SpaceX signs a computing deal worth up to $6.3B with Reflection AI for access to Nvidia GB300s at Colossus 2; Reflection will pay $150M per month through 2029
SpaceX has signed a major computing power agreement with Reflection AI, making the open-source artificial intelligence startup …
Context & Ripple Effects
This agreement extends a visible pattern in the related coverage: after SpaceX acquired xAI, its Colossus infrastructure was already being contracted to Anthropic, including an expansion toward Colossus 2 capacity.
Reflection is also seeking capacity from Nebius, indicating that the startup is assembling compute supply across more than one provider rather than relying solely on its SpaceX arrangement. The deal therefore matters as another large, multi-year commitment for Colossus 2 and Nvidia-based AI capacity.
First-order effects
- Reflection secures access to Nvidia GB300 systems at Colossus 2 under a payment commitment of $150 million per month through 2029, giving it a defined long-term source of training and inference capacity.
- SpaceX gains a contracted revenue stream worth up to $6.3 billion and another major tenant for Colossus 2, alongside its reported Anthropic capacity commitments.
Second-order effects
- Reflection’s subsequent Nebius agreement suggests that securing sufficient AI compute may require startups to diversify providers; this can reduce any one infrastructure operator’s leverage while increasing demand for available Nvidia-equipped capacity overall.
- Large, long-duration customer contracts make capacity planning more consequential for SpaceX: allocating systems to external customers competes with flexibility for xAI-related workloads and future tenants.
Third-order effects
- If similar contracts continue, frontier-model developers may increasingly treat compute procurement as a portfolio problem—locking in multi-year capacity across specialized operators instead of depending on a single cloud provider.
- The pattern would strengthen the role of vertically integrated AI infrastructure operators, but legal challenges cited around Colossus data centers show that buildout and siting constraints could become as important as chip access in determining usable capacity.
The trend: AI companies are moving toward multi-year, multi-provider reservations of high-end GPU capacity as compute becomes a strategic input rather than an on-demand utility.