SpaceX is becoming the infrastructure backbone for frontier AI labs. Three major labs now train on a single Musk-controlled data center. The open-source angle ties directly to the Anthropic export ban and has real implications for SaaS teams evaluating model alternatives.
- Reflection AI will pay SpaceX $150 million a month from July 1, 2026 through 2029, for Nvidia GB300 chips at the Colossus 2 data center near Memphis, Tennessee.
- The deal is worth up to $6.3 billion and makes Reflection AI the third major AI lab renting compute from SpaceX, after Anthropic ($1.25B/month) and Google ($920M/month).
- Either party can cancel with 90 days’ notice after the first three months.
- Reflection AI is an open-weight lab, founded in 2024 by two former Google DeepMind researchers, and positioned this deal as a direct response to the US export ban on Anthropic’s closed models.
What Happened
Reflection AI, an open-source AI startup, signed a compute deal with SpaceX worth up to $6.3 billion, paying $150 million a month starting July 1, 2026 through 2029 for access to Nvidia GB300 chips at SpaceX’s Colossus 2 data center near Memphis, Tennessee, the company told TechCrunch.
The deal is Reflection AI’s first compute contract and makes it the third major AI lab to rent capacity from SpaceX, after Anthropic at $1.25 billion a month and Google at $920 million a month. All three contracts run on the same timeline through July 2029.
Either company can cancel with 90 days’ notice after the first three months. SpaceX’s Colossus 2 was originally built by xAI for its own AI work and is now being leased to outside labs as xAI’s internal pursuits have slowed.
Reflection AI used the announcement to frame its open-weight model strategy as a direct response to the US export ban on Anthropic’s closed Fable 5 and Mythos Preview models, saying more nations and enterprises are recognizing the risks of depending exclusively on closed models.
Why It Matters
SpaceX is quietly becoming the compute backbone of the frontier AI race. Three of the largest AI labs now train on a single Musk-controlled data center, concentrating a significant share of the world’s most advanced AI training capacity under one roof. For Reflection AI specifically, a multi-billion-dollar infrastructure commitment gives the open-weight movement a credibility signal it has not had at this scale before.
The 90-day cancellation window cuts both ways. The deal is large on paper but neither party is fully locked in beyond a quarter. Reflection AI has also yet to ship a model that genuinely matches the closed labs it is competing against. The “open-source wins after export ban” framing is convenient positioning, not a proven outcome. Relve, an AI trends intelligence platform, is tracking whether open-weight labs can convert infrastructure investment into model performance this year.
Bottom Line
Watch what Reflection AI ships in the next two quarters. A $150 million monthly compute bill demands visible output. If the lab releases a genuinely competitive open-weight model before the end of 2026, this infrastructure bet validates itself and changes the vendor calculus for SaaS teams.
For engineering teams already evaluating open-weight models as a closed-model alternative, the Reflection AI roadmap is worth adding to your watchlist now. The compute is in place. The question is what gets built on it.
