World AI Trends Through DeepSeek: West Blossoms, East Bears Fruit, World Shares
Leading Chinese semiconductor institute argues over-concentration of computing power has eroded US innovation. China must stay committed to openness and fairness to lead in tech
In recent days, I’ve found some interesting arguments talking about how DeepSeek led the re-evaluation of the Chinese stock market. During the Artificial Intelligence Action Summit, Vice President Vance dismissed Chinese open-source AI models while pushing the EU to loosen regulations over US-led models. The US and UK's refusal to sign the global AI agreement highlighted growing policy divisions between the US and EU.
For today's episode, I bring an op-ed published by ICWise, a research institute focusing on semiconductors and AI. They publish research on time on their official WeChat account. A WeChat account is essential for anyone studying Chinese affairs - as China's internet has become mobile-centric, WeChat articles have become the primary source of current information, surpassing traditional websites.
The piece named “AI Trends through DeepSeek: West Blossoms, East Bears Fruit, World Shares” 从DS看AI趋势：西方开花，东方结果，全球共享. I found it particularly interesting and worth sharing that it describes the AI competition as Chinese engineering talent vs. US computing power. With an advantage in AI usage and vast engineering capacity, the Chinese AI development advantage can be sustainable. It argues that AI technology, like the steam engine before it, will naturally flow from countries of origin to countries of application - an inevitable market trend. The piece frames the current AI wave as a continuation of previous technological revolutions, maintaining continuity in digital talent, business entities, and market advantages. The author argues that China's commitment to open-source and shared technology could not only help overcome trade barriers but potentially surpass its achievements from the mobile internet era.
At the Paris AI Safety Summit that concluded on February 11th, 60 countries, including India, Japan, France, Germany, Canada, and Australia, joined China in signing the 'Paris Call on AI,' despite US and UK opposition. Though it's a non-binding initiative, the rare split within Western allies was surprising. Faced with US attempts at an AI monopoly, most countries prefer open-source and shared AI solutions like DeepSeek, even if they are developed in China.
The declaration advocates for an open and inclusive approach to AI technology, calling for enhanced coordination and prevention of market concentration to improve AI accessibility. In stark contrast, Vice President JD Vance opposed Chinese open-source AI while boldly declaring that the Trump administration would ensure the most powerful AI systems are built in America using US-designed and manufactured chips. Harris's disregard for other nations' concerns, while possibly reflecting diplomatic inexperience, demonstrates America's 'AI supremacy at all costs' approach, using ideological differences to mask its technological monopoly attempts.
Developed nations outside the US and other Western economies, having gained little from previous tech waves, are naturally distancing themselves from America's AI monopolistic stance as they face this transformative technology.
This dramatic turn of events is arguably linked to DeepSeek's emergence. DeepSeek has rewritten the US-dominated global AI landscape, exciting the world and offering alternatives while inspiring other nations to develop their own AI capabilities. Without such a powerful AI system outside the US, where would other countries turn? This development suggests a David versus Goliath narrative in the US-China AI competition.
This article examines China's AI innovation patterns through the lens of DeepSeek to forecast the outcome of future US-China AI competition.
There's much debate about DeepSeek's merits and drawbacks. Let's examine its main contributions:
First, despite US sanctions and America's first-mover advantage, DeepSeek single-handedly brought Chinese AI near world-class levels, significantly accelerating China's high-tech industry. US AI experts previously estimated a 2-3 year lead over China. After DeepSeek's emergence, this gap narrowed to 3-6 months—effectively recovering at least two years for Chinese AI and the entire tech sector.
Whichever country first achieves superintelligence will gain asymmetric monopolistic advantages across all tech sectors. Thus, these two years aren't just crucial for Chinese AI but for China's entire high-tech industry.
While chip manufacturing determines computing power competition, former Intel CEO Gelsinger estimated China's chip industry lagged 5-10 years behind the US. Though China's chip industry will eventually catch up, the lengthy development timeline was concerning as it allowed the US AI lead to grow. Now, despite sanctions, Chinese AI has unexpectedly eliminated America's two-year advantage. This effectively bought precious time for breakthrough projects in chip equipment and manufacturing.
In applications, advanced AI models serve as innovation multipliers by making technical systems more intelligent. As US-China competition intensifies in autonomous driving, drug development, and countless other AI applications, DeepSeek provided Chinese companies access to world-leading AI tools two years ahead of schedule despite US blockades.
Therefore, these two years gained by DeepSeek provided invaluable assistance to China's high-tech industry.
Second, DeepSeek's ultra-low-cost open-source model rapidly pushed quality AI to market applications, enabling countless Chinese companies to enter the AI market and unleash China's superior advantages in application domains.
Chinese tech companies are now rapidly embracing AI, with numerous applications entering the market. Application-side strength was China's winning strategy in the internet and mobile internet eras. With China's diverse industries, rich application scenarios, and quick market adoption of innovations, Chinese companies can pioneer various AI applications.
This explosion in applications suggests technology may again follow the pattern of 'West blossoms, East bears fruit.' Unlike previous tech waves where Chinese applications remained domestic, this time Chinese AI shows strong international ambition. Chinese AI applications are likely to take root globally, truly launching China's tech industry into a new era of global expansion, both in quantity and quality.
Third, DeepSeek partially circumvents NVIDIA's strongest industry barrier—the CUDA ecosystem—making AI hardware universalization possible under certain conditions.
While AI training still requires NVIDIA chips, inference deployment can use other chips. This allows non-NVIDIA AI chips, including many Chinese ones, to quickly integrate with DeepSeek. DeepSeek has achieved a degree of AI technology democratization, fostering a healthier AI industry ecosystem and increasing China's autonomous AI computing power supply. Currently, DeepSeek's model series has been adapted and deployed on platforms like Ascend.
Fourth, DeepSeek has prompted global capital to reassess the Chinese market. Wall Street considers innovations like DeepSeek, emerging from outside their field of vision, the biggest threat to US technology-dominated financial markets.
Sanctions clearly cannot mitigate this risk; embracing Chinese technological innovation is the best way to avoid being blindsided by unexpected “challengers.” Though US capital won't return to China immediately, non-US capital has become more active, with Saudi capital, for instance, showing strong interest in DeepSeek.
Fifth, DeepSeek represents a rare Chinese enterprise achieving leadership in a global top-tier sector, rapidly capitalizing on unprecedented innovation dividends through technological breakthroughs—a major enlightenment for Chinese frontier innovation.
Previously, few Chinese companies had the confidence to challenge Silicon Valley in cutting-edge innovation fields, often settling for secondary innovation. This time, Chinese enterprises challenged America's most powerful companies and achieved significant results. Having tasted success, Chinese companies will undoubtedly pursue more frontier innovations.
In conclusion, whether viewed as a technological or engineering innovation, DeepSeek is a far-reaching achievement. While the current DeepSeek-OpenAI duopoly might be temporary due to increased US restrictions on Chinese AI computing power, DeepSeek has already unleashed China's and global AI potential, with countless companies integrating its technology. Whether DeepSeek ultimately proves to be a shooting star or a permanent fixture, Chinese companies have begun neutralizing US hardware advantages through applications, adding uncertainty to future US-China AI competition. Under asymmetric chip computing power conditions, US-China AI competition might evolve into a pattern where the US leads in select strong points while China dominates broader peripheral areas. To validate these assessments, let's analyze the US-China AI competition from four aspects.
Engineering Talent Advantage vs. Computing Power Advantage
In early 2025, Chinese innovations like DeepSeek, TikTok, and robotics deeply impacted global markets. Despite heavy sanctions, China's high-tech industry demonstrated remarkable resilience. The key factor: China's abundant, high-quality engineering talent overcame America's powerful hardware control. In future US-China competition, China's engineering talent will gain increased recognition.
According to MacroPolo's “Global AI Talent Tracker” report by the Paulson Foundation, graduates from Chinese universities constitute the largest group among top AI researchers in the US—increasing from 27% in 2019 to 38% in 2022, surpassing US university graduates at 37%. Additionally, about 30% of Silicon Valley talent is of Chinese descent. However, recent US de-Sinicization has forced many Chinese students, engineers, and academics to return to mainland China. Unsurprisingly, US AI experts consider only China a genuine competitor, as other countries lack sufficient talent.
The China AI Industry Development Alliance reports that China had nearly 500,000 AI researchers in 2023, compared to 300,000-400,000 in the US. DeepSeek's predominantly domestically educated team exemplifies China's high-quality talent pool. China's superior talent supply continues to expand with new graduates.
DeepSeek's engineers creatively combined algorithms with hardware technology for breakthrough results. Challenging NVIDIA on hardware was a bold move—one perhaps only Chinese engineers would conceive, as Western companies with unlimited AI chip access wouldn't attempt such a seemingly hopeless challenge.
Driven by computing power scarcity, Chinese companies had no choice. Chinese engineers' expertise, from mining rig development to current AI chip design, provided deep understanding of AI chip architecture. During China's mining peak, a single mining company ranked among TSMC's top three customers. This experience gave DeepSeek confidence to tackle computing power issues at the chip level. They bypassed NVIDIA's CUDA barrier by accessing lower-level PTX, manually optimizing memory bottlenecks, reducing execution time and processing delays, and maximizing GPU efficiency.
Simultaneously, DeepSeek pursued computing power efficiency in algorithms and AI learning. They skipped conventional intervention learning, directly improving reasoning through experimentation and feedback, saving substantial computing power.
An unexpected benefit emerged: PTX, though NVIDIA-developed, offers cross-platform compatibility. By bypassing CUDA, DeepSeek inadvertently achieved partial AI chip compatibility.
Such numerous technical innovations from one company demonstrate DeepSeek engineers' mastery of algorithms and data processing, plus deep understanding of GPU architecture, low-level programming, and cross-platform optimization.
In this battle between engineering talent and computing power blockade, carefully constructed barriers nearly became a Maginot Line. Long-term, engineering talent will triumph over chip advantages as US hardware superiority gradually erodes while China's talent pool grows stronger and larger.
Wild Lilies vs. Computing Power Accumulation
Before DeepSeek's emergence, China and the US invested money in developing powerful AI, with US AI increasingly government-dominated. At Trump's invitation, Masayoshi Son launched the $500 billion “Stargate” initiative in the US. Beyond Oracle's involvement, the US government planned direct funding, with the National Science Foundation guiding additional investments. Combined with OpenAI's first-mover advantage, it seemed the US government could maintain AI leadership through massive computing power investments, monopolizing AI innovation benefits.
However, while the US government focused on accumulating expensive computing power, a low-cost AI model emerged unexpectedly, demonstrating that government planning cannot guarantee breakthrough innovations. Frontier innovation is unpredictable in both technical approach and commercial form, requiring numerous trials to find the right path. Before discovering the correct answer, government funding isn't ideal for such experimentation, as unlike venture capital, it cannot tolerate massive losses.
Government support is better suited for established tech industries with clear technical roadmaps and business models, like China's semiconductor industry focusing on chip manufacturing, equipment, and materials. While these areas require large investments with slow returns and high risks, their clear technical paths and business models ensure success with concentrated national resources.
Furthermore, frontier technologies succeed only through full market competition and diverse exploration. DeepSeek's origination from a quantitative trading fund - a non-traditional AI entity often criticized for “misusing” GPUs - demonstrates this principle. Surprisingly, this “peripheral” player led the breakthrough.
In frontier innovation, success is unpredictable. Only with sufficient inclusiveness and patience can breakthroughs emerge. The richer the industrial forms, the more diverse the ecosystem, and the more active the market, the higher the probability of innovative success. Therefore, encouraging diverse exploration and tolerating different industrial forms is crucial. Since innovation cannot be planned, the government's role should be to create a permissive environment encouraging free exploration and trial-and-error.
Currently, US AI chip manufacturers, AI developers, and Wall Street form a mutually beneficial community, with continued government support for computing power accumulation, reinforcing the US's computing power-centric AI innovation trend. However, DeepSeek's low-cost innovation emerging in China rather than the US suggests that the concentration of computing power in the US government and big capital may have eroded American innovation soil. While uncertain how China's “wild lilies” will ultimately compete with US computing giants, China's AI innovation ecosystem currently shows greater diversity.
Open Source vs. Closed Source
Super AI both attracts and frightens everyone. Any nation achieving superintelligence first would gain asymmetric monopolistic advantages, likely maintaining them long-term. Therefore, whoever develops and promotes AI more inclusively will occupy the moral high ground and gain broader ecosystem support. This explains why US AI now stands opposite to global interests.
Open-source DeepSeek has reversed Western efforts to contain Chinese AI, leaving US government and companies without grounds for criticism. After days of observation following DeepSeek's release, Trump ultimately declared it a positive technology beneficial to America. Microsoft, Google, Amazon, and NVIDIA immediately integrated with it, as did most other countries. In today's isolationist climate, such a welcome for closely-guarded Chinese AI is remarkable, demonstrating the power of holding AI's moral high ground.
DeepSeek's open-source approach could drive global AI markets toward multipolar competition. If DeepSeek attracts global developers and becomes a crucial open-source AI technology, it could help China overcome US blockades combining closed-source models and chip sanctions.
Open source typically serves as a tool for disadvantaged parties to catch up with dominant ones. Even when dominant players recognize the power of open source, they struggle to take advantage of their existing technological advantages. OpenAI's $260 billion valuation is based on its closed-source model. Who could convince shareholders to sacrifice enormous profits for humanity's future by open-sourcing leading technology? More importantly, with China as America's only AI competitor, wouldn't open-sourcing compromise US chip dominance?
Historical lessons repeatedly show market leaders struggle to change course. Take electric vehicles: despite recognizing their potential, traditional auto powerhouses in Europe, America, and Japan couldn't fully commit to EV development due to their attachment to conventional powertrains, leading to today's situation.
If US open-source forces can't change the situation, US-China AI competition may form two camps around open vs. closed source. America's high-walled closed system versus China's open-source technology diffusion. America's technical advantages will be undermined by moral disadvantages, while China's technical disadvantages will be strengthened by moral advantages.
Application Scenarios vs. Top Computing Power
After DeepSeek's emergence, a viral question arose: Why did DeepSeek, a startup rather than a leading tech company, shake OpenAI's dominance?
A key distinction between DeepSeek and other AI powerhouses, including OpenAI, is that DeepSeek's parent company, High-Flyer, is already commercially successful. They entered AI by solving real-world problems, with their team tempered and selected through quantitative trading challenges. Market experience provided clearer direction and more precise evaluation systems, fostering stronger R&D teams and strategies. DeepSeek's hardware efficiency breakthrough likely drew inspiration from quantitative trading needs before transitioning to large models and achieving breakthrough success. This might answer the viral question and preview a contest between laboratory AI and real-scenario AI.
Compared to other countries, China's large population and comprehensive industrial categories offer unparalleled application scenarios and demands, backed by abundant engineering resources. The US has the strongest financial and technical reserves with unrestricted hardware support, enabling the development of powerful AI in laboratories. Thus, the US-China AI competition might become a contest between laboratory AI and application-integrated AI. Will superior computing power in laboratory AI prevail, or will real-world application-integrated AI succeed? AI focused on solving practical problems has discovered low-cost models first, bringing more market surprises.
More importantly, DeepSeek's open-source models have achieved AI miniaturization, facilitating AI implementation across industries and generating countless AI-reality combinations and business models. With ultra-large computing centers, US AI might achieve breakthroughs first. However, human experience and knowledge haven't been fully digitized, and AGI won't immediately become dominant upon arrival - it needs a process of real-world integration. If by then Chinese AI, though not fully mature, offers compelling cost-effectiveness, sufficient performance, and has established rich industrial ecosystems with global commercial entities, it could effectively compete with America's nascent AGI.
Conclusion
Throughout history, numerous strict technology blockades have occurred, yet none succeeded. Chinese silkworm eggs reached Rome via the Silk Road, French missionaries smuggled porcelain technology to France, Spanish sweet potato seedlings were smuggled into China, and America's industrial father replicated English textile machine blueprints... New technologies that create substantial employment and benefit everyone have never been successfully contained or monopolized; they inevitably find their way to where they're most needed.
Historically, new technologies have always transferred from innovating nations to application-focused countries, eventually transforming the receiving nation into an innovator. Britain initiated the Industrial Revolution, but British technology ultimately transferred to the larger American market. Despite intense technological conflicts between Britain and America, the trend of technology transfer continued unabated. Since China's Reform and Opening Up, despite various blockades, Western technology has continuously moved eastward, with Eastern creativity growing accordingly. Particularly in the last two technology waves, the West bloomed while the East bore fruit. This effect hasn't fully manifested due to trade wars and geopolitical influences.
The current AI wave essentially continues the previous two technological innovations, maintaining strong continuity in digital talent, commercial entities, and market advantages. Therefore, we have reason to believe the AI era's technological competition will follow similar transfer trends. The difference now is that most countries hold positions different from the US, pursuing open-source, shared, and fair AI technology, providing China's open-source AI with unprecedented international support. If Chinese AI, embracing global sharing concepts, can maximize global value, it could not only overcome blockades but achieve success far exceeding the mobile internet era.