SpaceX inks compute deal with Reflection AI, an open-source AI lab

In a significant move within the AI landscape, Reflection AI, an emerging open-source AI startup, has secured a lucrative partnership with SpaceX. This collaboration will see Reflection AI investing $150 million monthly starting July 1, 2026, continuing through 2029, to gain access to Nvidia’s cutting-edge GB300 AI chips and other essential hardware from SpaceX’s expansive Colossus 2 data center located near Memphis, Tennessee. The total value of this agreement could reach up to $6.3 billion, offering Reflection a substantial boost in computing power as they advance their AI initiatives. Notably, both parties retain the flexibility to terminate the contract with a 90-day notice after the initial three months. This deal, while smaller than SpaceX’s existing contracts with other AI giants like Anthropic and Google, which command $1.25 billion and $920 million per month respectively, marks a critical development for Reflection AI. Reflection AI aims to promote its open-weight AI methodology, positioning itself as a viable alternative to the closed models of industry leaders such as Anthropic and OpenAI. The startup's strategy, which includes the public release of trained parameters, has gained traction following recent government restrictions on closed models such as Anthropic’s Fable and Mythos. Founded in 2024 by former Google DeepMind researchers, Reflection AI describes this compute arrangement as one of the most significant commitments to open AI infrastructure to date. A spokesperson for the company noted, “The recent shifts in the AI landscape underscore the necessity of open-source solutions, as more organizations recognize the risks of relying solely on closed systems. Our partnership with SpaceXAI exemplifies Reflection’s pivotal role within the AI frontier, and increased computing resources will empower us to develop superior open models at scale.” The Colossus data center, originally developed by xAI—Elon Musk's initiative now integrated within SpaceX—was designed for internal AI projects. However, as those objectives have slowed, SpaceX has capitalized on its valuable AI chip inventory by leasing it to leading AI laboratories around the globe.