Microsoft Shifts to In-House AI Models to Reduce Rising AI Costs

As the cost of developing and operating artificial intelligence continues to rise, technology companies are seeking ways to reduce costs. Microsoft appears to be the latest company to adopt that approach, with reports indicating it has begun relying less on AI software from OpenAI and Anthropic while expanding its use of internally developed models.

According to a Bloomberg report published on Tuesday, Microsoft has begun using its proprietary MAI models to handle a portion of user requests in two of its most widely used productivity applications, Word and Excel. Previously, the company had highlighted that significant parts of Microsoft 365’s AI-powered features were driven by models supplied by both OpenAI and Anthropic.

Although Microsoft continues to utilise third-party AI models where appropriate, it has been steadily increasing investment in its own AI capabilities and agents. During its annual Build conference last month, the company introduced seven new MAI models, including an AI coding agent designed to assist developers and a text-to-image generation model.

When TechCrunch contacted Microsoft for comment regarding the reported changes, the company declined to provide any additional information beyond its existing announcements.

Microsoft’s reported strategy reflects a broader shift taking place across the technology industry. Following an aggressive period of AI expansion and heavy investment earlier this year, many major technology companies have placed greater emphasis on cost control in recent months. Reports suggest that organisations including Amazon, Uber, Meta, and Accenture have also taken steps to reduce spending while continuing to invest in artificial intelligence.

The rising cost of building, operating, and purchasing AI services has become one of the industry’s most debated challenges. In parts of Silicon Valley, concerns over soaring costs have reportedly become significant enough that some companies are exploring lower-cost AI models developed in China as an alternative to agent-based solutions, despite ongoing questions about potential security and privacy risks.