SpaceX Reportedly Places $52 Billion Foxconn Order for NVIDIA GB300 AI Server Racks
SpaceX is reportedly making one of its biggest AI infrastructure moves yet, placing a massive order with Foxconn for NVIDIA GB300 AI server racks. The deal could dramatically expand the company’s computing power for artificial intelligence training and inference, strengthening its position in the fast-growing AI data center market.
According to a Taiwan-based report, SpaceX has awarded Foxconn an order worth around $52 billion for approximately 13,000 NVIDIA GB300 AI server racks. Shipments are expected to begin in the fourth quarter of 2026 and continue into the first quarter of 2027. Each rack is said to cost about $4 million, highlighting the enormous scale of the investment.
The order also marks another major win for Foxconn, which has been rapidly expanding its presence in the AI server business. The company is reportedly aiming to capture roughly 40 percent of the global AI server market by the end of 2026, and a deal of this size with SpaceX would significantly boost that ambition.
The purchase comes as demand for high-performance AI computing continues to surge across the technology sector. NVIDIA’s GB300 platform is expected to play a central role in next-generation AI workloads, offering the kind of accelerated performance needed for large-scale model training, real-time inference, and advanced data center operations.
SpaceX’s AI computing footprint has already grown substantially. As of May 2026, the company’s Colossus 1 data center reportedly supported more than 220,000 NVIDIA GPUs. That total included a mix of H100, H200, and roughly 30,000 GB200 AI accelerators.
Colossus 2 is said to be even larger, with more than 550,000 GPUs spread across NVIDIA GB200 and GB300 accelerators. The scale of these facilities shows how aggressively SpaceX is building out compute capacity as artificial intelligence becomes increasingly important across cloud services, automation, and advanced research.
SpaceX has also been signing major cloud and compute agreements with outside partners. Ahead of a potential public offering, the company reportedly secured a Cloud Service Agreement with Google, providing compute capacity equivalent to 110,000 NVIDIA GPUs, along with CPUs, memory, and related components. That agreement is said to be valued at around $920 million per month.
Another major agreement was reportedly signed with Anthropic, giving the AI company access to 220,000 NVIDIA GPUs, including H100, H200, GB200, and other accelerator types. That deal is valued at approximately $1.25 billion per month, or about $15 billion per year.
The Anthropic arrangement is believed to be tied mainly to the Colossus 1 data center. Because that facility uses a mixed collection of GPU types, it may be less ideal for training future versions of Grok AI compared with more uniform GPU clusters. Reports indicate that xAI consumes only about 11 percent of the total available FLOPs there.
By contrast, the Google agreement is primarily associated with the Colossus 2 data center, which appears to be positioned as a more advanced and powerful AI computing facility.
SpaceX also reportedly signed a $6.3 billion compute deal in June 2026 for the Colossus 2 data center with Reflection, a company linked to Pentagon-related work. That agreement further underlines the strategic importance of SpaceX’s expanding AI infrastructure.
If the reported Foxconn order moves forward as planned, it would represent one of the largest AI server rack purchases to date. It would also deepen the relationship between SpaceX, Foxconn, and NVIDIA at a time when demand for AI servers, GPU clusters, and high-performance data centers is reaching record levels.
For SpaceX, the $52 billion investment signals a long-term push into AI compute at extraordinary scale. For Foxconn, it strengthens the company’s position as a key supplier in the global AI hardware supply chain. And for NVIDIA, the deal would further confirm the dominant role of its GB-series accelerators in powering the next generation of artificial intelligence infrastructure.
