KUALA LUMPUR — When the first payment lands July 1, Reflection AI will be buying time on Nvidia GB300 chips housed inside SpaceX’s Colossus data center in Tennessee. $150 million a month buys a lot of processing power. The deal, which could run as high as $6.3 billion total, ties an Elon Musk company to an AI start-up barely a year old. Reflection AI was founded in 2024.
Its two founders came from Google DeepMind. Their stated goal: build open-source foundation AI models that lead to a “superintelligence” anyone can use.
That is a direct echo of the philosophy behind the internet and Linux — freely available building blocks that reshaped computing from the ground up. The structure of this contract matters. For the first three months, neither side can walk away.
After that, either Reflection AI or SpaceX can terminate with 90 days’ notice. That is an unusual degree of flexibility in a deal this large.
It suggests both companies know they are operating in terrain that shifts fast. Locking in for years might feel like a bet on a specific future that never arrives. A three-month minimum with a 90-day exit gives both parties room to pivot.
What does a space company gain from hosting AI hardware? SpaceX already operates the Colossus data center. Adding a tenant paying $150 million a month turns a fixed cost into a revenue stream.
The chips are Nvidia’s GB300s — the latest high-end hardware for training and running large AI models. SpaceX gets to keep those chips warm and running, paid for by someone else.
For Reflection AI, the math is straightforward. Building a superintelligence requires compute. Lots of it.
Buying and operating their own data center would take months or years. Renting from an existing facility with the best chips available means they start work immediately.
Time matters in AI. A six-month delay in compute access could mean watching another lab ship the breakthrough first. The open-source angle is what makes this different from other big AI deals.
Most frontier labs keep their models secret. Reflection AI plans to release theirs openly. That choice has consequences.
Open models can be audited, modified, and deployed by anyone. Governments, small businesses, researchers — all get access.
The report notes this could benefit communities across Asia in education, transportation, and communication. The logic is simple: if the best AI tools are free, they spread faster to places that cannot afford to build their own. There are risks.
Open-source models can be misused. A superintelligence accessible to anyone raises safety questions no one has fully answered.
The founders came from DeepMind, a lab that pioneered AI safety research. Whether that background shapes their approach to openness remains to be seen. The deal also signals something about the AI industry’s trajectory.
Compute is becoming a commodity you rent, not a fortress you build. The winners may not be the teams with the best algorithms but the ones who secure the best hardware contracts first. SpaceX, a rocket company, now plays in that game.
Reflection AI, a start-up with no products yet, just bought a seat at the table. July 1 is the start date.
That is eight days from the announcement. The speed suggests both sides have been working on this for a while. The clock is now running on a deal that could reshape how AI gets built — or end quietly after three months with a 90-day notice.
{“@context”: “https://schema.org”, “@type”: “NewsArticle”, “headline”: “Reflection AI to Pay SpaceX $150 Million Monthly for Nvidia Chips in Data Center”, “description”: “KUALA LUMPUR — When the first payment lands July 1, Reflection AI will be buying time on Nvidia GB300 chips housed inside SpaceX’s Colossus data…”, “datePublished”: “2026-06-23T13:42:00+00:00T09:30:00+00:00”, “author”: {“@type”: “Person”, “name”: “Erika Johnson”}, “publisher”: {“@type”: “Organization”, “name”: “InfoPulse Today”, “url”: “https://newsfeed.infopulsetoday.com/”}}
