In the latest development in China’s artificial intelligence (AI) race, DeepSeek has released a sneak peek of its new AI model built for Huawei chips. This development marks a shift away from the reliance on Nvidia chips and underscores China’s push for an independent AI industry.
The Chinese start-up DeepSeek, which captured the world’s imagination last year with its cheap but powerful AI model, is now bringing itself front and centre of China’s tech sovereignty drive. Through its latest V4 model, which is designed to run on Huawei’s Ascend chips, it is showing how high-end AI models can increasingly be trained and deployed on local hardware.
This is a trend that is closely followed by industry observers, international rivals and policymakers.
DeepSeek’s Rise in the AI World
DeepSeek first came to prominence when it shocked the global tech industry with a cost-effective AI model capable of delivering impressive results for less money than many other models.
In a field where many players have deep pockets and access to the latest and most expensive hardware, DeepSeek’s success implied that efficiency and innovation can upend the status quo in the world of AI.
But now, with the launch of the V4 preview version, DeepSeek is poised to focus more on the performance front.
Strong Benchmark Performance
DeepSeek claims the Pro version of V4 is competitive with many open-source AI models on world-knowledge tests. It says it is second only to Google’s closed-source Gemini-Pro-3.1 in these tests among the top models.
If these benchmarks hold up to further testing, V4 will be one of the best open AI models on the market.
This is important for benchmarking because it provides developers and enterprises the assurance that a model can perform the range of tasks in reasoning, knowledge tasks, code generation, and other productivity applications.
Why Huawei Chips Matter?
It’s not the model, though, that’s the most exciting part of this announcement.
Previously, DeepSeek has used Nvidia hardware, which is currently the leading AI hardware provider. Nvidia GPUs are the platform of choice for training and deploying AI models.
But with U.S. export controls and China’s long-term aim to achieve technological self-sufficiency, there is growing urgency for Chinese firms to come up with their own solutions.
Huawei’s Ascend chips are now seen as China’s best homegrown alternative to Nvidia’s GPUs.
In porting V4 to Huawei, DeepSeek is demonstrating that state-of-the-art Chinese AI models can be deployed on Chinese infrastructure.
A Big Moment for China’s AI Industry
The partnership is a big moment for the industry.
The ability to run a leading AI model on locally-made chips shows China is moving in the right direction in moving away from foreign technology. That’s significant for business and for government.
Over the past few years, China has been pouring resources into semiconductors, cloud computing and AI. The DeepSeek-Huawei partnership is an example of how advances in these areas might be beginning to show.
Rather than being dependent on imported chips, Chinese AI startups are creating an ecosystem of local models on local chips.
Fast Momentum Among Developers
The V4 model is said to have been one of the fastest models to reach the top of Hugging Face, one of the largest developer platforms for machine learning models.
This is important because developer interest is a good indicator of future use. If developers, academics and startups start creating tools to use DeepSeek V4, it could gain traction.
Popularity on sites like Hugging Face can lead to:
- Community trust
- Third-party integrations
- Faster experimentation
- Real-world use cases
- Global visibility
- Competitive Cost Advantage
- Cost is also reported to be an advantage of V4.
DeepSeek has gained popularity by offering good AI performance at lower cost than others. If V4 follows suit, it could become a hot commodity for companies hoping to use AI without breaking the bank on infrastructure.
Cheaper AI might be particularly attractive for startups, developers and developing economies where cost is a significant barrier.
Current Limitations
However, the model does have limitations.
V4 is rumored not to support multimodal intelligence such as image or video processing. Most modern AI systems are developing multimodal capabilities to integrate text, images, audio and video.
So without multimodal capabilities, V4 might excel at text-based tasks like:
- Writing assistance
- Coding support
- Long document analysis
- Research summarization
- Enterprise productivity tools
- International Concerns About AI
DeepSeek’s achievements have also drawn ire from the U.S. government and some U.S. rivals for potentially being too dependent on American technology or expertise.
While DeepSeek has previously acknowledged using Nvidia chips, it has not provided information about whether these chips were exported. It has also denied deliberately using OpenAI’s “faked” data.
These controversies are part of a geopolitical competition over semiconductors, AI, and intellectual property.
What This Means for Nvidia
Nvidia is still the world’s leading AI chip maker, but moves like DeepSeek’s Huawei announcement present longer-term concerns.
Dependence on local hardware by Chinese developers could mean Nvidia loses some ground in a key AI market.
This is particularly concerning given China’s potential as a chip market.
Final Thoughts
DeepSeek’s release of V4 for Huawei chips exemplifies China’s rapid move towards tech self-reliance.
China is forging a more self-reliant future for AI by linking competitive AI software with domestic semiconductor capabilities. DeepSeek’s rise also demonstrates innovation can come from outside the “Four Horsemen” of Silicon Valley.
The battle for global AI supremacy is now about more than just the best model. It is also about who has the chips, the ecosystem and the supply chain. In this race, DeepSeek and Huawei is making a strong play.
