智东西（公众号：zhidxcom）
编译 | 孙小雯
编辑 | 程茜
智东西10月30日消息，昨晚，加拿大大模型独角兽Cohere宣布完成E轮融资首次交割，本轮目标融资最高70亿加元（约合人民币334亿元）。Cohere在官方X账号上称，若达到募资目标，本次融资将创下加拿大初创企业史上最大单轮私募融资纪录。
据Cohere透露，其计划在一年内推进两项AI公司并购，推出面向企业网络安全的产品Cohere Cyber，并训练新一代前沿大模型。
本轮融资由德国零售巨头施瓦茨集团领投，英伟达、沙特阿美旗下Prosperity7 Ventures等老股东继续参投，加拿大政府等首次参投。去年9月，该公司估值约为98亿加元（约合人民币466亿元）。若其达到募资目标，Cohere估值将达357亿加元（约合人民币1702亿元），一年内增长至原来的3.6倍。
Cohere成立于2019年，主要为企业和政府提供可私有化部署的大模型、AI智能体及搜索工具。
其三位联合创始人均为多伦多大学校友。其中，联合创始人兼首席执行官Aidan Gomez是Transformer奠基论文Attention Is All You Need的八位作者之一。另一位联合创始人Nick Frosst则曾师从“深度学习教父”、2024年诺贝尔物理学奖得主Geoffrey Hinton。
▲Cohere三位联合创始人，从左至右依次为Ivan Zhang、Aidan Gomez、Nick Frosst（图源：Index Ventures）
此前，Cohere已完成多轮大额融资：2023年完成约3.8亿加元C轮融资，2024年完成约7亿加元D轮融资，2025年再融资约8.4亿加元，英伟达、甲骨文、Salesforce Ventures等曾参投。
据Cohere披露，截至今年第三季度末，其ARR同比增长165%。全球客户数量在过去一年增长至原来的20倍，达到300家，包括加拿大皇家银行、富士通、普华永道等企业。
一、自研模型接入企业客服，问题解决时间缩至2—5天
Cohere主要面向企业文档分析、推理及AI智能体任务开发模型。今年5月，该公司开源了Command A+大语言模型，其量化版本仅需两张英伟达H100 GPU即可运行，支持企业自行部署。
为了让AI参与企业日常工作，Cohere还推出了智能体平台North。该平台可以连接企业内部知识库和办公软件，让员工通过自然语言创建AI智能体，完成资料查询、客户服务等任务。今年10月5日，Cohere推出North 2，新增跨会话记忆、共享知识库等功能，让AI能够记住此前的工作信息，并连续执行多个任务。
据Cohere今年2月公布的客户案例，美国AI云厂商CoreWeave仅用90天便将此前版本的North接入客服系统，用AI收集客户信息、查找技术文档，并协助将问题转交给工程师。结合内部流程优化，CoreWeave的平均问题解决时间从4—8天缩短至2—5天。
▲Cohere North企业智能体平台演示界面，可调用Google Drive、Outlook等企业数据进行业务分析（图源：Cohere）
针对企业使用AI时的数据安全需求，Cohere提供私有化部署方案，并与英伟达、微软合作推出机密计算服务。今年10月，该公司推出基于英伟达H100 GPU和微软Azure的机密计算服务公测版，通过硬件加密技术保护AI模型在云端处理的敏感数据。
二、拟并购两家AI公司，加拿大政府最高投76亿元
Cohere计划在未来一年推进两项AI公司并购。10月15日，该公司与美国AI网络安全创企AISLE签署收购意向书，计划利用其技术推出Cohere Cyber，帮助企业自动发现、验证和修复软件漏洞。
据AISLE官方披露，过去一年，该公司已发现500多个获得CVE编号的软件漏洞，其中包括开源网络工具curl近期修复的24个漏洞中的18个。Cohere计划将AISLE的技术与自身大模型结合，让企业在本地服务器或隔离网络中完成漏洞排查，避免敏感代码外传。
▲AISLE AI网络安全平台工作流程，覆盖软件漏洞分析、风险筛选、修复及验证等环节（图源：AISLE）
另一项交易则面向欧洲市场。今年9月，Cohere与德国AI公司Aleph Alpha签署合并协议，计划整合双方的模型研发能力及欧洲政府、企业客户资源。本轮融资领投方施瓦茨集团也计划在旗下STACKIT云平台部署Cohere模型，为欧洲客户提供AI服务。目前，两项交易均尚未完成。
除并购外，Cohere还将获得加拿大政府的资金支持。根据加拿大总理办公室10月29日公布的协议，政府拟分阶段出资最多16亿加元（约合人民币76亿元）入股Cohere。作为投资条件，Cohere承诺在2030年前新增至少1000个加拿大技术岗位，并投资至少17亿加元（约合人民币81亿元）建设当地算力设施，同时保留加拿大总部及注册地。
结语：从模型研发延伸至企业部署，AI创企开始争抢企业软件订单
此次融资将支持Cohere继续开发大模型，并通过并购拓展网络安全、欧洲企业市场。加上本土算力建设和云平台合作，Cohere正在从模型研发进一步延伸至企业AI产品的部署和交付。
企业AI市场的竞争也已不局限于模型本身。本月，另一家获得英伟达投资的AI办公公司Pigment完成2.5亿美元融资，并披露要争夺微软、甲骨文、IBM等厂商的企业客户。
虽然两家公司的产品路线不同，但都在尝试让AI进入企业已有的业务系统，承担实际工作。对Cohere而言，接下来要争夺的不只是模型调用量，还有企业在办公、运维和网络安全等领域的软件采购预算。
来源：路透社、Cohere
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Intellectual (WeChat Official Account:zhidxcom)
Compiled by Sun Xiaoxuan
Editor, Concubine.
On October 30, according to the source, last night Cohere, the Canadian LLM unicorn, announced the completion of the first delivery of the E round of financing, which would be up to $7 billion (approximately RMB 33.4 billion). Cohere stated in the official X account that if the fund-raising target is met, the financing will result in the largest single-cycle private fund-raising record in the history of start-ups in Canada.
According to Cohere, it is planned to promote two AI mergers and acquisitions within one year, to launch the product Cohere Cyber for enterprise network security and to train a new generation of frontier LLM.
This round of financing was led by a German retail giant, the Schwartz Group, with the participation of older shareholders, such as British Weida and Saudi Arabia-United States Prosperity 7 Ventures, and the Government of Canada for the first time. Last September, the company was valued at approximately $9.8 billion (approximately 46.6 billion yuan). If the fund-raising target is met, the Cohere valuation will reach $35.7 billion (approximately $17.2 billion RMB), increasing to 3.6 times the original level in one year.
Cohere was established in 2019 to provide enterprises and Governments with privatized LLM, AIAI agent and search tools.
Three of its co-founders are students from Toronto. Among them, Aidan Gomez, co-founder and chief executive officer, was one of eight authors of the Transformer foundation paper Attention Is All You Need. Another co-founder, Nick Frost, was a former teacher from the “Father of Deep Learning” Nobel Prize winner of physics, Geoffrey Hinton, 2024.
Three co-founders of Cohere, Ivan Zhang, Aidan Gomez, Nick Frost (source: Index Ventures)
Before that, Cohere had completed multiple rounds of substantial financing: approximately $380 million in C-round finance in 2023, about $700 million in D-round finance in 2024, about $840 million in additional financing in 2025, with the participation of In Weida, Oracle, Salesforce Ventures and others.
Cohere disclosed that by the end of the third quarter of this year, its ARR had increased by 165 per cent over the same period. The number of global clients has increased 20 times over the past year to 300, including the Royal Bank of Canada, Fujito and PricewaterhouseCoopers.
Self-study model access to corporate guest clothes, with problem resolution time reduced to 2-5 days
Cohere is mainly oriented towards business document analysis, reasoning and AZAI agent task development model. In May this year, the company launched the Comand A+ Large Language Model, whose quantization version will only require two British Weibo H100 GPUs to be operational to support the enterprise ' s self-deployment.
In order to involve AI in the day-to-day business, Cohere also launched AI agent Platform North. The platform connects the enterprise ' s in-house knowledge base and office software, allows staff to create AZAI agent in their natural languages, perform information queries, customer services, etc. On 5 October this year, Cohere launched North 2, adding cross-conference memory, sharing knowledge base, etc., to enable AI to remember previous work messages and to perform multiple tasks on a continuous basis.
According to the client case released by Cohere in February this year, the American AI Cloudmaker CoreWeave used only 90 days to access the previous North version of the passenger service system to collect customer information, locate technical files and assist in the referral of problems to engineers. In conjunction with internal process optimization, the average problem resolution time for CoreWeave was reduced from 4 to 8 days to 2 to 5 days.
▲Cohere North Enterprise AI agent platform presentation interface, which allows business analysis using business data such as Google Drive and Outlook (source:Cohere)
In response to the data security needs of enterprises when using AI, Cohere provides a privatization deployment programme and works with Ingweida and Microsoft to introduce confidential computing services. In October this year, the company launched a public version of the confidential computing service based on the H100 GPU and Microsoft Azure to protect sensitive data processed by the AI model in the cloud through hardware encryption.
II. Two AI companies to be acquired and the Government of Canada to invest up to $7.6 billion
Cohere plans to move forward with two AI mergers and acquisitions in the coming year. On 15 October, the company signed a letter of intent for the acquisition of the United States of America ' s AI Network Security Enterprise AISLE, which plans to use its technology to launch Cohere Cyber to help businesses automatically detect, validate and repair software loopholes.
According to official AISLE, in the past year the company has discovered more than 500 software loopholes with CVE numbers, including 18 of the 24 recently repaired gaps in open-source network tool curl. Cohere plans to combine the AISLE technology with its own LLM, so that the enterprise can clear the loopholes in local servers or isolated networks and avoid the transmission of sensitive codes.
workflow, covering software gap analysis, risk screening, repair and validation (source: AISLE)
Another deal is directed towards the European market. In September this year, Cohere signed a merger agreement with the German company Alpha Alpha of AI, which plans to integrate their modelling capabilities and European government, corporate client resources. The Schwartz Group, the current financer, is also planning to deploy the Cohere model to the STACKIT cloud platform under the flag to provide AI services to European clients. Currently, both transactions have not been completed.
In addition to M&As, Cohere will receive financial support from the Government of Canada. According to an agreement published by the Prime Minister ' s Office of Canada on 29 October, the Government proposes to contribute up to $1.6 billion (approximately $7.6 billion RMB) in a phased manner to Cohere. As a condition for investment, Cohere committed to at least 1,000 new Canadian technical posts by 2030 and invested at least $1.7 billion (approximately $8.1 billion RMB) to build local computing facilities while retaining Canadian headquarters and registration.
Conclusion: From model development to enterprise deployment, AI started fighting for business software orders
This financing will support Cohere to continue the development of LLM and expand network security and the European business market through M&As. Together with local computing and cloud platform cooperation, Cohere is moving further from model development to the deployment and delivery of enterprise AI products.
Competition in the corporate AI market is also not limited to the model itself. This month, another AI office company, which received an investment from Weida, completed $250 million in financing and disclosed business clients competing for Microsoft, Oracle and IBM.
Although the product routes of the two companies differ, attempts are being made to bring AI into the business systems that exist in the enterprise and to take on practical tasks. For Cohere, the next challenge is not just model calls, but also the enterprise ' s software procurement budget in the areas of office, transport and network security.
Source: Reuters, Cohere
Auto-fetched and translated by AI Atlas. All rights belong to the original author.