Microsoft is making a significant adjustment to its AI strategy for its enterprise agent, Copilot Cowork.
The company is now testing a lower-cost, open-source AI model called DeepSeek as an optional engine for Copilot and is moving away from a flat-rate subscription to a usage-based pricing model. This change might seem technical, but it's a deliberate move to tackle two critical challenges in the AI industry today: unpredictable costs and supply chain risks.
So, why is this happening now? There are a few key drivers. First, the regulatory landscape has suddenly become more uncertain. Just last week, a major AI company, Anthropic, was forced to shut down its newest models due to a U.S. government order. This event sent a clear signal to the market: relying on a single, powerful AI model is risky. If a top-tier model can be turned off overnight, large enterprise customers need reliable alternatives. Microsoft's move to offer a cheaper, Azure-hosted open-source model provides exactly that—a hedge against regulatory and supply chain disruptions.
Second, the way people use AI is changing. Microsoft's recent developer conference heavily promoted 'agentic AI'—AI assistants that can perform complex, multi-step tasks on their own. While incredibly powerful, these agents can also call on AI models repeatedly, causing computing costs to spike unpredictably. A fixed subscription fee just isn't sustainable in this environment. By switching to usage-based pricing and offering a cheaper model, Microsoft can stabilize its costs (COGS) and prevent customers from getting 'bill shock.'
Finally, this isn't a completely new direction for Microsoft. The company had already integrated DeepSeek into its Azure AI platform earlier, establishing a compliance and safety framework for it. This existing groundwork makes it easier to now offer it as an optional, securely-hosted engine within Copilot. This strategy is less about replacing high-end models like those from OpenAI or Anthropic and more about adding a practical, cost-effective tool to its multi-model toolbox. It's a pragmatic response to the evolving economic and regulatory realities of enterprise AI.
- Glossary
- Agentic AI: AI systems that can autonomously plan and execute multi-step tasks to achieve a goal, rather than just responding to a single prompt.
- Usage-based pricing: A billing model where customers pay based on how much of a service they consume, rather than a flat monthly or annual fee.
- COGS (Cost of Goods Sold): The direct costs attributable to the production of the goods or services sold by a company. For AI, this is primarily the cost of computing power (inference).