SpaceX has finalized a substantial purchase order with Foxconn for NVIDIA GB300 AI server racks, aiming to significantly boost its artificial intelligence computing capabilities.
Sources in Taiwan report that the order involves a $52 billion investment for approximately 13,000 server racks, each equipped with NVIDIA GB300 AI technology and valued at about $4 million. The agreement covers 936,000 NVIDIA GPUs, with the first shipments anticipated between the fourth quarter of 2026 and the first quarter of 2027. It remains uncertain whether Foxconn will be the sole supplier or if SpaceX will involve additional rack assemblers in the procurement process.
Foxconn forecasts it will have close to 40 percent of the global AI server market by the end of 2026; notably, the latest SpaceX contract was not factored into this estimate, which could lead to increased market share for the company. The recent agreement also marks Foxconn’s entry into a market segment historically led by Dell and Supermicro.
Currently, SpaceX operates two major data centers. The Colossus 1 facility employs over 220,000 NVIDIA GPUs across various models, including the H100, H200, and around 30,000 GB200 accelerators. The newer Colossus 2 data center houses more than 550,000 GPUs focusing on GB200 and GB300 accelerators.
Leading up to its initial public offering, SpaceX secured a cloud service contract with Google, providing compute resources equivalent to 110,000 NVIDIA GPUs for $920 million monthly, mainly benefitting the Colossus 2 facility. Additionally, SpaceX reached an agreement with Anthropic to supply it with access to 220,000 NVIDIA GPUs for $1.25 billion per month; this arrangement applies to Colossus 1, where xAI utilizes 11 percent of available processing power.
