SpaceX taps NVIDIA for 1M-satellite AI plan

SpaceX has expanded its partnership with NVIDIA to power Starmind, a proposed network of orbital data centers designed to process artificial intelligence workloads in space.

• SpaceX will use NVIDIA's Vera Rubin platform for its planned Starmind satellite network.
• The company has requested FCC approval for up to one million orbital data-center satellites.
• NVIDIA says its Space-1 module provides up to 25 times the AI compute of an H100 GPU.
• SPCX gained 9.8% , while NVIDIA shares rose 2.5% following the announcement.

SpaceX plans to equip its Starmind satellites with NVIDIA's Rubin graphics processing units and Vera central processing units. The hardware forms part of NVIDIA's Space-1 platform, which was developed for AI processing and other computing workloads in orbit.

The partnership expands NVIDIA's list of space-computing customers after the chipmaker introduced the platform in March. Its initial launch partners included Aetherflux, Axiom Space, Kepler Communications, Planet Labs, Sophia Space and Starcloud.

SpaceX was not included in the original announcement but has now joined the companies working with NVIDIA on orbital computing. Musk later said SpaceX would build its AI infrastructure exclusively on NVIDIA platforms and described Vera Rubin as the strongest available option.

The companies also plan to jointly design computing payloads for future satellites, according to reports following SpaceX's investor call.

NVIDIA says its Space-1 Vera Rubin module combines GPUs, CPUs and high-bandwidth connections in a system designed for the power and weight limits of satellites. It can run large language models and process data in orbit instead of transmitting all raw information back to Earth.

SpaceX's broader plan remains subject to regulatory approval in the United States. The company filed an application in January seeking permission to launch and operate as many as one million non-geostationary satellites.

The proposed system would operate between 500 and 2,000 kilometers above Earth. SpaceX said optical links would connect the satellites through a high-capacity network capable of moving data between orbital computing nodes.

The Federal Communications Commission accepted the application for filing in February and opened it for public comment. That procedural decision did not constitute final approval, correcting reports that the FCC had already authorized the full constellation.

The requested scale is far larger than the existing satellite population around Earth. It could also face questions involving orbital congestion, collision risks, radio interference and effects on astronomical observations.

SPCX closed Tuesday at $125.33, up $10.80, or 9.43%, after investors responded to the NVIDIA partnership ahead of SpaceX's earnings report.

However, the stock reversed course after the closing bell. SPCX fell 6.82% to $116.78 in after-hours trading as investors assessed the company's quarterly results and AI spending plans.

NVIDIA shares gained 3.03% to $212.91 during Tuesday afternoon trading. The move reflected investor expectations that orbital data centers could create another market for the chipmaker's AI computing hardware.

Starmind could allow SpaceX to combine its launch capabilities, Starlink communications network and AI operations within one infrastructure project. However, deploying orbital data centers would require large investments in satellites, launches, power generation and thermal management.

SpaceX must also demonstrate that the system can operate safely alongside existing spacecraft before receiving final FCC authorization. Even if approved, the company would likely deploy the network in stages rather than launching the full requested number.

For U.S. investors, the NVIDIA agreement provides a clearer hardware path for SpaceX's AI strategy. The next tests will be regulatory progress, the cost of building the constellation, and whether orbital computing can produce enough revenue to justify the required spending.