SpaceX lands $6.3 billion AI compute deal With Reflection AI to power open-source models

SpaceX has signed a multiyear AI compute agreement with Reflection AI worth up to $6.3 billion, giving the open-source startup access to Nvidia GB300 chips and giving investors another look at how Elon Musk’s company plans to turn Colossus into a serious infrastructure business.

SpaceX is no longer just building rockets, satellites, and an AI chatbot. It’s now selling access to the computing backbone behind them.

The company has signed a major computing-power agreement with Reflection AI, the open-source artificial intelligence startup, in a deal that shows how far SpaceX is pushing beyond its traditional businesses into the AI infrastructure market. Under the agreement, Reflection will get immediate access to Nvidia GB300 chips from SpaceX’s Colossus system and will pay $150 million per month starting July 1, 2026, through 2029, according to materials viewed by CNBC.

If the contract runs its full course, the payments would total about $6.3 billion. After the first three months, either side can walk away with 90 days’ notice.

“The payments would total about $6.3 billion if the agreement runs through the end of its term. Either company can end the contract with 90 days’ notice after the first three months,” CNBC reported.

The deal comes about three months after reports that Reflection, the New York-based startup founded in 2024 by former Google DeepMind researchers Misha Laskin and Ioannis Antonoglou, was in talks to raise $2.5 billion at a $25 billion valuation.

Reflection AI Taps SpaceX’s Colossus in $6.3 Billion Open-Source AI Compute Deal

The headline number is big, but the larger story sits with SpaceX. This is another sign that Colossus is becoming more than internal muscle for Grok and Musk’s broader AI ambitions. It is turning into a product SpaceX can sell.

That matters for a company fresh off a record IPO and trying to convince investors it can become more than a launch business and a satellite internet provider. SpaceX has already used Colossus to strike compute-related deals with Anthropic, Google, and Cursor. Reflection now joins that list, adding a different kind of customer at a moment when open-source AI is starting to look less like an ideological alternative and more like a strategic one.

Reflection is pitching itself as part of that shift. The startup, last valued at $25 billion, is trying to build American open-source AI models that can compete with systems from OpenAI, Anthropic, and Google, with a pitch centered on control, transparency, and enabling customers to run and inspect models more directly.

That message has gained fresh traction in recent weeks after Anthropic cut off access to Fable and Mythos, a move that reignited debate over the risks of depending too heavily on closed-model providers for important workloads. For open-model companies, the episode gave them a clear argument: if a model becomes central to your business or government operations, relying on a provider that can change terms or cut access overnight carries real risk.

Reflection leaned into that point in a statement on the deal.

“Recent events highlight how important open source is to the AI ecosystem, with more nations and enterprises recognizing the risks and costs associated with exclusively depending on closed models,” a Reflection spokesperson said.

The company said the agreement will give it more compute capacity to accelerate what it calls “American open intelligence.”

Reflection has not yet released a public, open-source frontier model, but it has been building ties with government and national security customers. The startup is working with the Department of Energy’s Genesis Mission and has participated in broader Pentagon AI efforts, giving it a profile that differs from that of a typical consumer AI startup chasing chatbot traffic.

For SpaceX, that distinction matters. Reflection is not just another customer renting expensive GPUs. It gives SpaceX exposure to a part of the AI market that is attracting growing interest from governments, defense agencies, and enterprises seeking greater control over how models are built and deployed.

It is another reminder that in the AI race, compute has become its own form of leverage. Access to advanced Nvidia chips remains one of the biggest constraints for companies trying to train and serve frontier models at scale. The companies that control those chips, the data centers that house them, and the power needed to run them are increasingly shaping the market just as much as the companies building the models themselves.

SpaceX appears to understand that well. By opening Colossus to outside customers, it is starting to look less like a company building AI for its own use and more like one trying to become a major compute landlord for the industry.

That could become one of the most important parts of its post-IPO story. Investors have been watching whether SpaceX can stretch beyond rockets, Starlink, and Musk’s AI ambitions into something larger: a company that owns scarce AI infrastructure and rents it to the labs, governments, and enterprises racing to secure it.

Reflection AI Founders