Microsoft has abandoned flat-rate pricing for its Copilot Cowork agentic AI tool, moving to usage-based billing through Copilot Credits. The company launched the consumption-based model globally on June 16, 2026, signaling that the economics of autonomous AI workflows no longer fit traditional monthly subscription packages. Microsoft is also exploring a fine-tuned version of DeepSeek V4, a Chinese open-source AI model, as a lower-cost alternative to premium models from Anthropic and OpenAI.

Why Did Microsoft Switch Away From Flat-Rate Pricing?

Copilot Cowork is designed to handle complex, multi-step tasks across Microsoft 365 applications like Outlook, Teams, and Excel. Unlike a standard chatbot that answers a single question, Cowork plans workflows, retrieves data, calls multiple tools, and delivers completed results with minimal human intervention. This autonomous behavior comes with a hidden cost: each task burns through far more AI model tokens than a typical user prompt.

The financial pressure is real. According to EY's June 2026 analysis, the cost per AI interaction in enterprise settings jumped from $0.04 in 2023 to $1.20 in 2026, roughly a 30-fold increase driven by agentic workflows. Charles Lamanna, Microsoft's executive vice president for Copilot, explained the problem directly: "We have users who do hundreds of tasks a week, which is great, they're way productive, but the consequence is the costs can go very high," he stated.

Charles Lamanna, Microsoft's executive vice president for Copilot

"We have users who do hundreds of tasks a week, which is great, they're way productive, but the consequence is the costs can go very high," stated Charles Lamanna, executive vice president for Copilot at Microsoft.

Charles Lamanna, Executive Vice President for Copilot, Microsoft

Under the new pricing structure, Copilot Cowork sits on top of a $30 per user per month Microsoft 365 Copilot license. Beyond that base subscription, companies pay for actual compute consumed through Copilot Credits. The cost of each task depends on four key factors:

• Model Used: Premium models like Anthropic's Opus 4.8 cost more per token than standard alternatives
• Context Retrieved: The amount of data the agent pulls in to complete a task affects token consumption
• Tool Calls: The number of external applications and actions the agent invokes increases costs
• Runtime: How long the task takes to complete determines total compute time billed

Microsoft offers two payment options: Pay As You Go for flexible spending, and a prepaid P3 plan for organizations that can commit to a usage volume upfront in exchange for a discount.

How to Manage AI Spending in Your Organization?

For IT teams adopting Copilot Cowork, the shift from flat-rate to usage-based billing requires new governance practices. Microsoft has built financial operations controls into the Copilot admin center to help organizations monitor and manage spending. Here are the key steps organizations should take:

• Set Usage Budgets: Define spending limits per department or team, similar to how cloud compute budgets are managed in Azure or AWS
• Monitor Token Consumption: Track which workflows consume the most tokens and identify opportunities to optimize task design or model selection
• Test Model Options: Evaluate whether lower-cost models like DeepSeek V4 can handle routine workflows without sacrificing quality
• Establish Governance Policies: Create approval workflows for high-cost tasks and educate users on cost-conscious AI usage patterns

Is This Pricing Shift Happening Across the Industry?

Microsoft's move is not isolated. GitHub Copilot already switched to token-based billing on June 1, 2026, replacing its previous fixed Premium Request Unit system with AI Credits priced at $0.01 each. The pattern is clear: the entire AI industry is shifting the cost of agentic AI from vendor balance sheets onto customer budgets.

The underlying problem is universal. Goldman Sachs projects that agentic workflows could drive token consumption up by 24 times or more compared to standard chatbot interactions. Uber's chief technology officer publicly revealed that the company burned through its entire 2026 AI budget within the first few months of the year. OpenAI CEO Sam Altman acknowledged in June 2026 that questions about AI spending returns are "the most fair criticism right now of AI".

Sam Altman

These signals reflect not a technology failure, but an economic model that has not yet caught up with capability. Usage-based billing transfers risk from vendors to customers while providing granular cost control. For organizations already using or evaluating Microsoft 365 Copilot, the Cowork launch introduces both opportunity and complexity: access to powerful agentic automation balanced against the need for active cost management and governance.

What About the DeepSeek Option and Its Implications?

The most controversial aspect of Microsoft's announcement is its exploration of DeepSeek V4, a Chinese open-source AI model, as a lower-cost option. DeepSeek has gained significant traction among developers for strong performance at a fraction of the cost of proprietary alternatives. Microsoft has already fine-tuned a version of the model and added safeguards aimed at reducing bias, with a final deployment decision expected in the coming weeks.

If Microsoft proceeds, the company says DeepSeek would be optional for enterprise customers, not a default; fully hosted on Azure to keep data within Microsoft's cloud; and protected by Azure's full stack of compliance, security, and data sovereignty safeguards. The strategic logic is straightforward: agentic AI needs cheaper reasoning. Not every workflow requires a top-tier model, and forcing customers onto expensive models for routine tasks drives up costs without corresponding gains in output quality.

However, integrating a Chinese-developed AI model into a productivity suite relied on by over half of the Fortune 500 carries political risk and will likely attract scrutiny from regulators, security teams, and policymakers who view AI supply chains as matters of national technology competition.