Coding Platforms Threaten Anthropic

The real threat is that coding platforms can now turn frontier models into a hidden ingredient instead of a visible dependency. If Cursor can build Composer on Moonshot models and buy inference from providers like Fireworks, it can tune quality, latency, privacy, and price inside its own product. That weakens Anthropic twice, because Claude loses direct API share and Claude Code loses the chance to become the developer’s default workspace.1 2 3 4 5

• Open weight and partner served models change who owns the customer. Moonshot distributes Kimi through revenue sharing with inference providers, which lets app companies keep billing, interface, and workflow control while the model lab sits behind the scenes.3 5
• Anthropic had benefited from the opposite setup. Claude became core infrastructure for the vibe coding wave, powering tools like Cursor and helping drive Anthropic from $1.0B annualized revenue at the end of 2024 to $65B by July 2026, so substitution at the app layer now hits a very large revenue base.2
• This is the same squeeze hitting model labs more broadly. OpenRouter grew by routing developers across dozens of models with failover and analytics, while coding agents like Claude Code, Codex, and Cursor all moved up the stack into the end product, making raw model access less differentiated on its own.7 8 9

The coding market is heading toward a split where a few labs supply flagship models, but the durable margins sit with products that control workflow, routing, and user data. For Anthropic, that means the next phase is less about having the best model in isolation and more about making Claude hard to swap out inside the tools where developers already work.1 2 7 9