SpaceX is converting its Colossus data center into a pay-as-you-go compute platform and has inked a multi-year contract to supply Reflection with Nvidia GB300 GPUs starting July 2026, a commitment that runs through 2029 and carries a maximum value of $6.3 billion at $150 million per month.
SpaceX has signed a $6.3 billion dollar compute deal with Reflection. Reflection will gain immediate access to GB300s to train open source models, and will pay SpaceX $150 million per month beginning July 1, 2026, through 2029, according to materials viewed by CNBC.
Google DeepMind engineer Susan Zhang publicly questioned whether Reflection truly qualifies as open-source, leaving the characterization unverified in current reporting despite the company's stated focus on accessible frontier models.
The Reflection agreement joins existing Colossus deals with Anthropic, Google, and Cursor, showing the data center's shift from internal Starlink and vehicle workloads to external commercial compute sales.
Positive users praise SpaceX's $6.3B compute deal with Reflection as smart diversification and revenue growth while negative users claim it kills Grok and call the open-source claims a pump-and-dump scheme.
No Digg Deeper questions have been answered for this story yet.
#1766
can't even say open-source with a straight face here
(reflection release soon, right?)
SPACEX SIGNS COMPUTE DEAL WITH OPE-SOURCE AI STARTUP REFLECTION - CNBC
https://www.cnbc.com/2026/06/22/spacex-ai-colossus-data-center-reflection.html
SpaceX has signed a $6.3 billion dollar compute deal with Reflection. Reflection will gain immediate access to GB300s to train open source models, and will pay SpaceX $150 million per month beginning July 1, 2026, through 2029, according to materials viewed by CNBC.
SpaceX signs computing power deal with open-source AI startup Reflection worth up to $6.3 billion https://www.cnbc.com/2026/06/22/spacex-ai-colossus-data-center-reflection.html?taid=6a394e64098c630001d9b0ac&utm_campaign=trueanthem&utm_content=main&utm_medium=social&utm_source=twitter
At this point SpaceX just looks like CoreWeave with a bigger budget and a satellite company thrown in.
$SPCX signs $6.3B compute deal with AI startup Reflection; $150M/month through 2029 for capacity at Colossus 2 in Memphis.
Reflection is America's open-source answer to Deepseek (model weights are open for researchers and developers, but training data and the full training process remain proprietary), founded in 2024 by two ex-Google DeepMind researchers and funded by NVDA among others. Recently shopping a $2.5B round at $25B valuation, per WSJ.
Reflection's primary product is Asimov, a code-research agent designed to help engineering teams understand large, complex codebases rather than generate new code, based on the premise that ~70% of engineering time is spent reading existing systems, not writing new ones.
@AndrewCurran_ spaceX is gonna run the world huh
@EliasofIX Reflection holds Pentagon classified-network access (Anthropic excluded) and has the “foundational intelligence layer” role in DOE Genesis Mission
@AndrewCurran_ @aaronburnett @VladSaigau
@rand_longevity Based on current projections it will make Elon a quadrillionaire by around 2037.
@ns123abc Never heard of that company before, interesting. Their website is very vague, and apparently they have no released models? Might be a sleeper kind of company, could be very cool.
@CNBC SpaceX selling compute capacity is the detail nobody is talking about here, that's basically a new revenue line outside launches and Starlink
@AndrewCurran_ I assume others also want to look them up: @reflection_ai
@AndrewCurran_ damn...3rd major deal in a row
@AndrewCurran_ SpaceX and Xlab will lead the AI race by 2027.
@AndrewCurran_ I look forward to when starship containers filled with GPUs start landing on the moon. Long running compute jobs could run safely on the moon, unimpeded for very long periods of time
@ns123abc Elon, please don't forget about Grok Imagine bros. Our limits are still severely restricted.
@AndrewCurran_ Cursor's not going to have enough compute at this rate.
@ns123abc I want midjourney to get included here
@AndrewCurran_
@AndrewCurran_ is spacex basically giving up on training models or what? infra as a service should also do great.
@ns123abc let the cash flow
