Microsoft Shifts Copilot to In-House AI Models, Potentially Impacting Performance and Pricing for Indian Users

Microsoft is beginning to replace third-party AI models from OpenAI and Anthropic with its own internally developed MAI models across various Copilot products, including key applications like Excel and Outlook. This strategic shift is driven by a goal to reduce operational costs associated with licensing external AI technology. The move could have implications for the performance of Copilot features and future pricing models for users, including the growing number of businesses and individuals in India relying on Microsoft’s AI tools.

According to reports, Microsoft’s in-house MAI models are already processing tens of thousands of requests weekly within Excel and Outlook. While these represent a fraction of the total AI interactions, the company’s long-term objective is to significantly decrease its reliance on external AI providers. The MAI models are also being integrated into GitHub Copilot, with a proprietary transcription model expected for Teams soon.

Cost Reduction as Primary Driver

Microsoft’s head of AI, Mustafa Suleyman, has openly stated the company’s intention to “reduce and ultimately eliminate” the cost of using external models from partners like Anthropic. This cost-cutting measure is a significant factor behind the acceleration of in-house AI development and deployment. While Microsoft has previously advocated for platform neutrality and choice in AI models, this current strategy suggests a stronger push towards proprietary solutions to manage expenses.

The transition raises questions about the balance between cost savings and AI performance. At its Build conference, Microsoft introduced seven new AI models, including MAI-Thinking 1, positioned as its first reasoning model. While Microsoft claimed MAI-Thinking 1 could rival models like Sonnet 4.6 and Opus 4.6 in coding performance based on internal human evaluations, earlier benchmarks indicated it trailed OpenAI and Anthropic models, performing more comparably to Deepseek V3.2. This disparity suggests that the cost savings might come with a trade-off in raw performance for certain tasks.

Potential Impact on Users and Pricing

For existing Copilot and Office subscribers, this shift could mean paying the same subscription fees for AI capabilities powered by different, potentially less performant, models. The scenario could lead to a re-evaluation of Copilot’s value proposition for users who have grown accustomed to the capabilities offered by OpenAI and Anthropic models.

Microsoft CEO Satya Nadella has also hinted at a potential shift towards usage-based AI billing rather than flat-rate subscriptions. One possible future model could involve making the cheaper MAI models the default option, with premium third-party models from OpenAI or Anthropic available as add-ons at an extra cost. This approach would effectively transfer the expenses of external AI models to customers who require their advanced capabilities, potentially influencing how Indian businesses budget for and utilize AI tools.

Data Training and Commercial Licensing

Microsoft has asserted that its MAI models are trained on “clean, commercially licensed data,” promoting them as safe for business use. However, technical papers reveal that Microsoft, like many other AI companies, utilizes datasets such as Common Crawl. The legal standing of using Common Crawl data for AI training is not definitively settled, making Microsoft’s claims of uniquely clean data a point of ongoing discussion within the industry.

Key facts:

Feature	Detail
Shift From	OpenAI and Anthropic AI models
Shift To	Microsoft’s in-house MAI models
Primary Goal	Reduce external AI licensing costs
Impact	Potential changes in AI performance and future pricing structures

This development is particularly relevant for Indian businesses and developers who are increasingly integrating AI into their workflows using Microsoft’s ecosystem. Any changes in Copilot’s performance or pricing structure could influence technology adoption strategies and operational costs across various sectors in India, from startups to large enterprises. Users may need to assess whether the cost efficiency of the MAI models aligns with their performance requirements or if a premium upgrade for third-party models becomes necessary.

Source: The Decoder, https://the-decoder.com/copilot-goes-cheap-as-microsoft-phases-out-openai-and-anthropic-models-to-cut-costs/