投资者正向人工智能经济的一个热门新领域注入资金：模型路由——一种让开发者和公司能够轻松在不同人工智能模型之间切换的做法。
Investors are pouring cash into a hot new sector of the AI economy: model routing — a practice allowing developers and companies to easily switch between different AI models.
成立六周的初创公司TrustedRouter正处于一批旨在降低人工智能成本、减少企业对单一模型制造商依赖的模型路由公司之中。该公司本周完成了由Sierra Ventures领投的600万美元种子轮融资，它向Semafor透露了这一消息。
Startup TrustedRouter, which launched publicly six weeks ago, is among a growing crop of model-routing firms looking to cut AI costs and reduce businesses’ dependence on any single model maker. The company this week closed a $6 million seed round, led by Sierra Ventures, it told Semafor.
路由领域的最大玩家是OpenRouter，它在2025年2月完成了1250万美元的种子轮融资，此后同意以75亿美元的价格出售给支付服务提供商Stripe。
The biggest player in the routing space is OpenRouter, which started with a $12.5 million seed round in February 2025 and has since agreed to sell to payments provider Stripe in a $7.5 billion deal.
TrustedRouter的首席执行官约瑟夫·佩拉表示，该公司旨在通过为某些开放权重模型提供端到端加密和可验证的机密计算来实现差异化，这意味着既不是TrustedRouter也不是运行模型的数据中心能够读取客户的提示或输出，而来自OpenAI和Anthropic的封闭模型目前并不提供同样的保障。
TrustedRouter, CEO Joseph Perla said, aims to differentiate itself by offering end-to-end encryption and verifiable confidential computing for some open-weight models, meaning neither TrustedRouter nor the data center running the model can read a customer’s prompts or outputs, unlike with closed models from OpenAI and Anthropic, which don’t currently offer the same guarantees.
佩拉还表示，OpenRouter也不提供这些保护：“安全只是一个基本前提。它只需要是安全的。”佩拉曾是Facebook工程师，也是Lyft的早期员工，他创立并出售过多家初创公司，希望能够赢得那些人工智能支出激增但同时也高度重视数据隐私的企业。
OpenRouter also doesn’t offer those protections, according to Perla: “Security is just a pure baseline. It just has to be secure.”Perla, a former Facebook engineer and early Lyft hire who has founded and sold several startups, hopes to win over businesses that are seeing their AI spending balloon, but also putting a premium on keeping their data private.
“如果我们都对人工智能聊天进行加密，”佩拉说，“我们就能减缓”前沿实验室吸收用户数据以训练未来模型的循环。
“If we all encrypt our AI chats,” Perla said, “we can slow down” the cycle in which frontier labs absorb user data to train future models.
他说，包括DeepSeek、智谱和Moonshot在内的中国AI模型在他的平台上非常受欢迎，但他也希望看到一个更具活力的美国开源生态系统。最近几周，西方开放权重模型的发布呈现出一股热潮，包括来自Reflection AI和Mistral的模型；周四，Meta发布了它所谓的“迄今为止最强大的美国开放权重模型”。最终，Perla表示，他希望推出“纽约代币交易所”（New York Token Exchange），这是一个AI模型的市场，届时将有来自数千家托管提供商的数千种模型参与竞争——任何愿意运行开放模型的数据中心都可以加入——以此来处理每个请求，将AI代币转化为一种可通过期货、对冲和价格指数进行交易的商品，就像石油或电力一样。他希望为新发布的开放权重模型举办“模型IPO”。
He said Chinese AI models, including those from DeepSeek, Zhipu, and Moonshot, are among the most popular on his platform, but he wants to also see a more vibrant American open-source ecosystem. There’s been a flurry of Western open-weight releases in recent weeks, including from Reflection AI and Mistral; on Thursday, Meta released what it called “the most capable US open-weight model to date.”Eventually, Perla said he wants to launch the “New York Token Exchange,” a marketplace for AI models in which thousands of models from thousands of hosting providers — any data center willing to run an open model — would compete to serve each request, turning AI tokens into a commodity that could be traded with futures, hedges, and price indices, like oil or electricity. He hopes to stage “model IPOs” for new open-weight releases.
开发者对路由服务有明确需求，而TrustedRouter有望填补他们中最注重隐私群体的空白。它是否能从小众走向大众市场，可能取决于开放权重模型的竞争力。如果它们继续落后于OpenAI和Anthropic的闭源竞争对手，一些企业和开发者可能会接受更高的成本和缺乏加密的情况，以便为其最先进的任务获得更好的输出。
There’s clear demand from developers for routing services, and TrustedRouter could fill a void for the most privacy-obsessed among them. Whether it goes from niche to mass market could depend on how competitive open-weight models get. If they keep lagging behind their closed competitors at OpenAI and Anthropic, some enterprises and developers will put up with the higher cost and lack of encryption to get a better output for their most advanced tasks.
我上周监督的一个项目——我们利用AI分析了数千段关于数据中心的有线新闻节目——成本约为140美元，因为我们使用了先进的闭源模型。如果在加密、保密的基础设施上使用TrustedRouter并搭配性能较弱的中国模型运行，成本可能会低于20美元，尽管输出质量可能不会那么好。只要这种差距持续存在，这类权衡将一直困扰着科技领袖。
A project I oversaw last week, in which we analyzed thousands of cable news segments about data centers using AI, cost about $140, because we used advanced closed models. Running that through TrustedRouter on encrypted, confidential infrastructure with less advanced Chinese models would have cost under $20, though the output likely wouldn’t have been as good. These kinds of tradeoffs will keep confronting tech leaders as long as that disparity persists.