1 hour ago
3
Microsoft is overhauling how it charges for Copilot Cowork, its autonomous AI agent baked into Microsoft 365, moving away from the flat per-user fee structure toward a consumption-based model built around what the company calls Copilot Credits. At the same time, Microsoft is reportedly weighing the introduction of a Microsoft-hosted version of DeepSeek, the Chinese AI lab’s model, as a cheaper alternative to the Anthropic Claude models currently powering the tool.
From flat fees to pay-as-you-go
Copilot Cowork launched in March 2026 as a bundled feature for users already paying for the standard Microsoft 365 Copilot add-on, which runs around $30 per user per month. At launch, there was no extra charge. The tool was designed to handle complex, context-aware tasks across corporate data sources like SharePoint, OneDrive, and Outlook, essentially acting as an autonomous agent that could chain together multi-step workflows without constant human oversight.
The credit-based system is designed to integrate with existing consumption metrics already used in products like Copilot Studio and the new Work IQ API services. For IT departments managing sprawling deployments, this creates a more unified way to track and budget AI spending across Microsoft’s ecosystem.
Long-running agentic workflows, the kind Copilot Cowork specializes in, consume unpredictable amounts of compute. A flat per-user price doesn’t reflect the actual cost of running an autonomous agent that might spend hours chewing through a company’s document library versus one that handles a quick email summary.
DeepSeek enters the enterprise conversation
Microsoft’s apparent interest is in bringing a version of DeepSeek, potentially branded DeepSeek V4, into the Copilot Cowork stack as a lower-cost alternative to the Anthropic Claude models currently running the show. DeepSeek’s models have been available through Azure AI Foundry since January 2025, offered in both serverless and provisioned throughput configurations for developers.
The rollout could come within weeks of the June 16, 2026 announcement, though Microsoft hasn’t publicly confirmed exact timing. If it materializes, it would make Microsoft one of the first major enterprise software companies to embed a Chinese-developed AI model as a primary engine in a mainstream productivity suite.
What this means for the market
For enterprises, the upside is lower barriers to initial deployment. Companies that hesitated to roll out Copilot Cowork to thousands of employees at a fixed cost can now start small and scale based on actual value delivered. The downside is cost unpredictability, as usage-based models can spiral quickly if guardrails aren’t in place.
The DeepSeek angle introduces data sovereignty and security concerns around Chinese AI models, particularly for government contractors, financial institutions, and healthcare organizations operating under strict regulatory frameworks. Microsoft hosting the model on its own infrastructure mitigates some of those concerns, since the data wouldn’t flow to Chinese servers.
The model diversification strategy, running Anthropic Claude alongside a potential DeepSeek option, gives Microsoft negotiating leverage so that no single model provider becomes indispensable.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
English (US) · 