Why Microsoft is swapping out high-end AI for its own

Microsoft is moving away from a total reliance on outside specialists like OpenAI and Anthropic—the companies that build the heavy-duty AI systems behind tools like ChatGPT. Instead, Microsoft is embedding its own, in-house software, called MAI models, to handle some of the requests users make within Office apps. While they still work with these partners, the strategy is shifting toward controlling the underlying technology themselves to bring down the high price tag of running these tools.

To understand why this is a cost-cutting move, we have to look at how AI actually runs. Every time you ask an AI to summarize a document or write an email, it consumes computational resources, often measured in tokens. Think of a token as a small chunk of text, like a syllable. The more complex the model, the more energy and computing power it needs to process your request. Using a massive, high-powered model for a simple task is like using a fleet of heavy-duty semi-trucks to deliver a single envelope. It gets the job done, but it is an incredibly inefficient and expensive waste of fuel. By building smaller, custom models meant for specific tasks—like fixing a spreadsheet formula or proofreading a paragraph—Microsoft can deliver similar results using far less computing power. They are effectively swapping the semi-truck for a bike messenger.

This shift is a reality check for the entire tech industry. For the last couple of years, companies have been rushing to use the biggest, smartest models available as quickly as possible. Now, the novelty is wearing off and the reality of the electricity and hardware costs is setting in. For the average person, this means AI tools might start feeling more tailored and specific rather than generic and bulky. For tech giants, it marks the end of the initial rush to buy the most expensive AI power available and the beginning of a hunt for efficiency. When these companies get better at building their own lean models, they become less dependent on expensive outside labs, which could change who actually holds the power in the AI market.