From San Francisco to Hangzhou: How DeepSeek is shaping the next big shift

Featured Article by Veena V

The global AI race has shifted from a focus on innovation aimed at solving problems to a fight for dominance. For quite some time now, the U.S. has been the dominant player in the AI race, with China remaining relatively less prominent on the global stage. However, with the recent release of DeepSeek-R1, the spotlight is on China, signalling a shift in the battle for AI supremacy. DeepSeek, the Chinese AI company is making waves with its ambitious model R1 that has challenged Western AI models in capability, cost-effectiveness, and accessibility. Certain historical and cultural aspects of China may have laid the foundation for this advancement, positioning the country to challenge the global AI landscape in a new way. What cultural aspects of China could have acted as a catalyst? What does DeepSeek’s rise mean for the AI landscape? In this article, we’ll explore these and also what lies ahead.

China vs. America:

Certain key cultural, economic, technological differences exist between China and America.

• Manufacturing Driven by Cost Effectiveness: China has traditionally been an economy driven by cost-effective manufacturing and developments leading to high-quality output at competitive prices, while America is expensive due to the overall cost including those arising from regulations. China mostly manufactures the hardware and its components for the U.S. while the U.S. develops and provides the software and semi-conductors and they were complementing each other. The current dynamics are different in a way that there is direct competition.
• Reverse Engineering: China has a rich history of iteration and adaptation. They have been quick to reverse engineer and repurpose technology. Following a period of stagnation, post-1978 reforms under Deng Xiaoping were a turning point in China’s history, marking the transition from a command economy to a market-driven one. These reforms encouraged a culture of mass production, cost-efficiency, and global integration, which allowed China to become a global economic powerhouse. Companies like Huawei and Xiaomi are built on offering high-quality products at lower costs than their Western competitors. U.S. has been an innovation hub. Companies such as Apple, Microsoft, Google are known for their innovation.
• Leapfrogging: China is known to leapfrogging in various sectors like mobility, communication networks by skipping certain stages of technological development and incremental innovation. America’s approach to leapfrogging has been about pioneering new technologies that reshaped the global landscape.
• Talent Pool: China has a growing pool of top-tier talent in AI, machine learning, and other tech fields, partly due to its large number of universities focused on STEM (Science, Technology, Engineering, and Mathematics). “New Generation Artificial Intelligence Development Plan" of 2017 in China set ambitious goals to make China the global leader in AI by 2030. This plan includes a focus on building a strong AI talent through massive investments in universities, AI research centres and labs. DeepSeek has followed an unconventional approach of hiring humanities graduates and Gen-Z interns to develop DeepSeek R1. Liang Wenfeng was looking out for a creative talent pool with deep understanding of human behaviour. The U.S. houses many of the world's leading universities and research institutions such as MIT, Stanford and Harvard which offer cutting-edge programs in AI. Their hiring approach is mostly focussed on relevant technical prowess.
• Cost of Labor: China has a skilled workforce and relatively lower labor costs compared to the U.S.
• Data: Compared to the U.S., China has massive datasets to train their LLMs giving them a critical advantage.
• Population and network effect: China’s large population estimated at 1.41 billion people in 2024 allows for economies of scale. The volume creates a network effect accelerating adoption.

Competition landscape:

• Global players: LLMs from established global players such as OpenAI’s GPT models, Anthropic’s Claude, Meta’s LLaMA, xAI’s recent Grok 3 etc from the U.S., Mistral models from France, UK players such as Google DeepMind’s Gemini and many more.
• Local players: Top Chinese players who are quickly catching up are Moonshot AI (Kimi), Alibaba’s Qwen, Baidu’s Ernie, Hunyuan (latest being Turbo S) from Tencent, and several other start-ups have emerged backed by abundant venture capital.
• AI Tigers: Four fast-rising Chinese unicorns – Moonshot AI, MiniMax, Baichuan, Zhipu AI have been locally referred to as China’s new “AI Tigers”.
• Distilled models: AI researchers from Stanford and the University of Washington have built an AI reasoning model, S1, in 26 minutes for under $50 through distillation, a process by which a model (student) learns from larger models (teacher) without the need for extensive resources. Inference time could also reduce in distilled models.

The AI space is rapidly evolving, and while many companies are competing to develop the most advanced models, each of them will likely carve out its own niche and address different use cases. It could be industry specific such as healthcare, could contain tailored algorithms for a specific geography taking cultural aspects into account. They may also cater to different use cases such as computer vision, speech recognition or NLP, technological differences. Baidu’s Ernie models are tailored for NLP tasks making them suitable for complex language related challenges. The key here is to reduce costs and improve access in order to solve global problems. 

Opportunities:

• Synthetic data: With the reduction in AI development and usage costs, data could be the next big focus area, synthetic data generation marketplaces might flourish.
• Distilled models: DeepSeek is expediting its release of R2 model which is expected to improve the model capabilities of R1 significantly, and provide expanded language support. They also have several distilled model variants.

Distilled models might become a major trend for cost-cutting strategies in AI deployment.

• Reliance on domestic technologies: If China plans to ban LLMs of other countries to encourage domestic innovation and reduce reliance on foreign technologies - given its population, DeepSeek and other companies might be at a significant advantage.
• Quantum computing: With the integration of AI with quantum computing, AI is poised to become exponentially more efficient. Microsoft ‘s Majorana 1 and Amazon ‘s Ocelot are quantum chips that could pave the way for this advancement, potentially enabling us to solve some complex challenges that we haven’t been able to solve previously.

Impact:

• VC funding dynamics: AI investment strategies might change. About 33% of VC funding was directed to AI companies in 2024 according to natlawreview.com, accounting to an increase of over 80% in funding compared to 2023. This was driven by the desire to capitalise on the growing popularity of AI. This year with changes in AI landscape, VC evaluation and investment strategies might focus on affordability, sustainability, scalability, and business or real-world impact.
• Global competitive pressure: Competitive pressure on U.S. and other AI majors leading to more research on low cost LLMs.
• Broader impact on the tech sector: Oracle, Microsoft and a few other tech stock prices have fluctuated reflecting the changing sentiment among the investors given the market uncertainties.
• Impact on Chip firms: Nvidia, Broadcom and AMD stocks have experienced declines and fluctuations. Microsoft has recently cancelled several data center leases in the United States, totalling several hundred megawatts of capacity.
• Reality check: Industry has been pumping billions into AI anticipating the compute requirements, this serves as a reality check.
• Consumer orientation: AI Research was not consumer-oriented in terms of accessibility; it was more about development of LLMs which might now change.
• Geographical dynamics: DeepSeek has a strong presence in China (39%) and the U.S. (16%), OpenAI's user base is more concentrated in the U.S., India, France, Russia, Spain, and the UK. Therefore, DeepSeek has a substantial opportunity to expand its user base in the U.S. market. It is possible that some cost effective LLMs emerge in the U.S. meanwhile, and gain rapid momentum.
• Commoditisation of AI: With the market getting flooded with AI players and their LLMs, there is possibly an early sign of it getting commoditised in the near future.

Challenges:

• Global Pushback: While China has successfully leapfrogged in AI - privacy, data security, and ethical concerns exist.
• Current climate: Some countries like South Korea have temporarily suspended DeepSeek fearing that their data would fall in the hands of the Chinese government.

Closing and call to action: As we continue developing AI models, it’s crucial that don’t just build solutions to sell but consider addressing the world’s real challenges. There are areas where we seem to be aiming to replace human interaction, we should rather prioritize creating solutions that complement and enhance human interaction or capabilities. This will improve lives without diminishing the personal touch that is essential especially in areas like healthcare, education, customer service. By keeping humanity at the center of technological development, we can ensure that AI serves as a tool for progress, not just replacement. On the other hand, as users of AI technology, we must use these tools responsibly, ensuring that we don’t lose sight of the essential human touch that enriches our lives. If we don’t, we’ll reach a tipping point where a reality check is unavoidable, but undoing the damage could be very difficult.  

References:

https://www.reuters.com/technology/artificial-intelligence/deepseek-rushes-launch-new-ai-model-china-goes-all-2025-02-25/?_bhlid=3f2dea62f33cbc00ff08a2da1df4e8ba136a940d

https://www.scmp.com/tech/big-tech/article/3259499/chinas-four-new-ai-tigers-baichuan-zhipu-ai-moonshot-ai-and-minimax-emerge-investor-favourites?

https://www.demandsage.com/deepseek-statistics/?

https://timesofindia.indiatimes.com/technology/tech-news/ai-researchers-develop-reasoning-model-for-under-50-like-openais-o1-and-deepseeks-r1/articleshow/117986776.cms