China's DeepSeek is gearing up to launch its latest AI model, DeepSeek-R2, which is already being positioned as a fierce rival to OpenAI’s GPT-4o. Many see it as a growing concern for major chipmakers, especially Nvidia.
According to reports from both Chinese and Western media, the new model is built on a Mixture-of-Experts architecture.
This hybrid structure improves computational efficiency by activating different sub-networks based on the task, resulting in significantly lower resource consumption and cost.
In fact, estimates suggest that training costs for R2 are up to 97.3% lower than those for GPT-4o.
Superior Performance at a Lower Cost
One of R2’s standout features is its ability to operate with 1.2 trillion parameters, trained on a dataset of 5.2 petabytes.
This represents a massive leap forward compared to its predecessor, R1, which had already gained attention in the tech community for its efficiency and low cost.
Beyond its technical structure, R2 also stands out for its strategic vision. DeepSeek is committed to making its models open-source and freely accessible, directly challenging the dominance of companies like Meta and OpenAI.
This direction may position China as a global leader in open-source AI development—especially as tensions with the United States continue to intensify.
A Potential Blow to Nvidia
Recent developments have unsettled Nvidia investors, given the company’s strong reliance on demand for its chips to power large-scale AI models.
Back in January, Nvidia’s stock dropped 17% following the announcement of DeepSeek’s open-source R1 model.
Although the stock eventually recovered, the looming threat remains.
What’s more concerning now is DeepSeek’s shift toward using Huawei’s Ascend 910B processors instead of Nvidia’s chips. By doing so, the company is building a local supply chain that reduces dependence on U.S. hardware.
Reports suggest that Huawei-based systems can deliver up to 91% of the performance offered by Nvidia’s A100-based systems.
If DeepSeek successfully makes this transition, other Chinese tech firms may be encouraged to follow suit.
Opportunities and Challenges
Still, some analysts believe this trend toward more efficient AI models could benefit Nvidia in a different way.
By reducing technical and financial barriers, these models may broaden access to AI technology.
In turn, that could open up new market opportunities for Nvidia, even in the face of rising competition.
Currently, Nvidia’s stock is valued at around $101, about 10% below its market peak.
The performance of DeepSeek R2 may prove to be a decisive factor in shaping the company’s future in the global AI race.
