Microsoft made the OpenAI model optional on June 16, announcing it is exploring a fine-tuned version of DeepSeek V4 as a lower-cost alternative to the GPT-5 and Claude models currently powering Copilot Cowork.
DeepSeek V4 runs at $0.14 to $0.28 per million tokens, an order of magnitude less than the frontier OpenAI and Anthropic models currently inside Cowork. Charles Lamanna (Microsoft EVP, Copilot, Agents and Platform) told Axios that power users “do hundreds of tasks a week, which is great … but the consequence is the costs can go very high.”
Microsoft says it would host DeepSeek (a Chinese model) fully on Azure, covered by Azure’s enterprise security, compliance, and data-residency controls, which solves the geopolitical concern directly. DeepSeek’s weights are open source, but once they’re running inside Azure’s compliance boundary, they’re an Azure service. Historically, Microsoft wraps, brands, and sells what it doesn’t build, so this is business as usual for Redmond.
AWS Bedrock has been multi-model since 2023 and now offers 30+ model choices. Google Vertex AI does the same. Microsoft is the last major cloud provider to go multi-model on its flagship AI product.
The $13.75 billion Microsoft invested in OpenAI bought brand permission and credibility at a moment when enterprises needed someone to trust on AI. That moment served its purpose. Today, models are plumbing, and Microsoft is saying so publicly.
AI deployment at enterprise-scale begins and ends with your corporate knowledge base (by far the highest hill to climb). Where and how to run it just got a little easier. You can now choose from any of the big three hyperscalers.
Every company needs a Claw strategy. Do you have one?
Author’s note: This is not a sponsored post. I am the author of this article and it expresses my own opinions. I am not, nor is my company, receiving compensation for it. This work was created with the assistance of various generative AI models.