Microsoft will transition its enterprise AI agent, Copilot Cowork, to a usage-based pricing model when the tool reaches broader availability in June 2026. The move, confirmed by sources familiar with the planning, marks a significant shift from the per-user subscription approach that has defined Microsoft 365 AI pricing so far. At the same time, the company is reportedly evaluating a Microsoft-hosted version of DeepSeek’s large language models, potentially offering enterprises a lower-cost alternative for AI workloads inside Azure and Microsoft 365.

These two developments signal a pivotal moment for IT leaders. As AI usage explodes across the enterprise, Microsoft is seeking to align costs with actual consumption while expanding its model catalog with competitive, open-weight options. The combination could reshape budgeting, governance, and vendor strategy for organizations already deeply invested in the Microsoft ecosystem.

Copilot Cowork Moves Beyond the Per-Seat Model

Copilot Cowork was first introduced in limited preview as a specialized agent integrated into Microsoft 365 applications. Unlike the personal Copilot that summarizes emails or drafts documents, Cowork is designed for team-level collaboration—it can attend meetings, generate action items, track decisions, and even perform multi-step workflows across Teams, SharePoint, and Planner. Its enterprise orientation meant pricing was always a complex question. During the initial rollout, Microsoft charged a flat per-user monthly fee, similar to the standalone Copilot for Microsoft 365.

That model is now being retired. Starting with general availability in June 2026, Cowork will be priced according to \"AI units\" consumed—a combination of tokens processed, API calls made, and task completions executed. Though Microsoft has not yet published the exact rate card, early indications suggest the ticketing will follow a pattern similar to Azure OpenAI Service, where different models and operations carry distinct unit costs. The change is designed to give larger customers more flexibility, but it also introduces a variable cost structure that could surprise teams without proper controls.

For IT departments, the shift means abandoning predictable per-user budgeting. A heavy user of Cowork—say, a project manager who runs multiple daily standup summaries and action trackers—could cost significantly more than a light user. Conversely, an organization that rolled out Cowork per-seat to hundreds of employees might see overall spending drop if utilization is low. The unpredictability is the challenge.

Why Microsoft Is Making the Switch Now

Usage-based pricing is not an outlier in enterprise AI. OpenAI’s API, Google’s Vertex AI, and even Microsoft’s own Azure Cognitive Services all charge by consumption. By extending this model to Copilot Cowork, Microsoft is synchronizing its billing with the way large language models actually consume resources. Each Cowork session requires GPU-backed inferencing, and sustained, complex tasks can cost dollars per execution.

There is also a competitive angle. Google Workspace has been aggressively baking AI features into existing tiers without additional per-user surcharges, betting that higher retention and upsell will recoup the investment. Microsoft, by contrast, has layered Copilot as a premium add-on. Usage-based pricing could allow Microsoft to offer a lower entry point—perhaps even a small free tier—while charging power users more, making the service more accessible to smaller businesses and departmental teams who balk at the $30-per-user monthly fee.

The timing of June 2026 is also strategic. By then, many enterprises will have completed Copilot pilots and will be ready to scale. A usage-based model eases adoption: customers can start small and ramp up organically. Microsoft can also collect granular data on exactly which agent capabilities drive the most value, informing future development.

The Governance Gap: What IT Must Prepare For

Moving to meter billing places an immediate burden on IT and procurement teams. They will need to implement real-time monitoring of AI consumption, set spending caps, and create chargeback mechanisms for departments. Without these guardrails, a single runaway agent—perhaps one stuck in a loop re-generating meeting notes—could rack up thousands of dollars in a weekend.

Microsoft is expected to deliver enhanced admin controls alongside the pricing change. The Microsoft 365 admin center will likely gain dashboards showing usage by user, department, and task type. Integration with Cost Management in Azure is probable, allowing organizations to set budget alerts and enforce quotas. However, the native tools are rarely sufficient for complex enterprises. Third-party SaaS management platforms such as Zylo, Productiv, and even ServiceNow will need to add Copilot Cowork spend tracking to their portfolio.

Beyond cost, governance must address data privacy and compliance. Cowork processes sensitive enterprise content—emails, chat messages, documents. Under usage billing, every transaction becomes an auditable event. Logs of prompts and responses will need to be retained for security and e-discovery. IT must verify that the data does not leave the tenant boundary or get used for model training without consent. Microsoft’s Data Protection Addendum already covers Copilot, but the granular visibility of usage billing might expose new compliance requirements in regulated industries like finance and healthcare.

Enter DeepSeek: A Low-Cost Alternative in Microsoft’s Own Cloud

While reworking Cowork’s commercial model, Microsoft is simultaneously evaluating DeepSeek’s AI models for its enterprise platform. The Chinese AI startup made headlines with an open-source model that rivals GPT-4-level performance at dramatically lower training and inference costs. According to reports, Microsoft is exploring hosting DeepSeek models within Azure and making them available through the same APIs that power Copilot features.

If realized, this would give enterprise customers a powerful new option. DeepSeek’s models could be offered as a cost-effective backend for certain Cowork tasks—summarization, classification, basic Q&A—where the full power of GPT-5 is overkill. The combination could lower the per-usage rate for common operations while preserving the flagship model for high-value reasoning. It also hedges Microsoft’s dependence on OpenAI, which remains a key partner but is no longer exclusive.

However, the DeepSeek option raises significant geopolitical and regulatory questions. DeepSeek is a Chinese company, and many enterprises are prohibited from sending data to Chinese-controlled servers. If Microsoft hosts the models entirely within its own sovereign regions—Azure Government, EU data boundary, etc.—those concerns might be mitigated. But IT leaders will need to scrutinize the data flow carefully. Additionally, model access controls in Microsoft 365 would have to ensure that sensitive data never gets routed to a non-compliant model backend by mistake.

For IT, the prospect of multi-model AI within the same familiar Microsoft interface is appealing. Users could enjoy the same Copilot experiences while the system intelligently selects the most cost-effective model for each task, transparent to the end user. But the complexity of managing permissions, data residency, and model-specific performance characteristics will require new skill sets in the IT organization.

Preparing Your Enterprise: Actionable Steps for 2025-2026

With over a year until the pricing change takes effect, IT departments have time to prepare. The following steps should be on every CIO’s roadmap:

  • Audit current Copilot Cowork usage. If already in preview, start logging the number of interactions per user and the types of tasks performed. This data will be crucial for modeling future costs under usage billing.
  • Engage Microsoft on the upcoming rate card. Ask for indicative pricing and any commitment discounts. Early adopters may negotiate preferential terms if they help shape the product.
  • Implement a FinOps practice for AI. AI spend must be tracked like cloud compute. Establish a team responsible for monitoring, optimizing, and governing all AI consumption, starting with Copilot Cowork.
  • Develop internal showback or chargeback models. Without per-seat pricing, cost allocation becomes murky. Design a mechanism to attribute Cowork spend to business units, projects, or cost centers.
  • Evaluate model choice policies. If DeepSeek or other third-party models become available, define which workloads are approved for which model. Security, accuracy, and cost thresholds should guide that policy.
  • Update data governance frameworks. With model routing across potentially different providers, existing DLP (Data Loss Prevention) and Insider Risk policies may need revision. Map data flows for all AI interactions.
  • Train the helpdesk and end users. Users need to understand that AI is not free. Encourage efficient prompting and discourage leaving long-running agent tasks unattended.

The Bigger Picture: AI Is Becoming an Infrastructure Expense

The Copilot Cowork pricing shift is part of a broader trend where AI transitions from a software license to a utility-like service. Just as virtualization moved IT from buying servers to paying for VM-hours, AI is moving from fixed-price tools to metered intelligence. Gartner predicts that 70% of enterprise AI applications will be usage-based by 2027. Microsoft’s move is both a validation and an accelerator of that trend.

For the Windows and Microsoft 365 community, the developments reinforce the importance of staying aligned with Microsoft’s Azure ecosystem. The tighter the integration between Copilot Cowork’s metering and Azure billing—perhaps through a single MACC (Microsoft Azure Consumption Commitment)—the more strategic a customer’s cloud commitment becomes. Organizations already on an Azure Enterprise Agreement may find it simpler to absorb AI costs, while those with rigid per-user budgeting will face disruption.

Uncertainty and Opportunity

The dual-track strategy—usage pricing for Cowork, model diversification with DeepSeek—carries both risk and reward. Microsoft gains the flexibility to compete on price and performance while protecting itself from over-reliance on a single model provider. Enterprises gain choice and the potential for lower total cost of ownership. But realizing those benefits demands active governance, informed procurement, and a willingness to treat AI as a critical infrastructure layer rather than a one-off software purchase.

In the coming months, Microsoft is expected to release detailed technical and commercial documentation. IT leaders should watch for Ignite 2025 sessions and early adopter programs that offer granular insights. The decisions made in the next year will set the foundation for how organizations harness agentic AI for the rest of the decade.

Ultimately, the message is clear: the era of flat-rate AI is ending. Those who adapt their management practices now will turn June 2026 into an opportunity rather than a surprise invoice.