could exceed two million. This massive build-out follows the February 2026 merger of xAI into SpaceX, valuing the combined entity at $1.25 trillion.

Starmind: Datacenters in Orbit

The collaboration with Nvidia extends beyond Earth. The two companies are co-designing the compute payload for Starmind, a planned network of orbital data center satellites. Each satellite will feature Rubin GPUs and Vera CPUs, delivering a peak compute payload of 250 kW. This move solidifies the shift toward space-based computing, a frontier where US and Chinese firms are currently outpacing other global players.

Vertical Integration via Tesla Megapacks

To manage the volatile power demands of AI training and inference, SpaceX has invested $329 million this year in Tesla Megapacks. These industrial batteries smooth out energy peaks and provide critical backup power, illustrating the deep synergy between Musk's companies.

Financial Trade-offs

While revenue for SpaceX's AI segment tripled in Q2 2026, reaching $2.56 billion, operating losses remain steep at $1.26 billion for the quarter. Much of this growth is driven by cloud contracts leasing Colossus capacity, with a single major customer contributing roughly $1.52 billion.