SpaceX Secures $6.3B Compute Deal With Open-Source Startup Reflection AI
Reflection AI commits to $150M monthly for Nvidia GB300 chips at SpaceX's Memphis data center through 2029, positioning open-weights models as alternative to closed frontier labs.
Last verified:
Reflection AI Secures $150M Monthly Compute Commitment From SpaceX
Reflection AI, a 2024-founded startup led by former Google DeepMind researchers, inked a $6.3 billion compute agreement with SpaceX on June 22, according to TechCrunch. Beginning July 1, 2026, Reflection AI will pay $150 million monthly through 2029 for exclusive access to Nvidia GB300 chips and supporting infrastructure at SpaceX’s Colossus 2 data center near Memphis, Tennessee. The contract permits either party to terminate with 90 days’ notice after the first three months.
Positioning Against Closed-Model Competition
The deal amounts to a fraction of SpaceX’s other announced commitments: Anthropic pays $1.25 billion monthly and Google pays $920 million monthly under agreements running the same timeline through July 2029, per TechCrunch. Despite the lower monthly outlay, Reflection AI framed the arrangement as validation of its open-weights strategy. According to TechCrunch, a Reflection AI spokesperson said in an emailed statement that “our deal with SpaceXAI signals Reflection’s strategic importance within the frontier AI ecosystem, and more compute means more runway to build the world’s best open models at scale.”
The timing aligns with renewed regulatory scrutiny on proprietary systems. TechCrunch reports the U.S. government has banned Anthropic’s closed models, Fable and Mythos, and notes that this action has elevated attention to open-weights alternatives that release trained parameters publicly.
Infrastructure Origins and SpaceX’s Pivot
Colossus 2 was originally constructed by xAI, Elon Musk’s in-house AI venture now operating as part of SpaceX, for internal AI research. As xAI’s own efforts have stalled, SpaceX has leveraged its hardware capacity as a standalone business, renting chips to multiple frontier labs including Anthropic and Google alongside Reflection AI.
Why This Matters
Teams evaluating open-weights versus closed-model architectures now have a committed third-party compute pathway that lacked material backing six months ago. Reflection AI’s ability to secure multi-billion-dollar infrastructure funding signals investor confidence in open-source AI development, potentially reshaping how startups choose between licensing proprietary models and training custom open-weights alternatives. For SpaceX, the strategy converts stranded xAI hardware into recurring revenue—a model that works only if frontier labs continue competing for scarce Nvidia GPU capacity through 2029.
Frequently Asked Questions
How does Reflection AI's deal compare to other SpaceX compute contracts?
According to TechCrunch, Reflection AI pays $150M monthly compared to Anthropic's $1.25B monthly and Google's $920M monthly—making Reflection's commitment the smallest of the three announced SpaceX arrangements, though still valued at $6.3B over three years.
What regulatory backdrop prompted this deal?
The U.S. government has banned Anthropic's closed models, Fable and Mythos. TechCrunch reports this regulatory action has increased attention to open-weights alternatives, which Reflection AI positioned as the rationale for its open-source strategy.
Can either party exit early?
Yes. According to TechCrunch, both SpaceX and Reflection AI have the option to terminate with 90 days' notice after the first three months, allowing either side to exit by October 2026 if needed.