SpaceX has a new tenant for its mountain of AI chips. Open source startup Reflection AI has signed a compute deal worth up to $6.3 billion, paying SpaceX $150 million a month from July 1, 2026 through 2029 for access to Nvidia’s latest GB300 chips, according to TechCrunch AI. The company told TechCrunch the hardware sits inside SpaceX’s Colossus 2 data center near Memphis, Tennessee.
This is the third such deal SpaceX has struck, after Anthropic and Google. What stands out is the pattern: a rocket company has quietly become one of the most important landlords in AI.
What the deal covers
The numbers tell the story of where Reflection sits in the pecking order.
- Reflection AI: $150 million per month, up to $6.3 billion total
- Google: $920 million per month
- Anthropic: $1.25 billion per month
All three contracts run through July 2029. Each side can walk away with 90 days’ notice after the first three months, so the headline totals are ceilings, not guarantees. Elon Musk has publicly downplayed the three-year framing, stressing that the contracts can be canceled at any time.
Reflection’s deal is the smallest of the three. But for a startup founded in 2024 by two former Google DeepMind researchers, landing any nine-figure monthly compute commitment is a serious signal. It’s the company’s first compute deal, and Reflection calls it one of the largest open AI infrastructure commitments announced to date.
Why Reflection is making noise about “open”
Reflection used the announcement to push its open-weight strategy, per TechCrunch AI. Open-weight models publicly release their trained parameters, which lets anyone download, run, and build on them. That’s the opposite of the closed approach taken by Anthropic and OpenAI, where the weights stay locked behind an API.
The timing matters. TechCrunch AI reports that open-weight models have drawn more attention after the U.S. government banned Anthropic’s closed models, Fable and Mythos. When a government can cut off access to a closed system overnight, the case for models you actually control gets a lot stronger.
Reflection leaned into exactly that. “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 spokesperson said in a statement to TechCrunch. The company added that “more compute means more runway to build the world’s best open models at scale.”
How SpaceX ended up in this business
Here’s the twist. The Colossus data center was originally built by xAI, Musk’s AI company, for its own model training. xAI is now part of SpaceX. As those internal AI efforts faltered, SpaceX did the practical thing: it started renting out its valuable chip stockpile to the top labs that needed compute faster than they could build it.
So SpaceX isn’t really pivoting into AI. It’s monetizing hardware it already owns while the rest of the industry scrambles for Nvidia supply. With Anthropic, Google, and now Reflection on the books, those rentals are turning into a meaningful revenue line.
Why it matters
Three things to take away from this:
- Compute is the real moat. Reflection didn’t announce a model. It announced access to chips. In the current race, securing GB300 capacity through 2029 is the headline, because everything else depends on it.
- Open-weight is gaining ground as a hedge. The government ban on Anthropic’s closed models gave the open camp a concrete argument: control and continuity. Expect more nations and enterprises to weigh that risk.
- The supplier map is shifting. A launch company sitting on a chip hoard is now a frontier-AI power broker. That tells you how scarce this hardware really is.
Watch whether other open labs follow Reflection to SpaceX, and whether those 90-day exit clauses actually get used as the compute market keeps moving. For the full breakdown, see the original report at TechCrunch AI.