英伟达拟收购或投资 Reflection AI
FT just reported Nvidia is in early talks to acquire, or invest further in, Refl…
AI 产业链重大并购动态，关注英伟达如何通过资本运作整合前沿模型能力，直接影响行业格局。
FT just reported Nvidia is in early talks to acquire, or invest further in, Reflection AI, a $25B open-weight model start-up.
《金融时报》（FT）刚刚报道，英伟达正在就收购或进一步投资 Reflection AI 进行早期谈判。Reflection AI 是一家估值 250 亿美元的开源模型初创公司。
One option is an acqui-hire, in which Nvidia hires Reflection’s staff and licenses its technology without the antitrust review a full takeover would trigger.
一种选择是“人才收购”（acqui-hire），即英伟达聘用 Reflection 的员工并许可其技术，从而避免全面收购所触发的反垄断审查。
Nvidia used that structure for its $20B Groq deal last December, and a deal could come within weeks, though the talks could still fall apart.
去年 12 月，英伟达曾采用这种结构完成了对 Groq 的 200 亿美元交易。尽管谈判仍可能破裂，但交易可能在数周内达成。
Nvidia is already one of Reflection’s largest shareholders after investing $800M and the Trump administration also wants Reflection to compete with China.
在投资 8 亿美元后，英伟达已成为 Reflection 的最大股东之一。此外，特朗普政府也希望 Reflection 与中国展开竞争。
Last week Reflection released Beam, a sparse mixture-of-experts model with 501B total parameters, 23B of them active, built for coding, reasoning and agent work.
上周，Reflection 发布了 Beam，这是一个稀疏混合专家（sparse mixture-of-experts）模型，总参数量为 5010 亿，其中活跃参数为 230 亿，专为编码、推理和智能体工作而构建。
Reflection says Beam is competitive with the larger GLM 5.2 on coding and agentic tasks, and matches its reasoning scores while using 3x to 4x less inference compute.
Reflection 表示，Beam 在编码和智能体任务方面具有与更大的 GLM 5.2 相竞争的实力，且在推理算力使用量仅为后者 3 到 4 倍的情况下，达到了相当的推理得分。
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力