7月17日，月之暗面正式发布了其拥有2.8万亿参数的Kimi K3大模型，该模型在多项评测指标方面逼近甚至超越了美国顶尖的闭源模型，成功打破了海外高端大模型的技术垄断，在硅谷引起了强烈震动，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AI人工智能的“威胁论”，对开源模型的安全性和合规性提出了质疑，并联合相关人士向美国白宫施压，呼吁通过政策手段封杀Kimi K3。

当地时间7月24日，Meta、英伟达、微软等25家美国科技企业、投资机构以及开源组织联名发布公开信，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开源与闭源之争，表面上是技术路线与发展模式的分歧，本质上是商业利益的博弈，各方阵营的对立与表态，都围绕自身市场利益与行业地位展开。

OpenAI and Anthropic follow a core logic centered on preserving their elevated valuations. Prior to the emergence of Kimi K3, these two companies leveraged their cutting-edge closed-source models to secure a market technical advantage, thereby monopolizing the global high-end closed-source large-model market and attaining exceptionally high valuations, with Anthropic's valuation approaching 100 billion dollars and OpenAI's exceeding 850 billion dollars, while both are advancing their listing processes. Large models demand substantial capital inputs with extremely high funding requirements, and elevated valuations form the central support for their access to financing, maintenance of research and development iterations, and employee incentive mechanisms. For these two companies, the most favorable development pattern consists in establishing an industry oligopoly, relying on exclusive technical monopolies to control market pricing power and industry discourse authority, thereby generating super-normal monopoly profits. This represents the fundamental reason behind their resolute resistance to open-source models dividing the market.

Kimi K3 was released, which had a direct impact on the two giants. On one hand, compared to the high calling costs and limited cloud deployment modes of closed-source models, Kimi K3 and other open-source models support enterprise local private deployment, which not only greatly lowers the cost of AI landing but also effectively avoids the risk of core data leakage for enterprises, which can adapt to various high-security scenarios such as government and enterprises and high-end manufacturing, forming a strong alternative to closed-source models. On the other hand, Moonshot AI, which was valued at 31.5 billion dollars before investment, can produce models that are close to the top level, and the question mark that the two giants with valuations of ten billion dollars can also be hit by the market.

此外，硅谷众多企业支持开源模型的发布也是基于自身商业利益的考量。英伟达和AMD等硬件厂商充分认识到开源模型能够降低AI研发与应用门槛，让中小开发者以及中小企业广泛进入这一领域，从而全面激活算力需求，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。微软和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，从而提升平台竞争力与用户黏性。戴尔和IBM等下游应用企业则更依赖低成本且可定制的开源模型迭代产品，以拓展消费级市场。

OpenAI and Anthropic stand on one side while the other Silicon Valley enterprises stand on the other side, which itself points to a problem: the development of open-source models benefits the overall technology circle ecosystem, and suppressing the development of open-source models is harmful to the entire ecosystem.

当前全球AI产业仍处于早期发展阶段，技术迭代、场景落地、商业模式探索仍有极大空间，新兴产业的成长离不开多元主体参与和充分市场竞争。充分的竞争能够推动企业降本增效、优化技术、完善服务，加速AI技术普惠落地。更关键的是，科技创新从来没有唯一最优路径，单一闭源寡头垄断模式易固化技术路线、压缩多元创新空间。开源与闭源共存的多元格局，能够保留多样化创新方向，让行业在持续试错与迭代中筛选出高效、适配的技术路径，为AI产业长期发展注入活力。

Anthropic has repeatedly promoted the distillation theory to suppress open-source models, but in reality, model distillation is a conventional iterative method in the AI industry and represents a neutral technical approach. The top models from OpenAI and Anthropic are essentially products that integrate and distill massive amounts of publicly available human knowledge and industry technological achievements. Large-scale model distillation is a general technology in the industry, while the infringement controversy arises only from the illegal acquisition of competitor model weights for distillation purposes. Anthropic deliberately confuses the two scenarios to promote the narrative of distillation threats, which fundamentally represents a double standard employed to secure an oligarchic position in the industry.

A successful AI not only depends on its capabilities but also on its accessibility and inclusiveness, with policy restrictions able to influence the speed of open-source model development yet unable to alter its direction.

封面图片来源：每日经济新闻

来源：每经热评 | 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 | 每日经济新闻