The article argues that frontier AI companies function primarily as energy infrastructure businesses rather than pure software entities, creating a structural vulnerability for those lacking ownership of their compute resources. While Anthropic currently dominates the business AI coding market, its reliance on purchasing capacity from providers like AWS and SpaceX means it operates as a customer dependent on external agreements that may shift based on the provider’s strategic interests. Unlike Google or Meta, which control their own data centers and silicon, Anthropic does not own the wells, leaving it exposed when infrastructure priorities change.
This risk is concretized by SpaceX’s dual role as both a compute supplier and a direct competitor through its acquisition of Cursor. By funding SpaceX’s Colossus cluster, Anthropic inadvertently finances the expansion of a rival’s industrial base, while the Cursor merger provides a distribution channel capable of routing traffic away from Claude entirely. Furthermore, SpaceX’s investment in custom C-based training stacks promises to lower production costs significantly, allowing them to compete on price parity rather than requiring absolute model superiority to capture market share.
Ultimately, the piece suggests that model quality alone is insufficient without control over the underlying supply chain, predicting that Anthropic will face measurable decline by mid-2028 if it fails to secure deeper operational control. Enterprise customers exhibit weak switching moats because their code and workflows remain independent of the model itself, meaning vendor dependency fails when access becomes unreliable or expensive. Consequently, the competitive bar is set at acceptable parity at a better price, placing Anthropic at higher risk of becoming a niche product compared to vertically integrated providers.