Anthropic IPO 估值缩水：发行区间对应约 1 万亿美元，低于此前 1.5–2 万亿预期
这篇分析把 Anthropic IPO 降估值的账算得很清楚：OpenRouter 上开源 token 一年从 2 万亿涨到 146 万亿，九成跑中国模型，连 H100 租价都被推高了 25%。做模型选型或关注中美开源差距的人都该看看这组数据。
Anthropic 的 IPO 定价区间为 70–78 美元，对应市值约 1.0 万亿美元，完全稀释口径约 1.3–1.4 万亿美元，明显低于几个月前市场谈论的 1.5–2 万亿美元。
作者认为 Anthropic 急于上市的核心原因是开源模型带来的双重挤压：一方面 OpenRouter 上的开源模型 token 用量从 1 月的每周约 2 万亿涨到 11 月的约 146 万亿，其中 90% 以上跑在中国模型上，客户借此规避闭源前沿模型的溢价并控制数据留存；另一方面自托管开源模型的需求推高了 GPU 租赁价格，H100 一年上涨 25%，B200 自 5 月以来上涨 50%，数据中心不再只能把算力卖给 Anthropic 这类大买家。
叠加 OpenAI 和 Gemini 的竞争，Anthropic 需要在 ARR 仍为正增长时，通过上市把持续增长的算力账单转化为永久资本。这篇文章对理解闭源厂商的资本市场逻辑和中国开源模型的真实市场渗透率都有参考价值。
Anthropic’s valuation came down a lot before the IPO. The $70–78 range works out to about $1.0T, or $1.3–1.4T fully diluted, against the $1.5–2T people were talking about a few months ago. A big haircut on the expectation.
Why do they have to go IPO sooner or later?
Apart from the political uncertainty, open source models are giving Anthropic pressure from two fronts.
- On one side, OSS models are ready to be used, and the usage is growing exponentially: open-model tokens on OpenRouter went from ~2T a week in January to ~146T in November, 90%+ of it on Chinese models, as clients look for ways to avoid the frontier model tax and turn to usable open models with controlled data retention.
- On the other side, the growth of open source models has moved the market price of GPUs. The I-can-earn-money-by-moving-up-the-stack thinking plus huge demand to run self-hosted open models pushed rental prices up: H100 +25% in a year, B200 +50% since May. Previously datacenters had no choice but to sell to big players including Anthropic. Now they can host models themselves, or rent to the people who do.
This, alongside the competition from @OpenAI and @GeminiApp, is the pressure on Anthropic, and it explains why they need to list soon: turn a recurring growing compute bill into permanent capital before the next tranche comes due, while the ARR growth is still positive.
The OpenRouter China vs US chart was inspired by @natolambert's recent post.
This report is categorized under AITOP's "Industry" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include Anthropic, OpenRouter, 开源模型, IPO.
Source attribution: curated from Xianbao Qian (@Xianbao_QIAN). Cross-source corroboration: 11 independent sources have covered this development (including The Diplomat, Decoder, rohanpaul_ai), confirming industry-wide signal strength.
