GitHub的Universe大会周三在旧金山开幕，会上传出消息称，其基础设施版图正在扩张，以应对AI智能体带来的惊人需求。2026年9月，GitHub记录的代码提交量达到74亿次，是去年同期服务器所承受提交量的五倍。过去一年流量的激增——源于开发者运行AI智能体并生成越来越多的编程代码——让公司措手不及，并导致了一系列宕机和服务可用性问题。
GitHub's Universe conference kicked off in San Francisco on Wednesday with news that its infrastructural universe is expanding to address staggering demand from AI agents. GitHub recorded 7.4 billion code commits in September 2026, five times more than hit its servers a year earlier. This surge in traffic over the past year, the result of developers running AI agents and generating increased amounts of programming code, caught the company unprepared and led to a series of outages and service availability woes.
AI驱动的代码提交去年12月开始成为问题，并在2月、4月、8月引发故障，更不用说其间的其他月份了。根据对GitHub状态页面数据的重建，5月7日，该网站90天可用率降至约84%的低点。GitHub高管为此道歉，并承诺改进。这些承诺已在其基础设施路线图中列入400万个CPU核心和160拍字节的高速存储，同时为GitHub Actions增加额外算力——这是该公司备受困扰的持续集成与持续交付（CI/CD）平台。
AI-driven code commits started becoming a problem last December, and led to problems in February, April, August, not to mention the months in between. On May 7, the site's 90-day availability hit a low of about 84 percent, according to reconstructed data from GitHub's status page. GitHub execs apologized and promised to do better. Those commitments have put four million CPU cores and 160 petabytes of high-speed storage on its infrastructure roadmap, alongside extra compute capacity for GitHub Actions, the company's beleaguered continuous integration and continuous delivery (CI/CD) platform.
这家由微软拥有的代码托管公司还表示，正在重新设计其技术架构（此前已有讨论），以将计算与存储分离，预计这将带来35倍的吞吐量提升。“我们还在迁移到Azure的多区域部署，同时继续为全球开发者提供惊人规模的流量服务，”微软CoreAI执行副总裁杰伊·帕里克在提供给The的一份博客文章中表示。“我们预计将在明年初完成迁移。这些投资为我们提供了更大的增长空间。下一步是构建处理这些工作的系统。”
The Microsoft-owned code hosting biz also says it is redesigning its technical architecture (an effort discussed previously) to separate compute and storage, with the expectation that will enable 35x more throughput. "We’re also migrating to a multi-region deployment in Azure, while continuing to serve incredible levels of traffic from developers around the world," said Jay Parikh, EVP of CoreAI, Microsoft, in a blog post provided to The. "We expect to complete the migration early next year. Those investments give us more room to grow. The next step is building the systems that handle the work."
同样在其Universe活动上，GitHub宣布Copilot Fusion正式全面可用，这是该公司的模型路由服务，此前以HydraFusion的名称进行预览。
Also at the its Universe event, GitHub announced general availability of Copilot Fusion, the company's model routing service that was previewed under the name HydraFusion.
这样做的想法是，允许 GitHub 将请求路由到不同的模型可以提高任务完成率，同时降低代币消耗成本。“不同的任务需要不同的模型，”Parikh 说。“开发者不必管理每一个模型选择，就能以合适的价格获得正确的结果。Copilot Fusion 会评估一项任务在推理、代码生成、调试和工具使用方面的需求，然后选择如何跨一个或多个模型执行。在人工智能分析编码代理指数中，Fusion Max 的得分与 Opus 5.5 Max 仅相差两分，但成本减半，每项任务的时间缩短了 40%。”
The idea is that allowing GitHub to route requests to different models can improve task-completion rates while reducing token consumption costs. "Different tasks need different models," said Parikh. "Developers shouldn’t have to manage every model choice to get the right result at the right price. Copilot Fusion evaluates what a task needs across reasoning, code generation, debugging, and tool use, then chooses how to execute across one or more models. In the Artificial Analysis Coding Agent Index, Fusion Max scored within two points of Opus 5.5 Max at half the cost and 40% less time per task."
GitHub 正通过其 Copilot 应用提供更多附加功能，包括用于协调并行工作的 Copilot Projects；用于简化 Copilot 连接到 MCP 服务器方式的 Copilot Connectors；以及可重用的代理任务 Copilot Automations。为了促进 Copilot 在工作场所的使用，GitHub 正在将 Copilot 的访问权限扩展到所有 GitHub Enterprise 用户，而无需额外按席位购买。
GitHub is making more bells and whistles available through its Copilot app, including Copilot Projects, for coordinating parallel work; Copilot Connectors, for simplifying how Copilot connects to MCP servers; and Copilot Automations, which are reusable agent tasks. And in what appears to be an effort to boost Copilot usage in the workplace, GitHub is extending Copilot access to all GitHub Enterprise users without requiring any additional per-seat purchases.
客户仍需为他们使用的代币付费。微软的代码协作业务还通过一项名为 AI Scans 的服务扩展了其面向开发者的 AI 服务套件，该服务可以检查拉取请求中用规则扫描无法解决的语言编写的漏洞，以及已合并代码中的漏洞，此外还有新的可达性分析工具，使开发能够专注于可能被利用的问题。另一项新服务是一个包防火墙，在获取 npm 包之前会检查企业策略设置。
Customer still get billed for the tokens they use. Microsoft's code collab biz is also expanding its suite of AI services for developers with an offering called AI Scans, which checks pull requests for vulnerabilities in languages not addressed by rule-based scanning and in code that's already merged, plus new reachability analysis tools that allow development to focus on issues that are plausibly exploitable. Another new offering is a package firewall that checks enterprise policy settings before fetching npm packages. ®