© ROOT-NATION.com - Use of content is permitted with a backlink.
Microsoft has officially confirmed that it is currently considering the possibility of integrating a specially optimized version of the DeepSeek V4 model on its own servers as a significantly more cost-effective alternative to tools from Anthropic and OpenAI, which currently power the Copilot Cowork system.
It is worth noting that just today, Microsoft made Copilot Cowork generally available to a wider range of Microsoft 365 users, while simultaneously transitioning the tool to a new pricing model based on actual resource consumption. Both of these steps confirm the overarching conclusion that maintaining the operational viability of complex autonomous AI agents at an enterprise scale while retaining the old financial model is economically unfeasible and impossible in the long term.
Read also: AERONAUT – everything that flies above the ground: aviation, UAVs and drones, rockets, and space
The main reason that fully justifies such a search for new solutions lies in the specific cost figures. Anthropic’s latest flagship product requires a fee of approximately $50 for every million tokens processed, whereas using the advanced version, DeepSeek V4 Pro, costs developers approximately $0.87 for the same volume of data. Such a colossal difference in cost – about 57 times – is not a mere statistical error or a minor discrepancy in reports.
It is a serious structural obstacle for any enterprise attempting to implement large-scale automated workflows, where a digital assistant is forced to sequentially run dozens of interrelated queries to the model in order to perform a single operation, such as sorting incoming emails, preparing a text report, or coordinating actions between the corporate messaging platforms Teams and Outlook.
Large enterprises’ financial expenditures on integrating with AI technologies have skyrocketed 30-fold compared to 2023 figures, and according to analytical forecasts by Goldman Sachs bankers, the further development and mass deployment of autonomous work agents could increase overall demand for tokens by another 24 times from current levels. Under these conditions, traditional mathematical calculations of business profitability simply cease to apply, and Microsoft’s leadership has apparently concluded that this critical moment has already arrived.
To better understand the context, it should be explained that Copilot Cowork is a specialized feature based on autonomous AI within the Microsoft 365 ecosystem, developed in close collaboration with experts at Anthropic. The main purpose of this tool is to independently perform complex, multi-step operations within various office applications, including Excel, Teams, and Outlook.
The system was released worldwide on June 16. To activate this feature, customers must have a standard Microsoft 365 Copilot license, which costs $30 per user per month; however, from now on, additional charges for resources actually used will be added to this amount through a special Copilot Credits system. At the same time, the total cost of performing a specific task will depend directly on which language model was used, how much related context had to be retrieved from the database, how many tools were activated during the computation process, and how long the system was in operation overall.
It is this last detail that represents the most radical change in the tech giant’s approach. The previous fixed pricing plans seemed entirely justified and logical at a time when the intensity of use of intelligent systems remained at a stable and predictable level. However, the behavior of autonomous AI agents cannot be easily predicted, since the amount of resources consumed scales in parallel with the complexity of the tasks at hand, and solving precisely these kinds of complex corporate problems is the main reason why large companies purchase such software.
The huge difference in token costs between Anthropic’s flagship product and the DeepSeek V4 Pro model is not merely an abstract theory, but a real barrier that determines whether autonomous AI agents will become an effective tool for boosting business competitiveness or whether they will turn into a massive expense in the ITbudget that will send corporate CFOs into a panic.
Microsoft’s intention to integrate the new model could lead to serious disputes not only with corporate cybersecurity departments in private companies, as well as with representatives of the White House, which has been making significant efforts for many months to completely isolate U.S. digital infrastructure from the infiltration of Chinese AI technologies. The DeepSeek model was designed and created by developers from China, and this fact carries enormous political significance regardless of which specific servers will perform the actual mathematical computations.
In response to such concerns, Microsoft representatives note that any implementation of DeepSeek will be carried out exclusively on a voluntary basis at the customers’ own discretion, and the model itself will be hosted entirely within the Azure cloud platform. This will allow all users’ confidential information to be kept within Microsoft’s closed storage environment, protected by the company’s enterprise-grade security systems.
Whether this step will be sufficient to resolve all disputes remains a subject of active debate among experts. After all, physically hosting a copy of the model in the Azure cloud only solves the problem of local data storage; it does nothing to change the fact of who exactly conducted the initial training of this neural network, where it took place, and what source data was used.
However, the American corporation is not limiting itself to this single option. It is expected that, over time, Microsoft will expand the list of available tools in Copilot Cowork by adding other solutions, which could potentially include Meta’s Llama 4 model, as well as the latest version of Mistral’s development. Such plans clearly indicate the formation of a flexible, multimodal strategy, rather than a blind reliance on just one specific hardware or software provider.
In addition, insider reports suggest that Microsoft has already imposed certain restrictions on the use of the Claude Fable 5 model within the Copilot Cowork ecosystem. According to available information, this move was prompted by serious concerns regarding internal policies on the storage and retention of users’ confidential data. This detail points to certain tensions and misunderstandings in the relationship with Anthropic, which go far beyond the usual financial calculations of service costs.
It is also worth noting that the GitHub Copilot service made an identical transition to a pay-as-you-go consumer model on June 1, 2026, which confirms the systematic and coordinated nature of Microsoft’s efforts to overhaul its entire portfolio of AI products, rather than merely a spontaneous reaction to current financial pressures.
Read also:
- Instead of Microsoft Flight Simulator: Google Earth launches free flights right in your browser
- Microsoft Build 2026: From Language Models to New OS-Level Security Architecture