Microsoft and OpenAI broke up — now they’re ready to fight Microsoft 和 OpenAI 分手了 —— 现在他们准备开战

Microsoft’s annual Build conference was never just about developer tools; this year, it served as a coronation ceremony. The company unveiled a super app, custom-built reasoning models, a dedicated cybersecurity AI, and a fleet of autonomous agents. But the real story isn’t in the product lineup. It’s in the timing. These announcements crystallize a strategic pivot years in the making: Microsoft is formally, decisively, and publicly moving out of OpenAI’s shadow. The partnership isn’t over—it’s now one among many, and Microsoft is finally acting like the owner of its own AI destiny, not merely the primary investor and distribution channel for someone else’s.

For the better part of five years, Microsoft’s AI strategy was a masterclass in leveraging a single, brilliant asset. The partnership with OpenAI was visionary and mutually transformative. It gave Azure an unparalleled cloud AI narrative and gave OpenAI the infrastructure and enterprise reach it desperately needed. But like any relationship built on a singular, explosive connection, it eventually hit limits. The drama—the boardroom coups, the corporate restructuring, the shifting equity stakes—exposed a fundamental tension. Microsoft was building an empire, but its crown jewel was a separate kingdom with its own royal court and caprices.

The shift began subtly, with Microsoft’s own smaller, more specialized models like Phi. But Build was the public declaration of independence. The “reasoning models” aren’t just another GPT wrapper; they’re an explicit play to own the intellectual engine of AI reasoning, a core competency Microsoft previously outsourced. The super app vision, integrating AI agents into the fabric of Windows and its 365 suite, is about creating an ecosystem where the intelligence is native, not borrowed. You don’t build a bespoke superstructure to house someone else’s foundation unless you plan to eventually pour your own concrete.

This is a maturation from platform partner to platform owner. Satya Nadella’s Microsoft has always understood that control of the platform—and the developer experience atop it—is the ultimate moat. By diversifying its AI portfolio with internal research (from Microsoft Research), acquisitions (like Inflection’s team), and new partnerships, it insulates itself. It reduces the existential risk of relying on a single, volatile external provider. The message to enterprises is now clearer and more compelling: we don’t just sell you access to OpenAI’s models; we offer a comprehensive, resilient, and deeply integrated AI fabric for your entire business, some of which we built from the ground up.

One can’t help but view this through the lens of a long-term corporate divorce settlement. The "separation" in late April was just the legal paperwork. Build was the first holiday where the kids (the Azure customers and Windows users) had to choose which house to go to. Microsoft’s answer was to build a bigger, more fully equipped house of its own. The cybersecurity tool is a perfect example: it’s not a generic chatbot, but a specialized agent trained on Microsoft’s own unparalleled threat intelligence. That’s a value proposition OpenAI simply cannot replicate.

The risk, of course, is hubris. Building your own frontier models is astronomically expensive and technically perilous. Microsoft could stumble, or its models could underperform, leaving it with a half-built house. Furthermore, OpenAI is not standing still. Its partnership with Apple and its own enterprise pushes are a direct threat. But Microsoft’s play isn’t necessarily to beat OpenAI at its own game of generating the most dazzling, general-purpose chatbot. Its play is to weave AI so deeply into the workflows of global industry that the underlying model becomes a commoditized ingredient, while the platform, integration, and proprietary data loops become the true value. In that game, having your own models isn’t just an option; it’s a prerequisite for control and margin.

What we witnessed at Build was a tech giant finally stepping fully into its own narrative. The years of deferential partnership are over. Microsoft is now wielding its formidable engineering resources, distribution muscle, and enterprise trust to build an AI ecosystem on its own terms. It’s a brasher, more self-assured, and frankly more dangerous Microsoft. The era of the co-dependency is done; the era of competitive coexistence—and all the bitter battles that entails—has just begun.

微软在Build大会上摆出的姿态，像极了一个长期依赖输血的病人突然拔掉针管，自己站了起来，还顺手抄起几把手术刀。这次发布的一系列AI工具——超级应用、自研推理模型、网络安全方案、Agent平台——与其说是产品堆砌，不如说是微软对整个行业发出的独立宣言：我不再只是OpenAI最大的买单者，我要成为规则制定者本身。

过去几年微软的AI战略，本质上是一场精心计算的“借壳上市”。通过巨额投资绑定OpenAI，微软用对方的技术光环掩盖了自家研究院不温不火的尴尬。但当OpenAI董事会风波暴露出这段关系的脆弱性，当GPT系列模型的进化开始明显滞后于宣传，微软终于意识到：把核心命脉押注在一家时刻可能“暴雷”的初创公司身上，就像在流沙上盖摩天大楼。四月底的所谓“分手”早有预兆——表面仍是云服务合作，实则微软已经在OpenAI的围墙外疯狂培育自己的种子。

最耐人寻味的是自研推理模型的推出。微软此前在基础大模型领域几乎毫无存在感，Bing Chat所用的Prometheus模型始终被视为“GPT的套壳”。如今突然推出具备链式推理能力的自研模型，摆明了是要在最关键的智能核心层建立自主权。这就像一个长期代理别人品牌的工厂，突然宣布自己的专利设计投产——技术自主权的话语权，远比分账比例更重要。

至于那个“超级应用”的构想，则暴露了微软真正垂涎的战场。在移动互联网时代错失先机的微软，如今想用AI重新定义“入口”。把搜索、办公、创作、编程工具全部塞进一个由AI驱动的超级界面，这分明是在对苹果的封闭生态和谷歌的分散布局同时宣战。但危险在于，微软历史上从不擅长打造消费者级的优秀体验。把几十个功能强行糅合的结果，很可能是一个臃肿又分裂的怪胎。

网络安全工具的发布则完全是另一回事，这才是微软最扎实的牌。在企业安全领域，微软本就拥有庞大数据和场景优势。将AI威胁检测深度集成到Azure生态中，就像给企业网络装上了全天候的神经中枢。这个领域没有炒作，只有实打实的攻防较量，而微软在这里突然出手，显然是看到了比消费端更稳固的盈利前景。

Agent平台的战略意图最为明显，这基本是给OpenAI的“GPTs”生态直接上强度。微软利用Windows和Office的装机量优势，试图把每个企业用户都变成AI Agent的开发者。但历史经验告诉我们，微软的开放生态承诺往往虎头蛇尾——还记得Windows Phone的开发者承诺吗？真正的考验在于，微软能否建立起比OpenAI更健康的激励体系，而不仅仅是把现有软件用户强行迁移过来。

整个发布会最讽刺的对比在于：当OpenAI还在为Sam Altman的演讲服装和公司治理结构焦头烂额时，微软已经用冰冷的工程化节奏，把AI野心拆解成可落地的产品模块。这或许揭示了AI竞赛的本质——最终胜出的可能不是最有灵感的研究机构，而是最能把技术模块化、系统化、产业化的巨头。

微软此刻的独立宣言，与其说是技术突破，不如说是战略觉醒。它承认了过度依赖单一技术合作伙伴的风险，开始用帝国自身的资源浇灌AI野心。但这条自立之路布满荆棘：既要避免重蹈Windows Mobile时代封闭生态的覆辙，又要与亚马逊、谷歌在企业市场短兵相接。当微软AI不再躲在OpenAI的光环背后时，它将真正接受行业最残酷的质问：剥离了GPT的魔法外衣，你究竟能为世界提供什么不可替代的价值？答案可能藏在下一个季度的财报里，而不是今天舞台上的炫目Demo中。