Prime Intellect 完成 1.3 亿美元 A 轮融资，助企业自建 AI Agent

Prime Intellect，一家提供算力和专用软件工具以帮助企业构建 AI Agent 的初创公司，已获得 1.3 亿美元 A 轮融资，估值达到 10 亿美元。

本轮巨额融资由 Radical Ventures 领投，Nvidia Ventures、Intel Capital、Dell Technologies Capital、Iconiq 参投，并有一长串知名企业创始人作为天使投资人加入，包括 Perplexity 的 Aravind Srinivas、Box 的 Aaron Levie、Harvey 的 Winston Weinberg、Cognition 的 Jeff Wang 以及 Mercor 的 Brendan Foody。

Prime Intellect 成立于 2024 年，其目标是让各类组织能够自行训练智能体系统，而无需依赖前沿 AI 实验室。虽然这一使命在几年前会难以实现，但随着强化学习技术的兴起——该方法通过不断奖励任务成功和惩罚错误来迭代——企业可以通过针对特定业务任务微调模型，成为“自己的 AI 实验室”。

尽管如今绕开闭源 AI 实验室已成为可能，但底层基础设施依然极为复杂，大多数公司缺乏将这些组件整合为生产就绪系统所需的专业知识。

这正是 Prime Intellect 的用武之地。

这家初创公司开发了所谓的 AI Agent 开发“全栈”，包括算力访问、强化学习框架和评估工具。

Prime Intellect 的平台如同一个市场，提供模块化访问，客户可以按需挑选所需工具，不必被锁定在全有或全无的系统中。

Radical Ventures 合伙人 David Katz 表示：“他们把这一切整合起来，打造的方式既前沿又经济实惠。” 他补充说，尽管其他公司仅提供零碎功能，Prime Intellect 的独特之处在于能够将一流 AI 实验室的能力作为“一站式服务”提供给开发者。

该初创公司的方法已吸引了 Ramp、Zapier 和 Flapping Airplanes 等客户，他们付费使用其托管工具版本。快速的采用已使公司年化经常性收入达到 1 亿美元。

这一增长得益于切实可见的成果。例如，Ramp 借助 Prime Intellect 构建了一个智能体，帮助这家金融科技公司在电子表格中查找答案。Ramp 联合创始人兼联席 CEO Karim Atiyeh 在一份声明中表示：“结果在准确性上击败了前沿模型，同时运行速度更快，成本仅为前沿模型的一小部分。”

推动 Prime Intellect 增长的另一个关键因素是，企业近来意识到，在前沿 AI 实验室之上构建系统存在诸多风险。

越来越多的公司不愿向 OpenAI 和 Anthropic 提供专有信息，因为存在失去数据控制的风险。它们也担心依赖那些可能会突然被关停的模型，就像上个月 Anthropic 的 Fable 那样。

Katz 说道：“我怎么知道我不会与一家试图取代我并泛化我所作所为的公司合作。所有这些因素促使人们思考：‘我如何才能拥有自己的企业智能，并避免这些风险’。”

Prime Intellect 联合创始人兼 CEO Vincent Weisser 在接受 TechCrunch 采访时表示：“不应该只有旧金山玻璃塔里的少数书呆子才具备训练 AI 模型的能力。而应该是每一家企业，每一个国家。”

Prime Intellect, a startup that provides computing power and specialized software tools that help companies build AI agents, has raised a $130 million Series A at a $1 billion valuation.

The massive round was led by Radical Ventures, with participation from Nvidia Ventures, Intel Capital, Dell Technologies Capital, Iconiq, and a long list of angel investors who are founders of notable companies, including Aravind Srinivas (Perplexity), Aaron Levie (Box), Winston Weinberg (Harvey), Jeff Wang (Cognition), and Brendan Foody (Mercor).

Founded in 2024, Prime Intellect’s goal is to give organizations capabilities to train their own agentic systems without relying on frontier AI labs. While this mission would have been hard to achieve just a few years ago, the rise of reinforcement learning techniques, which iteratively reward successful task completion and penalizes errors, can allow companies to become their “own AI lab” by refining models for specific business tasks.

Although it is now possible to bypass closed AI labs, the underlying infrastructure remains so complex that most companies lack the expertise to assemble these pieces into a production-ready system.

That’s where Prime Intellect comes in.

The startup has developed what it calls a “full stack” for AI agent development, which includes compute access, a reinforcement learning framework, and evaluation tools.

Prime Intellect’s platform functions like a marketplace, providing modular access so customers can pick and choose the specific tools they need without being locked into an all-or-nothing system.

“They’ve stitched this together and built it in such a way that they’re operating at the frontier in a way that’s affordable,” said David Katz, a partner at Radical Ventures. He added that while others offer bits and pieces, Prime Intellect is unique in providing the capabilities of a top-tier AI lab as a “one-stop shop” for development.

The startup’s approach has attracted customers like Ramp, Zapier, and Flapping Airplanes, who pay the startup for a hosted version of its tools. This rapid adoption has propelled the company to an annualized revenue run rate of $100 million.

This growth is driven by the tangible results. For example, Ramp used Prime Intellect to build an agent that helped the fintech find answers inside spreadsheets. “The result beat the frontier models on accuracy while running at faster speeds and a fraction of the cost,” Ramp’s co-founder and co-CEO Karim Atiyeh said in a statement.

Another key factor driving Prime Intellect’s growth is the recent realization by companies that building on top of frontier labs carries a number of risks.

Companies increasingly don’t want to provide their proprietary information to OpenAI and Anthropic due to the risk of losing control over their data. They are also wary of depending on models that can be suddenly turned off, as happened with Anthropic’s Fable last month.

“How do I know that I’m not working with a company that is going to try to replace me and generalize to what I’m doing,” Katz said. “All of these things are causing people to think, ‘How do I own my own enterprise intelligence and not have these risks’.”

Prime Intellect co-founder and CEO Vincent Weisser believes enterprises are looking to move away from closed source frontier models, and his company provides the infrastructure to make that transition possible.

“It shouldn’t just be a few nerds in a glass tower in San Francisco that have the capability to train AI models,” he told TechCrunch. “It should be every enterprise, every nation state.”