Microsoft is set to transform its collaborative AI tool, Copilot Cowork, with a new usage-based pricing model starting June 2026, according to internal documents seen by Windows News. In parallel, the company is quietly testing a Microsoft-hosted, fine-tuned version of DeepSeek V4 — or a similar open-source model — as a cost-effective alternative backend for enterprise customers. The dual move signals a strategic shift aimed at offering more flexible cost structures while hedging against rising AI model costs.

Copilot Cowork, an AI-powered collaboration tool deeply integrated into Microsoft 365 and Teams, has until now been sold as a per-user add-on subscription. Designed to streamline meetings, co-author documents, and automate routine team tasks, it quickly became a flagship example of Microsoft’s generative AI ambitions. But the flat monthly fee per seat has drawn criticism from organizations with uneven usage patterns, where some months see heavy Copilot activity and others little to none.

The upcoming switch to usage-based pricing — often called consumption pricing — will charge enterprises based on the actual AI interactions, such as queries answered, documents generated, or meeting minutes processed. While exact rates remain undisclosed, the model mirrors what cloud providers like AWS and Google have long done for compute services. A source familiar with the plans said Microsoft aims to provide granular metering, possibly down to per-token pricing for text generation and per-minute pricing for meeting transcription features.

“The idea is to align cost with value,” the source explained. “If a team uses Copilot Cowork heavily during a product launch but less during quieter quarters, they shouldn’t pay a flat fee all year.” Early indications suggest that existing enterprise agreements may grandfathered into a hybrid model for a transition period, with new contracts moving entirely to consumption billing after June 2026.

This pricing pivot comes as Microsoft faces pressure from competitors like Google’s Duet AI and a growing ecosystem of open-weight models that are drastically undercutting proprietary AI services. By decoupling the tool from a per-user license, Microsoft hopes to attract cost-conscious mid-market customers and startups that previously balked at the $30 per user monthly price tag for full Copilot features.

Enter DeepSeek V4: A Budget Backend Bet

Perhaps more surprising is Microsoft’s quiet experiment with DeepSeek V4, a model developed by the Chinese AI lab known for achieving state-of-the-art results at a fraction of the cost of models like GPT-4. According to two people briefed on the project, Microsoft engineers are hosting a fine-tuned version of DeepSeek V4 specifically optimized for enterprise collaboration tasks — summarization, action-item extraction, and email thread decoding — and offering it as an optional, lower-cost backend choice for Copilot Cowork.

The fine-tuned model runs entirely inside Microsoft’s Azure infrastructure, addressing data sovereignty concerns that might arise from using a third-party model. Customers who opt for the DeepSeek backend would see lower per-interaction charges compared to the default OpenAI-based models that currently power Copilot. The company is also testing integration with other open models, but DeepSeek V4 emerged as the leading candidate due to its superior performance on structured business reasoning benchmarks.

“We are exploring multiple paths to bring affordable AI to every organization,” a Microsoft spokesperson said in a written statement, neither confirming nor denying the DeepSeek trial. “Our priority remains delivering enterprise-grade security, compliance, and choice.”

Why Usage-Based, Why Now?

The shift reflects a maturing market where AI is no longer a luxury add-on but a utility. According to a 2025 Gartner survey, 60% of large enterprises were renegotiating AI software contracts to include consumption elements, citing the 2024–2025 wave of “AI shelfware” — expensive tools that teams paid for but rarely used. By moving to usage-based pricing, Microsoft can also better compete with the rising tide of “bring your own model” architectures, where companies run open-source models themselves on cloud GPUs and avoid per-seat charges altogether.

For Microsoft, there’s also a defensive angle. The company has invested heavily in Azure AI infrastructure, and a consumption model for Copilot Cowork turns that fixed cost into a variable revenue stream tied directly to customer usage. It also paves the way for dynamic pricing based on model size — using a cheaper model like DeepSeek V4 for simple tasks and only spinning up a larger model for complex ones.

What It Means for Enterprise Customers

Reaction among early testers has been mixed. IT managers who spoke to Windows News on condition of anonymity appreciated the added flexibility but worried about budget predictability. “We can easily forecast per-user subscriptions, but consumption costs can spiral if a department suddenly ramps up AI use without guardrails,” said one IT director at a midsized insurance firm.

To address this, Microsoft is reportedly building granular cost controls and alerting into the Microsoft 365 Admin Center. Admins will be able to set monthly spending caps per user, group, or tenant, and view detailed breakdowns of which Copilot actions are driving charges. A public preview of these controls is expected in early 2026.

Another point of friction is the model choice. While the DeepSeek option promises lower costs, it may exhibit different behavior than the primary OpenAI model. Microsoft has assured that both backends will undergo rigorous red-teaming and quality evaluation, but some enterprises, particularly those in regulated industries, may be wary of using a model whose training data and guardrails are less transparent.

“We need consistency,” said the CTO of a financial services firm piloting the new pricing. “If the cheaper model suddenly starts hallucinating in board meeting summaries, the savings won’t be worth it.” Microsoft is considering a model-selection toggle within the Copilot interface, allowing teams to pick the backend per task — a “smart routing” feature that could automatically default to the cheaper model unless a higher level of sophistication is required.

The Open-Source Threat and Microsoft’s Answer

The DeepSeek move also acknowledges the growing viability of open-source models in the enterprise. DeepSeek V4, released in late 2025, stunned the industry by matching or beating proprietary models on numerous benchmarks while costing an estimated 80% less to run. By hosting and fine-tuning it itself, Microsoft can offer a “wrapper” service with enterprise support while maintaining the lock-in to Azure — much like its successful strategy with the OpenAI service.

This “model garden” approach could also insulate Microsoft from potential disputes with OpenAI over pricing or exclusivity. Recent reports have suggested tensions between the two companies over commercialization of new models, and Microsoft’s move to diversify its backend portfolios might be interpreted as a subtle hedge.

Competitive Landscape Shifts

The AI assistant market is hurtling toward consumption pricing. GitHub Copilot, also owned by Microsoft, introduced per-request pricing for certain API-style integrations in 2025. Google’s Duet AI for Workspace already offers a mix of per-user and consumption plans. And a slew of startups like Vercel’s v0 and Anthropic’s Claude for Work have undercut incumbents with purely usage-based billing.

By tying Copilot Cowork directly to Microsoft 365’s collaboration suite, Microsoft has a distribution advantage none can match. The transition to usage-based pricing removes the last barrier — cost uncertainty — for the hundreds of millions of knowledge workers already using Teams, Outlook, and Office daily.

The Road Ahead

With the June 2026 deadline, Microsoft has less than a year to finalize rates, build the metering infrastructure, and complete its DeepSeek trial. Industry analysts expect the company to announce full details at its Build conference in May 2026, with early adopter programs opening in the spring.

For enterprises, now is the time to start auditing current Copilot Cowork usage to model costs under the new regime. Microsoft’s own Copilot Dashboard already provides some metrics, and third-party tools are emerging to help map consumption patterns. The deeper question — whether to trust a cheaper, externally sourced model with sensitive boardroom discussions — will take longer to answer.

One thing is clear: the days of one-size-fits-all AI pricing are ending. Microsoft’s dual-track strategy of flexible billing and model choice positions Copilot Cowork as both a premium and affordable tool, potentially expanding its addressable market just as the AI arms race enters its next phase.