Microsoft официально запущен Copilot Cowork for general availability worldwide, and in the same breath, the tech giant revealed it is actively exploring the use of a fine-tuned version of ДипСик V4 to power a lower-cost tier of the enterprise AI agent.
The development, first reported by Axios on Tuesday, signals a significant strategic shift in how Microsoft is thinking about the economics of always-on AI in the enterprise — and raises fresh questions about model diversity, cost control, and the role of Chinese-developed AI inside sensitive corporate environments.
Soaring Token Costs Are the Root of the Problem
The driving force behind this exploration is straightforward: enterprise usage of Copilot Cowork is expensive. Чарльз Ламанна, Microsoft’s Executive Vice President for Copilot, Agents, and Platform, told Axios that some enterprise users are running hundreds of tasks per week through Cowork, sending costs soaring to levels he described as "очень высоко."
Unlike a conventional chatbot that processes a single prompt and returns a single response, Cowork operates fundamentally differently — it plans multi-step workflows, retrieves contextual information, calls external tools, and iterates in loops until a task is fully completed.
Every single one of those loops consumes model tokens and compute resources, making the cumulative cost substantially higher than standard AI interactions.
At its general availability launch, Copilot Cowork runs primarily on Anthropic models including Opus 4.8 and Sonnet 4.6, где GPT 5.5 от OpenAI available in a separate Frontier preview tier.
Microsoft has not yet formally committed to DeepSeek V4 as its cheaper engine, but the company confirmed that a lower-cost model option would be announced and made available в течение нескольких недель. Alongside the DeepSeek option, Microsoft is also developing its own proprietary “Cowork 1” branded model — described as a secure, in-house solution built for everyday tasks at substantially reduced cost.
Azure Hosting as the Security Shield
Microsoft is acutely aware that the prospect of a Chinese-developed AI model sitting inside an enterprise productivity suite will raise eyebrows — particularly among regulated industries, government contractors, and security-conscious organizations.
To address those concerns head-on, the company has confirmed that any DeepSeek-based model option would be fully hosted on Microsoft’s own Azure infrastructure, with all customer data remaining strictly within the tenant boundary and covered by Microsoft’s existing enterprise security, compliance, and data-residency controls.
The company has also confirmed it has fine-tuned the model and applied additional safeguards, including specific adjustments aimed at уменьшение предвзятости.
The approach is familiar — it mirrors the strategy Microsoft has deployed through Лаборатория искусственного интеллекта Azure to offer third-party models under its own operational and contractual umbrella, essentially laundering third-party risk through Microsoft’s own enterprise guarantees. Whether that will satisfy the most security-conscious enterprise customers, however, remains to be seen.
Usage-Based Pricing Changes the Enterprise AI Calculus
Alongside the DeepSeek news, Microsoft’s decision to shift Copilot Cowork from flat-rate licensing to usage-based pricing through “Copilot Credits” represents a broader and potentially more consequential change for the enterprise AI market.
Costs under the new model are driven by a combination of model use, context retrieval, tool calls, and runtime. The tool ships выключено по умолчанию, and administrators are given granular spending controls at the tenant, group, and individual user levels.
Microsoft has claimed internally that Cowork is 30 to 40% cheaper than Anthropic’s Claude Cowork when paired with a Microsoft 365 connector, though the company acknowledged costs vary significantly depending on configuration.
The push toward a cheaper model tier sends an unmistakable message: even at those savings, the economics of deploying always-on AI agents at enterprise scale remain a genuine barrier to widespread adoption — and Microsoft is betting that разнообразие моделей, rather than model exclusivity, is the path to making the math work for its customers.
3 Key Takeaways at a Glance:
💸 Cost Is the Core Problem: Enterprise users running hundreds of tasks per week through Copilot Cowork are hitting unexpectedly high costs due to the token-intensive nature of multi-step agentic workflows — forcing Microsoft to urgently explore cheaper model alternatives like DeepSeek V4 and its own proprietary “Cowork 1” model.
🔒 DeepSeek, But on Microsoft’s Terms: To neutralize security concerns around a Chinese-developed model, Microsoft plans to host any DeepSeek V4 option entirely on Azure infrastructure, keeping customer data within enterprise boundaries and applying its own fine-tuning and bias safeguards — effectively the same playbook used across Azure AI Foundry.
📊 Usage-Based Pricing Reshapes Enterprise AI: The shift from flat-rate licensing to Copilot Credits-based consumption pricing is a major structural move, signaling that Microsoft views flexible, multi-model pricing — not a one-size-fits-all subscription — as the key to unlocking broad enterprise adoption of always-on AI agents.
Relevant Twitter/X Links:
🐦 Techmeme — Microsoft is moving Copilot Cowork to usage-based pricing and considering a Microsoft-hosted version of DeepSeek as a cheaper option с помощью @Techmeme
Быстрые ссылки:
