For three years, Microsoft's artificial intelligence story has been inseparable from OpenAI. The partnership — cemented by a cumulative investment exceeding $13 billion — gave Microsoft early access to the most advanced AI models on the planet, catapulting its Copilot products into the enterprise mainstream and adding hundreds of billions of dollars to its market capitalization. To the outside world, Microsoft's AI strategy was OpenAI. Mustafa Suleyman wants to change that narrative. In an exclusive sit-down interview with VentureBeat at Microsoft Build 2026, the CEO of Microsoft AI disclosed that a contractual change with OpenAI roughly six months ago granted his division the formal authority to pursue what he openly calls "superintelligence" — using Microsoft's own researchers, its own data pipelines, and its own custom silicon. "We were only sort of set free from our contract with OpenAI about six months ago to formally pursue superintelligence," Suleyman said. "So this is very early days." The comment, delivered matter-of-factly backstage at the Fort Mason Center here, offers the clearest signal yet of a strategic inflection point unfolding inside the world's most valuable public company. Microsoft is not abandoning OpenAI. But it is building something alongside it — and, eventually, something that could stand entirely on its own. Microsoft's first in-house model family signals a new level of AI ambition The most tangible evidence of that shift arrived the same day. Microsoft announced a family of seven new AI models developed entirely in-house by its AI Superintelligence Team, spanning reasoning, code generation, image creation, transcription, and voice synthesis. The models — branded under the "MAI" family name — are Microsoft's most ambitious first-party AI release to date. The flagship, MAI-Thinking-1, is a 35-billion-active-parameter reasoning model that Microsoft says matches leading models in its weight class on key software engineering benchmarks and demonstrates advanced mathematical reasoning. Suleyman emphasized one point repeatedly: the model was trained from scratch on clean, commercially licensed data, without distillation from third-party frontier models — a direct, if unstated, contrast to the widespread industry practice of using outputs from competitors' systems to train cheaper alternatives. "We train our reasoning models from scratch," Suleyman wrote in a blog post accompanying the announcement. "We don't distill from other labs and we don't rely on unlicensed or opaque data." The rest of the family fills out a multimodal portfolio designed for enterprise deployment: MAI-Code-1-Flash, a lightweight coding model built specifically for GitHub Copilot and VS Code; MAI-Image-2.5, which supports both text-to-image and image editing; MAI-Transcribe-1.5, which Microsoft claims is the most accurate transcription model available, operating across 43 languages; and MAI-Voice-2, a multilingual speech-generation system. All of the models ship through Microsoft Foundry, the company's model-hosting and deployment infrastructure, and for the first time, developers can tune model weights themselves through third-party platforms including OpenRouter, Fireworks, and Baseten. But Suleyman made clear in the interview that the seven models are a proof of concept, not a finished product. The real project is the lab itself. "Our job is to make sure that when we look out to 2030 and beyond, we have the capacity not just to buy models from third parties, but to build the absolute frontier, the best models in the world," he said. "That's a long transition." What "set free" from OpenAI actually means for Microsoft's AI future To understand what Suleyman means by "set free," you need to understand the unusual contractual architecture that has governed Microsoft's AI efforts for years. When Microsoft invested billions into OpenAI beginning in 2019, the partnership came with a specific arrangement: OpenAI would build the frontier models, and Microsoft would serve as the exclusive cloud provider, integrating those models into its products and reselling them through Azure. The deal gave Microsoft extraordinary commercial leverage — access to the world's most advanced AI without having to build it — but it also created a dependency. Microsoft was explicitly barred from pursuing its own AGI research, and the agreement even capped how large a model the company could train, restricting it from building systems beyond a certain computing threshold measured in FLOPS. That arrangement was formally renegotiated. As Fortune and Axios reported in November, a revised deal with OpenAI removed those restrictions, clearing the way for Suleyman to launch the MAI Superintelligence Team and pursue what he calls "humanist superintelligence." The result, in Suleyman's telling at the time, was a "best-of-both environment, where we're free to pursue our own superintelligence and also work closely with them." By the time he sat down with VentureBeat at Build 2026, roughly six months had passed since that self-sufficiency effort formally began. Microsoft had already started shipping in-house models — including MAI-Image-2-Efficient, a lighter-weight image generation model released in April — but the seven MAI models announced at Build are the team's most ambitious release yet: a full multimodal family spanning reasoning, code, image generation, transcription, and voice. Even so, Suleyman does not view the shift as a rupture with OpenAI. He described Microsoft's current position as one of abundance, not scarcity. "There's no immediate urgent need to fill a gap in three months' time or six months' time," he said. "We have OpenAI, we have Anthropic, we have thousands of models inside Foundry. So there's already a huge amount of optionality available to us." The framing is telling. Microsoft's push into first-party frontier models is not born out of a crisis in the OpenAI relationship but out of a strategic calculation: as AI becomes the most consequential technology layer in enterprise computing, the company cannot afford to depend entirely on partners for the foundational capability. "Over the next five years, we have to be able to produce state-of-the-art frontier-scale models," Suleyman said. "That's our mission." Suleyman says the shift from chatbots to autonomous AI agents has already begun If the seven MAI models represent the technical ambition, a new capability called Frontier Tuning represents the commercial logic. Announced alongside the models at Build, Frontier Tuning allows enterprise customers to customize MAI models using their own proprietary

O cofundador e CEO da Anthropic, Dario Amodei, disse que isso estava chegando, mas ainda parece um marco: mais de 80% do código

Embora muitos fornecedores de modelos de código aberto de IA busquem modelos maiores e mais poderosos, o Google ainda está prestando atenção ao

A Alibaba lançou esta semana o Qwen3.7-Plus, o mais recente modelo de IA em linguagem grande (LLM) em sua família Qwen, globalmente amada

Os agentes corporativos de IA têm um novo modo de falha de produção, e esse não é o modelo. À medida que as

Apresentado por Floco de Neve Muitas vezes, a história da segurança empresarial tem sido uma história de tornar as coisas mais difíceis de

Permitir que os LLMs adquiram novos conhecimentos após a formação continua a ser um grande obstáculo para a IA empresarial – as soluções

Os agentes de IA empresarial estão paralisados ​​– não por causa do desempenho do modelo, mas por causa do licenciamento. Todo fluxo de

À medida que os agentes empresariais de IA entram em produção, as organizações enfrentam um problema crescente de confiabilidade. Muitas equipes estão descobrindo