Microsoft Considers Integrating DeepSeek Models to Build a Low-Cost Version of Copilot Cowork
Taylor Wilson
Microsoft is exploring DeepSeek V4 for a cheaper version of Copilot Cowork, with a final model decision expected within weeks — this means DeepSeek is now competing inside Microsoft's enterprise AI workflow for the first time, as model-call costs become the real bottleneck for scaling enterprise AI.
Why does Microsoft need a cheaper Copilot Cowork?
Copilot Cowork executive Charles Lamanna spelled it out: some users complete hundreds of tasks per week — great for productivity, but the inference costs add up fast.
This means → enterprise AI has moved past "does it work?" and landed on "can we afford it?" — at high usage volumes, model-call fees become a real financial burden.
In plain terms = the tool works so well that employees keep using it more, but the bill grows just as fast. Microsoft needs a cheaper option.
What role does DeepSeek V4 play here?
Copilot Cowork currently runs on models from Anthropic and OpenAI. Microsoft confirmed to Axios it is exploring a version of DeepSeek V4 as a third-party low-cost option.
Microsoft previously disclosed that Copilot Cowork's per-prompt cost is already 30%–40% lower than Claude Cowork with Microsoft 365 connectors. If DeepSeek V4 is adopted, it would push costs down further for specific workloads.
This means → DeepSeek would not replace existing models. It would add a cheaper tier to the mix, matching different tasks to different cost levels.
How does Microsoft's own Cowork 1 relate to DeepSeek V4?
Microsoft is simultaneously developing its in-house model Cowork 1, also targeting low-cost workloads — positioned to deliver "the best balance of cost, quality, and enterprise-grade use."
Whether the two compete or complement each other depends on Microsoft's final decision, expected in the coming weeks. Both are being explored in parallel.
In plain terms = Microsoft holds two low-cost cards — one built in-house, one brought in from outside. Which card gets played, or whether both do, is still undecided.
What does this mean for the broader market?
This reflects a larger shift: enterprise AI is moving from "pick the strongest model" to a multi-model strategy that matches each task to its best cost-performance option.
The final model choice for the low-cost tier will be a key test of whether Microsoft's multi-model strategy can deliver on its cost promises at scale.
This means → whether DeepSeek enters Microsoft's enterprise product line is not just about one model's fate — it is a bellwether for the cost-competition landscape across the entire AI industry.
Content is for reference only, not financial advice.