Microsoft and OpenAI have gone their separate ways, and both companies are now preparing to compete head to head in the artificial intelligence market.
At its annual Build conference on Tuesday, Microsoft unveiled a series of new AI initiatives including a reasoning model, a cybersecurity tool, and AI agents that resemble the popular open source platform OpenClaw. The announcements send a clear signal that Microsoft intends to be a major independent player in AI, no longer relying on its former partner.
For years, Microsoft's AI strategy depended heavily on its early and exclusive partnership with OpenAI. But that relationship deteriorated over time and the two companies effectively separated in late April, though Microsoft remains OpenAI's primary cloud partner for now. This year's Build had the feel of a company finally striking out on its own. CEO Satya Nadella told the audience that developer conferences are about "coming to grips with the new opportunity."
Microsoft aims to become a top AI lab
AI chief Mustafa Suleyman was even more direct in an interview with The Verge. "The goal is to prove that we can become one of the top four labs in the world," he said. "There's three labs that matter, Google DeepMind, OpenAI, and Anthropic. We are not one of them at the moment, and that's always been my intention. It's why I came here. I want to build the very best frontier models in the world, fully multimodal, and in order to do that, we have to prove that we can do everything that we need to from the ground up, and we're not just going to take from others."
Suleyman introduced MAI-Thinking-1, Microsoft's first reasoning model, along with six other models focused on image, voice, transcription, and coding. The medium sized model is built for serious math, coding, and enterprise deployment, according to Microsoft. Suleyman said it is cheaper than OpenAI equivalents on some tasks, a significant advantage in the current AI market where customers are pressuring companies to lower costs.
Suleyman emphasized that MAI-Thinking-1 was developed without distillation, meaning it was not trained using another company's AI model. If the model performs well, Microsoft does not want anyone attributing its success to OpenAI's technology. He noted that the pivotal moment for Microsoft was renegotiating its contract with OpenAI, which allowed the company to train larger models with its own intellectual property and data, without distillation.
New cybersecurity tool and AI agents
Stay updated
Get the day's AI and automation news in your inbox. No spam, unsubscribe anytime.
Nadella also highlighted MDASH, a recently launched AI cybersecurity tool that brings together 100 AI agents to find exploitable bugs. He said it works "better than any single model." The announcement appeared to be a direct challenge to Anthropic's Claude Mythos Preview and OpenAI's own cybersecurity system. All three companies are competing for the government and enterprise market.
Microsoft's approach to AI agents is more complex. The open source platform OpenClaw has shown a potential path forward, and after OpenAI hired its creator Peter Steinberger, Microsoft is trying to catch up. At Build, Nadella said the company is committed to OpenClaw support. Steinberger made a surprise appearance on stage, saying that users can now run OpenClaw inside companies and that the harness itself is now a plug in. "You can totally run OpenClaw inside your company now," he said. Users can run OpenClaw on top of Copilot, Codex, or other coding platforms via Windows.
Suleyman repeatedly mentioned Microsoft's "humanist superintelligence" as an AI that prioritizes humanity first, part of a broader industry effort to rebrand AGI in a less threatening way.
Nvidia CEO Jensen Huang appeared via video call and touted how Nvidia's RTX Spark chip is powering Microsoft's AI agent ambitions. "The idea that the PC evolved from a personal computer to a personal AI is just really exciting," he said.
Advantages and unanswered questions
Microsoft spent years betting on OpenAI, which in some ways left it behind in the AI race. But it has significant advantages as competitors turn to enterprise to make money. Microsoft already has a large client base and a reputation for security. Like Google, it has deep pockets and a diversified revenue stream. Suleyman said the company can operate with more humility and long term optimization, adding that Microsoft has the money to buy models from Anthropic when needed and offers 11,000 models on Azure for customers to choose from.
However, there are many unanswered questions. Microsoft touted benchmark wins and advancements for its seven new models, but that does not always translate to real world adoption. AI super apps are largely untested, and Microsoft is entering a crowded but still underwhelming AI agent marketplace with a product that has not been seen in action. There is still room for its promises to fall short.
Related on Neura Market