Microsoft launched three proprietary artificial intelligence models this week under the MAI brand, taking direct aim at both OpenAI and Google.
The move marks the most assertive step yet in Microsoft's effort to reduce its dependence on OpenAI, the San Francisco-based company it has backed with billions of dollars. The three new models are designed to integrate directly into Microsoft's flagship enterprise products, including Teams and Copilot, with the explicit goal of replacing functions currently handled by OpenAI's technology.
Mustafa Suleiman, Microsoft's chief of AI, has indicated that the company's ambitions extend well beyond this initial rollout. He has pointed to 2027 as the target for Microsoft to develop frontier-level large language models across text, image, audio and video. If achieved, that would place Microsoft in direct competition with the very partner powering much of its current AI ecosystem.
What the MAI models do
The models are aimed primarily at developers and enterprise workflows. Performance benchmarks suggest that MAI-Transcribe-1, the speech recognition model in the lineup, delivers highly competitive multilingual accuracy, outperforming comparable offerings from rival providers in certain tests.
Microsoft is also using pricing as a strategic lever. The MAI models are positioned to undercut competitors, with a cost performance balance designed to attract developers currently building on rival APIs.
Alongside the models, Microsoft has introduced multi-model workflow capabilities within Copilot. These allow outputs from different AI systems, including OpenAI's GPT models and Anthropic's Claude, to be compared or validated against each other. The feature is designed to address a key enterprise concern, reliability of AI-generated outputs.
The OpenAI question
Microsoft's relationship with OpenAI is entering a more complex phase. While its investment has secured access to OpenAI's models for products like Copilot and Azure AI, the MAI launch signals a shift toward building in-house capabilities that could eventually replace those dependencies.
Internally, Microsoft is said to be pursuing "AI self-sufficiency," with the MAI models representing an early step in that direction. The company has not clarified how this strategy might affect its commercial agreements with OpenAI, and there has been no public response from OpenAI on the development.
The 2027 frontier goal
The ambition to build frontier-level models by 2027 presents significant challenges. Such systems require vast computational infrastructure, proprietary datasets and top-tier research talent, resources currently concentrated among only a handful of global players.
While Microsoft's Azure cloud platform provides substantial computing capacity, matching or surpassing the most advanced models from OpenAI or Google within that timeframe remains an open question.
Security concerns are also emerging. Some enterprise IT specialists argue that proprietary models require rigorous, independent audits before deployment in regulated sectors. Microsoft has not yet released a detailed security framework specific to the MAI models.
Market implications
The broader AI market context adds weight to Microsoft's move. OpenAI and Google currently dominate enterprise AI adoption conversations, particularly among large corporations.
By developing its own models, pricing them aggressively, and embedding them into widely used software products, Microsoft is leveraging a distribution advantage that standalone AI firms cannot easily replicate.
Whether that strategy translates into large-scale enterprise adoption, and ultimately allows Microsoft to reduce its reliance on OpenAI by 2027, will become clearer over the next few quarters.
Disclaimer: This article was produced with the assistance of artificial intelligence tool but vetted by human editor.
Also Read: Google Launches Gemma 4 Open-Weight AI Models Under Apache 2.0 License
