新京报贝壳财经讯（记者罗亦丹）9月13日，智谱发布公告称，完成约50亿美元融资，包括约20亿美元股份配售及约30亿美元可转债发行。根据公告，智谱此次所得款项净额主要将用于下一代GLM、完全自训练体系及相关算力基础设施。
在公告中，智谱将完全自训练描述为下一代GLM在上一代GLM构建的环境中训练，形成递归式自我改进循环。具体投入包括自动化生成和筛选训练数据、构建任务环境、提升长程推理能力，以及国产芯片适配、算子开发和推理优化。
持续推出前沿模型，需要研发能力与长期投入相互支撑。智谱已通过连续升级展示模型迭代能力，约50亿美元资金的补充，则扩大了其跨多个研发周期安排算力、实验与人才投入的空间。
今年以来，GLM系列保持连续迭代，2月至8月，智谱先后推出GLM-5、GLM-5.1、GLM-5.2和GLM-5.3，保持约每两个月一次的重要升级。MaaS月度收入年化计算的ARR于8月底达到16亿美元。
本轮融资将进一步支持模型训练、生产推理和底层工程建设，为后续技术迭代配置资源。随着算力扩容获得更充足的资金支持，智谱将有更大空间同步推进训练实验与生产推理部署，加快将模型能力转化为规模化服务，为下一阶段技术突破和收入增长创造条件。
本文来源：https://www.sohu.com/a/1075682991_114988
若有版权问题，请来信告知：13951431913#139.com 或在右方提交评论。
Xinhua Beijing News Shell Finance reported on September 13th that Zhipu announced the completion of a $5.0 billion financing, including a $2.0 billion share placement and a $3.0 billion convertible bond issuance. According to the announcement, the proceeds from this round of financing will mainly be used for the development of the next-generation GLM model, a fully self-training system, and relevant computational infrastructure.
In the announcement, Zhipu described full self-training as the next-generation GLM model being trained in the environment built by the previous-generation GLM, forming a recursive self-improvement loop. Specific investments include automated generation and screening of training data, construction of task environments, enhancement of long-range reasoning capabilities, adaptation of domestic chips, operator development, and inference optimization.
Sustained development of cutting-edge models requires a synergy of research capabilities and long-term investment. Zhipu has demonstrated its model iteration capability through continuous upgrades, and the infusion of approximately $5.0 billion expands its capacity to allocate computing power, conduct experiments, and invest in talent across multiple development cycles.
Since the beginning of the year, the GLM series has maintained a rapid pace of iteration, with Zhipu launching GLM-5, GLM-5.1, GLM-5.2, and GLM-5.3 from February to August, representing significant upgrades roughly every two months. The Annual Recurring Revenue (ARR) from the Model-as-a-Service (MaaS) business has reached $1.6 billion by the end of August.
This round of financing will further support model training, production inference, and underlying engineering construction, providing the necessary resources for subsequent technological iterations. With stronger financial backing for computational power expansion, Zhipu will have greater flexibility to simultaneously advance training experiments and production deployments, accelerating the transformation of model capabilities into large-scale services and laying the groundwork for future technological breakthroughs and revenue growth.