月之暗面于7月17日正式发布了参数规模达到2.8万亿的Kimi K3大模型，这一发布使得多项评测指标逼近甚至超越了美国顶尖的闭源模型，从而成功打破了海外高端大模型的技术垄断，同时在硅谷引发了强烈的震动，也直接触动了OpenAI以及Anthropic的核心利益。两家闭源巨头的高管随即大肆渲染了人工智能的“威胁论”，质疑开源模型的安全性以及合规性，并联合相关人士向美国白宫施压，呼吁通过政策手段来封杀Kimi K3。

与两大闭源巨头极力打压的架势不同，硅谷多数科技企业坚守开源理念，立足产业整体利益，坚决反对技术封锁和行业垄断，抵制针对Kimi K3的歧视性、封杀式政策。当地时间7月24日，Meta、英伟达、微软等25家美国科技企业、投资机构与开源组织联合发布公开信，呼吁放开AI模型权重开放，支持开源AI生态自由发展。

纵观这场席卷硅谷的AI开源闭源之争，表面是技术路线与发展模式的分歧，本质上是商业利益博弈，各方阵营的对立与表态，皆围绕自身市场利益与行业地位展开。

OpenAI and Anthropic adopt core logic that centers on maintaining their own high valuations. Prior to the release of Kimi K3 these two companies leveraged frontier closed-source models to secure market technical advantages, thereby monopolizing the global high-end closed-source large-model market and obtaining extremely high valuations, with Anthropic valuation approaching 100 billion dollars and OpenAI exceeding 850 billion dollars, and both advancing the listing process. AI large models require high investment with extremely large funding needs, and high valuations constitute the core support for their acquisition of financing, maintenance of research and development iteration, and promotion of employee incentives. For these two companies the most favorable development pattern consists in forming industry oligopolies, relying on exclusive technical monopolies to master market pricing power and industry discourse authority, and thereby earn super-normal monopoly profits. This also constitutes the fundamental reason for their resolute resistance to open-source models that fragment the market.

Kimi K3 comes into being and poses a direct impact on the two giants. On the one hand, compared with the high cost of calling the closed-source models and the restricted cloud deployment mode, the open-source models such as Kimi K3 support enterprise local private deployment, which not only greatly lowers the cost of AI landing, but also can effectively avoid the risk of enterprise core data leakage, which adapts to various high-security demand scenarios such as government enterprises and high-end manufacturing, and forms a strong alternative to the closed-source models. On the other hand, the latest pre-investment valuation of 31.5 billion dollars by Moonshot AI, can make models that are close to the top level, then whether the two giants can reach a trillion dollar valuation will also be questioned by the market.

此外，硅谷众多企业力挺开源模型，也是基于自身商业利益的考量。英伟达、AMD等硬件厂商深知，开源模型能够降低AI研发与应用门槛，让中小开发者、中小企业广泛入局，全面激活算力需求，带动GPU（图形处理器）、算力芯片等硬件销量增长，持续拓宽自身营收空间。微软、亚马逊等云厂商则希望打破头部模型垄断，依托更多的开源模型丰富云端服务体系，为客户提供更多选择，提升平台竞争力与用户黏性。戴尔、IBM等下游应用企业，更依赖低成本、可定制的开源模型迭代产品、拓展消费级市场。

OpenAI and Anthropic stand on one side, while the other companies in Silicon Valley stand on the other. This itself reveals a problem. The development of open-source models benefits the overall ecosystem of the technology industry, and suppressing the development of open-source models is harmful to the entire ecosystem.

Current global AI industry remains in its early stage of development, with technology iterations, scenario deployments, and commercial model explorations still offering considerable space for advancement. The growth of emerging industries depends on the participation of diverse entities and adequate market competition. Adequate competition enables enterprises to reduce costs and increase efficiency, optimize technologies, and improve services, thereby accelerating the ubiquitous adoption and practical deployment of AI technology. More importantly, technological innovation has never possessed a single optimal path. A single closed-source oligopoly model tends to solidify technological routes and compress the space for diverse innovations. The diversified pattern of coexistence between open-source and closed-source can retain various directions for innovation, allowing the industry to screen for efficient and adaptable technological paths through continuous trial and error along with iteration, thus injecting vitality into the long-term development of the AI industry.

为了打压开源模型，Anthropic曾反复炒作蒸馏论，但是实际情况是，模型蒸馏是AI行业常规迭代手段，是一种中性的技术方法，OpenAI以及Anthropic的顶尖模型本质上也是整合蒸馏人类海量公开知识与行业技术成果的产物，合规模型蒸馏是行业通用技术，非法窃取竞品模型权重蒸馏才存在侵权争议，Anthropic刻意混淆两类场景炒作蒸馏威胁，本质是双重标准，以此谋求行业寡头地位。

成功的AI不仅在于其能力，更在于其可得性以及普惠性。政策限制或许可以影响开源模型发展的速度，却无法改变它的方向。

免责声明：本文内容与数据仅供参考，不构成投资建议，使用前请核实。据此操作，风险自担。

每经评论员| 杜恒峰

编辑 | 张锦河

校对| 程鹏

｜每日经济新闻 nbdnews 原创文章｜

未经许可禁止转载、摘编、复制及镜像等使用

来源：开源闭源之争实为商业利益博弈，美国两大巨头想成寡头，不利于AI发展 | 每经热评 | 每日经济新闻