Your railroad analogy is exactly right. But I'd push it one layer deeper. Between the model builders going broke and the application builders getting rich, there's a middle layer quietly making fortunes: the inference providers. Together AI at $3.5 billion, Fireworks at $3.4 billion, OpenRouter processing 30 trillion tokens a month. The pickaxe sellers in this gold rush are the ones hosting commodity models, not building them. I traced the full economics of who profits when models are free: https://medium.datadriveninvestor.com/who-profits-when-ai-models-are-free-b71ae03f4167
Your railroad analogy is exactly right. But I'd push it one layer deeper. Between the model builders going broke and the application builders getting rich, there's a middle layer quietly making fortunes: the inference providers. Together AI at $3.5 billion, Fireworks at $3.4 billion, OpenRouter processing 30 trillion tokens a month. The pickaxe sellers in this gold rush are the ones hosting commodity models, not building them. I traced the full economics of who profits when models are free: https://medium.datadriveninvestor.com/who-profits-when-ai-models-are-free-b71ae03f4167
Ooh, I’ll have a look. Thanks!