Third-party cyber evaluations involving OpenAI models
openai.com·14h ago
Data Science
Simon Willison's Blog·May 20·3 min read
Quoting SpaceX S-1
TL;DR
SpaceX operates large-scale compute infrastructure (COLOSSUS II) that supports both internal AI model training and external cloud services. The company built a dual-use compute strategy: running proprietary models like Grok 5 internally while selling excess capacity to third-party AI research firms.
✦ Why It Matters
Engineers can learn how to monetize excess compute capacity and structure partnerships with AI research organizations.
Key Takeaways
- Evaluate your compute resource needs for AI applications.
- Consider entering cloud service agreements for additional revenue.
- Monitor the performance of COLOSSUS and COLOSSUS II for insights.
- Stay informed about AI partnerships in the industry.
Full Summary
SpaceX is leveraging its compute resources to enhance its proprietary AI applications, including Grok 5, which is currently being trained on its advanced systems COLOSSUS and COLOSSUS II. In May 2026, SpaceX entered into a significant Cloud Services Agreement with Anthropic PBC, a public benefit corporation focused on AI research and development.
Under this agreement, Anthropic will pay SpaceX $1.25 billion per month for access to these compute resources, with the capacity increasing gradually over the next few months at a reduced fee. The contract is set to last until May 2029, but either party can terminate it with 90 days' notice.
This partnership highlights the growing demand for high-performance computing in AI development and the potential for substantial revenue generation in the cloud services market.