Microsoft Shifts Focus to In-House AI Development for Cost Savings and Competitive Edge

Deep News
Yesterday

Recent reports indicate that Microsoft is pursuing a strategy of self-reliance alongside a major restructuring of its Copilot AI assistant, aiming to cut costs and rebuild market competitiveness by replacing third-party AI models with its own technology in key products.

According to a Tuesday media report, Microsoft has begun using its internally developed MAI model within core office applications like Excel and Outlook, replacing models from OpenAI and Anthropic. The MAI model is now handling tens of thousands of AI prompts weekly for these apps.

This follows a separate report from earlier in the week, which revealed an internal memo from Microsoft executive vice president Jacob Andreou. The memo announced the consolidation of consumer and enterprise Copilot apps, the removal of underperforming features, and a goal to "earn the right to exist" in the eyes of customers.

These moves highlight the dual pressures facing the company. Reliance on external AI models comes with significant procurement costs, while Copilot faces ongoing market pressure. Microsoft's stock has fallen nearly 20% year-to-date, underperforming its Magnificent Seven peers, leading some major shareholders to reduce their holdings.

In-House Model Rollout Aims for Cost Reduction

A recent report cited an anonymous source stating that Excel and Outlook now process tens of thousands of AI prompts weekly with the MAI model, a task previously more reliant on models from OpenAI and Anthropic. A Microsoft spokesperson declined to comment on the report.

The scale of MAI's deployment within Microsoft Office products had not been previously disclosed. While still representing a small portion of overall AI usage, this progress indicates tangible breakthroughs in Microsoft's lower-cost, in-house AI development path.

Microsoft's AI model business lead, Mustafa Suleiman, stated publicly in June that the company "pays a lot of money to Anthropic" with the goal of reducing and eventually eliminating that cost. At a June developer conference, the company released seven new AI models, one of which was claimed to offer code generation capabilities comparable to Anthropic's popular Opus 4.6 model but at a lower cost.

Microsoft consumes vast amounts of AI compute tokens. While a long-term agreement with OpenAI currently provides discounted technical support, this arrangement is not permanent. Suleiman's team is working to build in-house capabilities to avoid future dependency on external pricing.

Beyond Excel and Outlook, the MAI model is already available for the GitHub Copilot code assistance service. Suleiman also noted that Microsoft's in-house speech transcription model will be deployed to the Teams video conferencing app and other products in the coming months.

Copilot Consolidation: Trimming Features and Betting on a "Super App"

According to a recent report, Jacob Andreou's internal memo announced plans to merge consumer and enterprise Copilot into a single, unified application. This consolidated app will also integrate AI coding tools and a new generation of AI agents called AutoPilot. These agents are designed to be "always-on," handling routine tasks for users. One agent, named Scout, can manage schedules and automatically summarize inbox emails.

The restructuring also involves removing underutilized features. Sources indicated that Copilot Podcasts, which could generate podcasts from user content, and the experimental developer platform Copilot Labs will be shut down due to lack of user traction. Andreou had also previously ordered the removal of some Copilot entry buttons from Windows apps, as data showed most user clicks were accidental and followed by immediate closure.

The new, consolidated Copilot is scheduled for launch in August. Beyond basic features, advanced capabilities like AI coding tools, AutoPilot agents, and Copilot Cowork will be offered as paid add-ons. Microsoft also announced the formation of a $2.5 billion AI consulting unit, Microsoft Frontier Company, which will deploy 6,000 industry and engineering experts to help customers implement AI systems.

Copilot User Growth Accelerates but Gap with Rivals Remains

Copilot's commercial progress has shown recent improvement. Microsoft reported 15 million paying users in January, a figure that grew 33% to 20 million by April. The product has a starting price of $30 per employee per month. Management stated that Copilot has helped improve the revenue growth rate of the Office business in recent quarters.

However, a significant gap remains compared to key competitors. ChatGPT boasts over 50 million paying subscribers. In terms of monthly active users, estimates suggest Copilot has around 38.5 million, while ChatGPT has approximately 1 billion.

In his memo, Andreou acknowledged that Copilot's over-expansion of features had caused user confusion and "become an embarrassing problem internally." He noted that enterprise customers are increasingly scrutinizing AI investment returns, and "the bar is rising across the enterprise software landscape." Andreou emphasized that Copilot should focus on "real work" and be "results-oriented," not "intelligent for the sake of being intelligent."

Andreou joined Microsoft in the spring of last year and was promoted by CEO Satya Nadella earlier this year to lead the Copilot overhaul. His responsibilities now span both consumer and enterprise product lines, as well as Bing, MSN, and the Edge browser business, overseeing a team of more than 11,000 employees. His performance metrics from Nadella include overall Copilot growth, user retention, and customer satisfaction.

Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.

Most Discussed

  1. 1
     
     
     
     
  2. 2
     
     
     
     
  3. 3
     
     
     
     
  4. 4
     
     
     
     
  5. 5
     
     
     
     
  6. 6
     
     
     
     
  7. 7
     
     
     
     
  8. 8
     
     
     
     
  9. 9
     
     
     
     
  10. 10