Decagon CEO Jesse Zhang dropped a spicy new theory: open source AI isn’t actually eating frontier models’ lunch. Instead, they’re two sides of the same coin. As enterprises find success with pricey frontier models, they often switch to cheaper open source versions for production, but the overall spending on top-tier models stays steady because new use cases keep popping up. Hard data backs this up: on Vercel, DeepSeek now handles over a third of tokens, but Anthropic still claims more than half the spend. Same story on OpenRouter, where DeepSeek V4 Flash processes 5.3 trillion tokens weekly versus Opus 4.8’s 2 trillion, but Opus costs 23x more per token. Frontier labs like Anthropic aren’t hurting because they own the discovery phase, while open source scales production. Think of it like a conveyor belt: frontier models prove what’s possible, then hand off to open source for the heavy lifting. Why it matters: This flips the narrative from a zero-sum battle to a symbiotic cycle, explaining why frontier AI spending won’t collapse even as open source adoption explodes.