Microsoft Corporation has faced an unprecedented surge in expenses related to the use of advanced third-party artificial intelligence models, such as Claude from Anthropic and ChatGPT from OpenAI, forcing its leadership to fundamentally rethink its internal strategy. According to data from the authoritative publication Bloomberg, tens of thousands of AI requests that weekly originate from popular enterprise applications like Excel and Outlook are now being processed exclusively by Microsoft's own proprietary MAI model lineup. Previously, these key productivity tools relied heavily on the computational power and algorithms of OpenAI and Anthropic to solve a wide range of tasks. Nevertheless, this transition to internal capabilities currently accounts for only a minor fraction of the tech giant's total AI token consumption, as the flagship Copilot work assistant still requires colossal resources for its stable operation across the ecosystem.

The financial aspect of this issue has reached critical proportions, as openly stated by the head of Microsoft AI, Mustafa Suleyman, who noted that numerous employees within the company are literally spending millions of dollars on AI tokens. As a preventive measure, Microsoft introduced seven new internal developments last month, including the first reasoning model MAI-Thinking-1, as well as specialized neural networks for working with images, voice, and code. The price difference in the market is truly striking: if the Chinese company DeepSeek offers its advanced models at highly democratic prices (about $0.43 per million input tokens), Anthropic charges up to $10 for input and a full $50 per million output tokens for its flagship Fable 5 model. OpenAI is also not a cheap option, although Microsoft does receive partnership discounts until the agreement expires in 2032.

Such a strategic pivot by Microsoft clearly demonstrates a global industry trend in which more affordable and energy-efficient AI solutions are rapidly capturing the market and attracting the attention of major players. The resounding success of the Chinese company DeepSeek earlier this year, which released a lineup of budget-friendly yet powerful models, clearly showed that the industry urgently needs alternatives. Suleyman directly emphasized that Anthropic's services have become extremely expensive, and the company has set an ambitious goal to completely eliminate these costs. Against the backdrop of modern AI agents consuming hundreds of times more energy and computing resources per request compared to regular chatbots, developing proprietary cost-effective models is becoming not just a matter of saving money, but a crucial issue of technological sovereignty and long-term financial sustainability for the world's leading technology corporations.