Microsoft is shaking up its AI pricing and tech mix. The company is ditching flat-rate fees for Copilot Cowork, its enterprise AI assistant, and shifting to usage-based billing. Now, customers pay for the compute they actually use, not a fixed monthly charge.

This change comes because agentic AI—tools that juggle multiple tasks autonomously—are expensive to run. Some users run hundreds of tasks weekly, driving compute costs through the roof. Microsoft can no longer swallow those bills silently.

To tame expenses, Microsoft is testing a self-hosted, fine-tuned version of DeepSeek V4, a Chinese open-source AI model. It’s cheaper to run than the current Anthropic and OpenAI models powering Copilot Cowork. The plan is to host DeepSeek on Azure, so data stays inside Microsoft’s secure cloud, sidestepping Chinese servers.

Choosing a Chinese AI in today’s tense geopolitical climate is bold. The U.S. government is cracking down on Chinese AI firms and restricting access to foreign models. Microsoft insists DeepSeek would be optional and tightly controlled, with bias-reducing tweaks and compliance safeguards.

This move reflects a broader strategy to diversify AI suppliers. Microsoft no longer wants to rely solely on a handful of providers. By mixing models, it gains pricing leverage and flexibility. If DeepSeek proves viable, it could lower costs without degrading user experience.

Microsoft CEO Satya Nadella admitted internally the company is “addicted” to token consumption—the units that track AI usage—illustrating how AI workloads quickly balloon compute expenses. The shift to pay-per-use pricing acknowledges this reality and forces heavy users to shoulder their fair share.

Enterprises stand to benefit from this pricing model, gaining more control and transparency over AI costs. But consumption-based billing adds uncertainty, making cost management crucial. Companies handling sensitive data also face tough questions about security and compliance, even with Azure hosting.

The DeepSeek experiment highlights how AI models are commoditizing. Microsoft invested billions in OpenAI, yet it’s open to swapping in a cheaper, open-source alternative from China. This signals that AI’s value increasingly lies in how tools are integrated and managed, not just the underlying model.

Microsoft plans to announce its new Copilot Cowork pricing tier and AI model choice soon. This marks a significant pivot for enterprise AI economics. As agentic AI grows more capable and costly, price structures and supplier strategies must adapt or risk breaking budgets.