Reflection AI, a startup founded by former Google DeepMind researchers with a mission to build open-weight frontier models, has signed a $6.3 billion computing contract with SpaceX, according to a report by The Next Web published June 23, 2026.
Under the agreement, Reflection will pay SpaceX $150 million per month starting July 1, 2026, for access to Nvidia GB300 chips at the Colossus data center complex in Memphis, Tennessee. The contract runs through 2029, with either party able to exit after the initial three months on 90 days notice.
Who is Reflection AI
Founded in 2024 by Misha Laskin and Ioannis Antonoglou – both veterans of Google DeepMind, where Antonoglou contributed to the AlphaGo project – Reflection AI is positioning itself as an alternative to the closed-model approach taken by OpenAI, Anthropic, and Google. The company has pledged to release its models as open weights, meaning developers and researchers can download and run them without a commercial API.
The startup has attracted heavyweight backers: Nvidia has invested $800 million, and Sequoia and Lightspeed are also on the cap table. A June 2026 fundraising round valued the company at $25 billion, up from $8 billion in October 2024. Reflection has not yet released a public frontier model, but has been serving government and national security customers.
The Colossus Facility
Colossus was originally built in Memphis for xAI, Elon Musk’s AI company. The facility now houses over 220,000 Nvidia GPUs across its first phase and is one of the most densely concentrated AI compute sites in the world. SpaceX, which operates the data center, has secured more than $80 billion in compute revenue through 2029 across external deals.
Reflection’s statement on the deal was direct: “More compute means more runway to build the world’s best open models at scale.”
Why it matters
Frontier AI capability has until now been locked behind proprietary APIs controlled by a handful of American companies. The prospect of a comparably powerful open-weight model – one that anyone can download and deploy – would dramatically shift the balance between access and restriction. Whether Reflection can close the benchmark gap against Fable 5 and GPT-5.6 remains to be seen, but the compute foundation to try is now in place.
Why it matters for Georgia
Georgian developers and researchers have faced growing friction in accessing cutting-edge AI. When the U.S. government this month ordered Anthropic to suspend access to its most powerful models for all foreign nationals – citing cybersecurity concerns – Georgian users lost access overnight. Open-weight models remove that single point of failure entirely: they run locally, without an API gatekeeper, and are not subject to export control enforcement in the same way.
If Reflection delivers a competitive open-weight frontier model, Georgian universities, startups, and public institutions could deploy it directly – without depending on access decisions made in Washington or San Francisco.