There’s fresh movement in the AI space: Microsoft is evaluating the integration of DeepSeek V4, the V4 model from the Chinese open-source AI company DeepSeek, into its enterprise-grade Copilot Cowork as a lower-cost alternative. This development is worth serious consideration for Hong Kong enterprises and developers, especially as AI usage costs continue to climb.
Background and Facts
According to Axios reporting, Copilot Cowork currently relies primarily on high-end models from OpenAI and Anthropic. These models perform strongly in agentic AI tasks, but token consumption and compute costs are very high. As usage scales up, Microsoft has shifted Copilot Cowork to usage-based pricing and is exploring a multi-model strategy, including self-hosted and fine-tuned versions of DeepSeek V4.
Microsoft’s Foundry Blog highlights that DeepSeek V4 is known for high performance and highly competitive pricing, delivering impressive results on benchmarks in coding, mathematical reasoning, and more. Microsoft plans to run the model on Azure to ensure data compliance and security. This reflects a broader shift in the AI industry from “pursuit of the absolute strongest model” to a more pragmatic balance of performance and cost.
Real Impact on Hong Kong Enterprises
Hong Kong, as an international financial and technology hub, has seen many companies deeply integrate Microsoft 365 + Copilot for document processing, Teams meeting summaries, data analysis, and automation workflows. If DeepSeek options are added, the following changes could emerge:
- Cost Optimization: For mid-sized enterprises and startups, the cost of frontier closed-source models is a noticeable burden. Efficient models like DeepSeek can reduce daily task expenses, making advanced AI agent capabilities affordable for more teams.
- Flexible Model Selection: Users may be able to switch models based on the task — using GPT or Claude for creative writing and complex analysis, and DeepSeek for code generation or repetitive work — achieving optimal resource allocation.
- Open Source Mindset: As someone who has long followed open source, I believe DeepSeek’s open-source nature is noteworthy. It helps empower local developers to participate in fine-tuning and building custom solutions, reducing reliance on a single vendor.
Of course, geopolitical factors cannot be ignored. Although hosted by Microsoft on Azure, models originating from China may prompt some enterprises to pay extra attention to data privacy, supply chain risks, or compliance policies. Hong Kong companies should make decisions in line with the Personal Data (Privacy) Ordinance and internal risk assessments.
Advantages and Considerations
Advantages:
- Excellent value for money, ideal for Hong Kong SMEs to accelerate digital transformation.
- Microsoft platform integration experience remains unchanged, with a low learning curve for users.
- Promotes AI technology diversification and indirectly encourages the growth of the local open-source community.
Points to Watch:
- Output style and knowledge consistency may vary between models.
- Policy variables in the US-China tech environment.
- Some teams may prefer to maintain a purely Western model ecosystem.
My Thoughts
This move shows Microsoft being more pragmatic under commercial pressure and highlights the rising trend of open-source AI. For Hong Kong, this is a good time to re-evaluate AI strategies: in addition to relying on Copilot, consider evaluating DeepSeek’s independent API or on-premise deployment options, combined with your own data for fine-tuning.
What do you think? Do you support a multi-model hybrid approach to control costs, or do you place more emphasis on the stability of a single ecosystem? Feel free to leave a comment and let’s discuss together while keeping an eye on AI adoption in Hong Kong.