Microsoft Copilot: goodbye to OpenAI and Anthropic, MAI models are here

What changed: Microsoft cuts ties with OpenAI and Anthropic

Microsoft has initiated a quiet but significant transition. Within products like Excel and Outlook, AI models from OpenAI and Anthropic are being progressively replaced with proprietary MAI (Microsoft AI) models. According to reports from The Decoder, tens of thousands of weekly queries are already going through these new internal models.

Mustafa Suleyman, Microsoft's head of AI, has stated the intention to «permanently eliminate» the cost of external models. Therefore, the direction is clear: reduce dependence on third-party providers and lower the marginal cost of each AI interaction. However, this financial optimization raises a direct question for user companies: will performance remain unchanged?

The MAI Model: Internal Architecture and Strategic Positioning

MAI models are not entirely new. Microsoft has been developing them internally for some time, with a focus on computational efficiency and native integration with the Microsoft 365 ecosystem. In fact, their adoption in products like Excel and Outlook suggests specific optimization for structured tasks: data analysis, email summarization, and formula generation.

However, OpenAI's models—particularly GPT-4o—and Anthropic's models like Claude 3 have been chosen in the past precisely for their complex reasoning and natural language processing capabilities. In contrast, MAI models seem to prioritize speed and cost per query over processing depth. This trade-off is not neutral for companies using Copilot for high cognitive value tasks.

We of SHM Studio we closely monitor these types of developments. In particular, we evaluate how infrastructural changes in AI models impact Artificial intelligence services that we integrate for our clients.

Immediate impact for companies using Copilot

For marketing and IT managers, the operational question is concrete. Those who have integrated Copilot into their workflows—from report writing to campaign management, from sales data analysis to brief summarization—may notice qualitative variations in the outputs. Furthermore, there is currently no official communication from Microsoft guaranteeing equal performance between old and new models.

According to Gartner, the quality of the underlying model is one of the critical factors in evaluating the ROI of enterprise AI tools. Consequently, even partial degradation of Copilot's capabilities could translate into concrete inefficiencies, especially for companies that have built automated workflows around these tools.

Beyond this, there's a transparency issue. Companies pay a Copilot subscription without knowing precisely which model is processing their requests. Therefore, the silent replacement of models poses an AI governance problem that IT managers cannot ignore.

The construction site still open: Microsoft navigating savings and reputation

Microsoft's move should be seen in a broader context. Last year, the race for AI investments put pressure on the margins of all major players. Microsoft, which has invested billions in OpenAI, now finds itself needing to balance operating costs with shareholder expectations. Therefore, internalizing the models is a financially rational response.

However, the reputational risk is real. Copilot was sold as a tool powered by the best models available on the market. If the perception of quality were to decline, companies might reconsider adoption or reduce purchased seats. Similarly, competitors – from Google with Gemini to Salesforce with Einstein AI – could leverage this moment to position themselves as more transparent alternatives.

According to an analysis by Harvard Business Review, trust in the AI provider is a critical asset for enterprise organizations. Therefore, Microsoft will need to accurately communicate the capabilities of its AI models to avoid erosion of the Copilot customer base.

What to do now: operational guidelines for marketing managers

First, it is advisable to conduct an internal review of workflows that depend on Copilot. In particular, it is recommended to identify tasks where the quality of AI output is critical — for example, content generation, semantic analysis, or the synthesis of complex documents.

Subsequently, it is useful to compare Copilot's outputs from recent weeks with previous ones to detect any qualitative variations. This type of audit, even if informal, provides concrete data before making decisions about renewals or alternatives. Furthermore, it is worth exploring whether Microsoft will offer options for selecting the underlying model for enterprise plans, as already happens in other API contexts.

• Audit of AI-dependent workflows: Map critical tasks that use Copilot daily.
• Qualitative benchmark: Compare recent outputs with those produced in previous months on identical tasks.
• Alternative assessment consider tools like digital marketing solutions that integrate selectable AI models.
• Microsoft Communications Monitoring: follow official updates on the MAI models roadmap.
• Internal AI Governance: Update the AI tool usage policies to include criteria for evaluating the underlying model.

For companies using Copilot in the workplace LinkedIn campaign o Google Ads campaigns, . The quality of text generation is a factor directly linked to the performance of creatives. Therefore, active monitoring is advisable from the outset.

Perspectives: Towards a More Fragmented AI Ecosystem

This Microsoft transition is not an isolated case. Google has also developed its own Gemini models to reduce its reliance on third-party providers. Thus, the enterprise AI market is evolving towards a model where large players prefer vertical integration to open partnerships.

For Italian SMEs, this scenario has direct implications. First, the choice of AI tools can no longer be based solely on the supplier's brand, but must consider the underlying model and its suitability for specific use cases. Furthermore, reliance on a single ecosystem—like Microsoft 365 with Copilot—becomes a strategic risk if model performance is not contractually guaranteed.

Finally, the need for in-house expertise to assess the quality of AI outputs emerges. Companies that invest in AI-assisted copywriting in SEO strategy powered by language models must be able to distinguish quality output from mediocre output, regardless of the tool used. To delve deeper into these topics, you can refer to the SHM Studio Blog o Contact our team for a dedicated consultation.