Microsoft has been one of OpenAI’s biggest supporters, investing nearly $14 billion to help the AI powerhouse grow. But now, things are shifting. Reports suggest Microsoft is building its own in-house AI—possibly called MAAI (Microsoft AI)—to reduce its reliance on OpenAI. Why the sudden change? Let’s break it down.
The Rising Costs of AI
One major factor seems to be the skyrocketing costs associated with OpenAI’s technology. OpenAI is rumored to be launching AI agents that could cost a jaw-dropping $20,000 per month, with top-tier research agents reaching $220,000 per month. Companies like SoftBank are already committing billions to access these AI agents, believing they can replace highly paid finance and research professionals.
For Microsoft, these costs add up fast. Running large AI models isn’t cheap—data center expenses, specialized AI chips, and infrastructure costs are enormous. Some AI chips alone range from $10,000 to $30,000 apiece, and a full server rack can cost over $500,000 before even powering it up.
Microsoft’s Own AI Models: MAAI
Instead of relying entirely on OpenAI, Microsoft has reportedly been training its own MAAI models, which may match the capabilities of OpenAI’s GPT models and competitors like Anthropic. These models could power Microsoft’s Copilot, which helps users with tasks like writing documents, coding, and managing spreadsheets.
Microsoft is also testing third-party AI models from Meta, xAI, DeepSeek, and Anthropic, aiming to find the best balance between performance and cost.
Microsoft & OpenAI: Frenemies?
Microsoft and OpenAI started out as close partners, with Microsoft being OpenAI’s exclusive cloud provider on Azure. However, that exclusivity has changed. Now, OpenAI is free to use other cloud providers, such as Oracle. This shift indicates Microsoft’s desire for flexibility and independence.
Meanwhile, Microsoft itself is investing heavily in AI infrastructure, planning to spend $80 billion in AI data centers by 2025, with significant investments in India. Additionally, major tech players like SoftBank, Oracle, and Nvidia are reportedly considering a $500 billion investment in next-generation AI hardware.
The Belief State Transformer: A Game-Changer?
Beyond MAAI, Microsoft Research is working on a revolutionary AI technique called the Belief State Transformer (BST). Traditional AI models process text left to right, meaning they predict words based only on past context. BST, however, processes both past and future tokens, leading to better planning and reasoning.
This approach solves a common problem in AI: models often “cheat” by making the next best guess without truly understanding long-term goals. BST forces AI to predict both next and previous tokens, preventing shortcuts and leading to better decision-making. Early tests show BST significantly outperforms traditional AI methods in tasks like navigation and storytelling.
What This Means for the AI Industry
With Microsoft developing its own models, OpenAI’s exclusive dominance is being challenged. The AI landscape is shifting, and we could see:
- More AI options: With multiple strong AI providers, businesses may have more choices beyond OpenAI.
- Lower prices?: If competition heats up, AI services might become more affordable—or at least, alternatives may emerge.
- Enterprise focus: AI companies may target deep-pocketed businesses instead of casual users, keeping costs high.
Either way, AI is evolving rapidly, and the competition between Microsoft, OpenAI, and others will likely drive faster innovation.
What do you think? Will Microsoft’s MAAI be a real competitor to OpenAI, or is this just a cost-cutting move? Let’s discuss in the comments!