SpaceX Pushes Custom Starlink Silicon

The custom silicon is critical for solving the physics of direct-to-cell service, including managing the significant Doppler shift from satellites moving at tens of thousands of kilometers per hour and compensating for the low antenna gain and transmit power of standard mobile phones. SpaceX's silicon design efforts are supported by a decade-long partnership with STMicroelectronics. The two companies co-design custom BiCMOS chips for the phased-array antennas used in millions of user terminals and over 10,000 satellites, with STMicro scaling production to deliver over 5 million chips per day. On the ground, Starlink gateways that connect the satellite constellation to the terrestrial internet rely on high-performance computing. These gateways utilize AMD EPYC processors, which are designed for the high I/O and compute density required to process massive data volumes for 5G-level networking and cloud-native environments. The collaboration with Samsung's System LSI division is focused on creating an AI-enhanced Exynos modem. This chip reportedly integrates a neural processing unit (NPU) to predict satellite trajectories and optimize signal links in real time, a foundational technology for future 6G non-terrestrial networks (NTN). AMD's involvement extends into orbit as well. CEO Lisa Su confirmed that SpaceX is using AMD's Versal AI Core adaptive SoCs in its latest generation of broadband satellites, enabling in-space AI inference and real-time data processing. This strategy of designing critical components in-house is a core tenet of SpaceX's vertical integration model. This approach began after early attempts by Elon Musk to buy refurbished Russian rockets failed, forcing the company to develop its own hardware from the ground up to control innovation speed and cost. Future capabilities are tied directly to this hardware roadmap. The next-generation V2 and V3 Starlink satellites, which depend on this custom silicon, are expected to provide at