An Nvidia-backed AI startup will pay $6 billion for access to a SpaceX supercomputer

Elon Musk’s aerospace and AI company, SpaceX, has signed a multibillion-dollar agreement with Reflection AI—a startup supported by, among others, Nvidia, to provide it with computing power, Bloomberg reports, noting that this deal is the latest in a series that Musk’s company has entered into as it seeks to become an AI infrastructure provider.

Previously—ahead of its initial public offering—SpaceX had signed similar contracts with Alphabet and Anthropic. Both deals gave them access to the computing power of SpaceX’s supercomputers, which Bloomberg and CNBC are already calling Musk’s data centers.

Under the contract with Reflection AI, the Nvidia-backed startup agreed to pay SpaceX’s AI division—now known as SpaceXAI—$150 million per month, starting in July 2026 and continuing through 2029. Thus, the total value of the deal, if it lasts until 2029—and is not terminated earlier—can be estimated at approximately $6.3 billion, according to CNBC’s calculations. Reflection AI will pay SpaceX this amount for the right to use SpaceX’s supercomputer, Colossus 2, located in Memphis, Tennessee. Under the terms of the deal, both companies have the right to terminate the agreement by providing 90 days’ notice.

The agreement demonstrates how SpaceX is utilizing its large-scale computing infrastructure following its record-breaking IPO. The company originally built the Colossus supercomputer to power, among other things, its own chatbot, Grok—an AI model that competes with ChatGPT. Now, however, SpaceX is using this infrastructure, among other things, to sell computing power to third-party companies engaged in AI development, notes CNBC. Bloomberg points out that under similar contracts, Google has agreed to pay Musk’s company approximately $30 billion by mid-2029, while Anthropic has agreed to pay $45 billion over the same period. Reflection is another such client for SpaceX, though “strategically different,” CNBC points out. The reason is that the company specializes in developing open-source AI models, which provide users with virtually free access to them and accelerate their adoption among developers and companies, with the ability to fine-tune them for specific needs. Such open-source AI models drew attention after Anthropic restricted access in June to its cutting-edge—proprietary — Fable and Mythos models in June, prompting questions from users (including government agencies) about the risks of using such models for mission-critical tasks, CNBC notes.

Reflection, an American AI startup backed by Nvidia, has placed its bet specifically on open-source AI models. Reflection’s latest valuation stood at $25 billion, according to CNBC. The company aims to develop open-source AI models in the U.S. that can compete with cutting-edge systems from OpenAI, Anthropic, and Google—while also offering governments and businesses greater flexibility compared to proprietary AI models.

While Reflection has not yet released any public open-source models, it is steadily expanding its presence among government customers and clients in the national security sector, CNBC notes. In particular, the company is collaborating with the U.S. Department of Energy on the Genesis mission and is participating in broader Pentagon initiatives in the field of AI.

This article was AI-translated and verified by a human editor