宝玉 (@dotey) · 10/21/2026, 12:12:00

Fin Argues Open Source Pricing Won't Undermine Closed Model Profitability

By 宝玉 (@dotey)Original title: fin 的 AI 半导体终局推演系列都写到 4 了,这次讲的核心问题是:DeepSeek 这类谁都能下载的开源模型越来越便宜,OpenAI、Anthropic 会不会赚不到钱、没钱买 GPU?
68AI Score
Executive Summary

Analyst Fin argues that despite falling prices of open-source models like DeepSeek, hardware iterations and software optimizations keep costs declining in tandem, preserving profit margins. Open source dominates the low-end market, while the flagship tier remains stable under OpenAI and Anthropic. Furthermore, although open source accounts for over 20% of token usage, it represents only 5-10% of actual spend, with overall demand growing as model capabilities improve.

SOURCE COVERAGEFull Original Report

fin 的 AI 半导体终局推演系列都写到 4 了,这次讲的核心问题是:DeepSeek 这类谁都能下载的开源模型越来越便宜,OpenAI、Anthropic 会不会赚不到钱、没钱买 GPU?

fin 认为不会。

第一个理由:价格降,成本降得一样快。

每代新芯片让同样的电多产出约一倍的 token,软件优化每年再砍一半成本,利润率守得住。1GW 数据中心跑旗舰模型一年能卖 800 亿美元,硬件摊销每年只要约 80 亿。

第二个理由:便宜模型和旗舰模型是两个市场。

开源只能在低端打价格战,旗舰层只有 OpenAI 和 Anthropic 在争,价格稳。

第三个理由:开源占 token 用量 20% 以上,按实际花的钱只占 5% 到 10%。

模型越强,能接的长任务越多,需求还会涨。

↳ 引用 @fi56622380 (fin): https://x.com/i/article/2112404770057240576

Fin Argues Open Source Pricing Won't Undermine Closed Model Profitability | Today for AI