en.Wedoany.com Reported - SpaceX, the American space exploration company, has reached a large-scale computing cooperation agreement with Reflection AI, an open-source artificial intelligence startup. Under the agreement, Reflection AI will use hardware at SpaceX's Colossus 2 data center near Memphis, Tennessee, and pay SpaceX $150 million per month starting July 1, 2026. If the agreement runs through the end of 2029, the total amount will be approximately $6.3 billion.
Reflection AI will gain immediate access to NVIDIA GB300 chips and supporting hardware. The GB300 is a high-end AI chip designed for advanced AI training and inference tasks, capable of supporting large model training, model inference, open-weight model iteration, and high-concurrency experiments. For an open-source AI startup, stable and large-scale computing power is a crucial foundation for competing with closed-source frontier model labs.
Founded in 2024 by two former Google DeepMind researchers, Reflection AI positions itself as an open-source AI model company. The company emphasizes an open-weight model approach, aiming to leverage stronger computing power to drive the development of next-generation open models. As enterprises, governments, and developers grow increasingly concerned about the risks, costs, and controllability of relying on closed-source models, open-source model companies are accelerating their competition for computing resources, R&D talent, and ecosystem developers.
The Colossus 2 data center provided by SpaceX was originally a key component of Musk's AI computing infrastructure. As SpaceXAI-related infrastructure expands, SpaceX has begun leasing out some of its AI chips and data center resources, offering large-scale computing capabilities to AI companies. This agreement with Reflection AI also indicates that SpaceX is moving beyond being a space launch and satellite communications company to further enter the AI infrastructure supply chain.
The value of this agreement is lower than some of SpaceX's previous computing partnerships with companies like Anthropic and Google, but it still ranks among the largest disclosed infrastructure commitments in the open-source AI field. According to TechCrunch, either party can terminate the contract after the first three months with 90 days' notice. Therefore, the approximately $6.3 billion figure should be understood as the total upper limit if the agreement runs to its full term, rather than a one-time recognized revenue.
For Reflection AI, securing computing power is just the starting point of the competition. To gain market recognition, open-source AI models need to build sustained advantages in model capability, inference cost, context length, tool calling, multimodal capabilities, safety, and developer ecosystem. Large-scale GB300 resources will enhance its model training and experimental iteration capabilities, but whether this translates into usable products and community influence will depend on the pace of model releases and actual performance.
For SpaceX, AI computing leasing is becoming a new infrastructure business. High-performance AI chips, data center power, cooling systems, and network interconnects are becoming some of the scarcest resources for frontier AI companies. If SpaceX can continue to secure large client contracts, its Colossus series data centers could evolve from internal AI resources into a computing platform available for external sale.
Key areas to watch going forward include Colossus 2's delivery capacity, Reflection AI's model release progress, whether the contract is sustained, and whether SpaceX continues to expand its external AI computing leasing scale. As AI companies' demand for computing power continues to rise, the boundaries between aerospace, cloud computing, chips, and data center resources are further blurring.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
