AI
In the early 2025 global tech world, an algorithm - driven "storm" is quietly rewriting the power map of the AI industry. Chinese AI startup DeepSeek has taken the open - source large - model DeepSeek - R1 as its spear, and "low - cost + high - performance" as its shield, triggering a global shock across technology, capital, and geopolitics in just two months. Nvidia's stock price plummeted 17% in a single day, Meta's market value dropped by $100 billion, and the Nasdaq index fell 3%... The source of the storm is DeepSeek's innovation that stunned Silicon Valley — its training cost was only $6 million, yet its performance rivals OpenAI's GPT - 4o. By February 1, 2025, DeepSeek had already topped the Chinese productivity application chart.
When DeepSeek reached Wall Street, Silicon Valley engineers were faced with a harsh reality: when "Open R1" received 1.9k stars on GitHub in just one day, global developers chose the Chinese open - source framework with their actions. The rise of DeepSeek is not only a milestone for Chinese AI technology going global, but also signals that huge funds and powerful computing power no longer tightly control the fate of AI. Chinese entrepreneurs are rewriting the global AI landscape.
On January 27, 2025, DeepSeek made history by becoming the first Chinese AI product to simultaneously top the free app charts in both the Chinese and American App Stores, marking its official entry into the global mainstream market. The key to this rapid success of the tool lies in its extremely low cost — it's cheap, open - source, and user - friendly. The development cost of DeepSeek - R1 was only $6 million, saving over 98% compared to OpenAI's O1 model (about $500 million). Moreover, the DeepSeek - R1 API entered the market at a highly competitive price, costing only 0.5 yuan per million input tokens (cache hit) or 2 yuan (cache miss), and 8 yuan per million output tokens, significantly reducing usage costs compared to OpenAI o1. This significant cost advantage is mainly due to innovative training methods and a carefully designed architecture that reduces training costs while maintaining model performance. Compared to previous computer revolutions, this change is happening much faster, reshaping the cost structure and application model of AI with unprecedented efficiency.
This breakthrough has shattered the long - held perception of high AI technology thresholds and triggered a series of far - reaching impacts. First, the success of DeepSeek proves that large models can be open - sourced and are no longer the exclusive domain of a few tech giants. This means that researchers, independent developers, and startups worldwide can enter the AI track at a lower cost, accelerating the popularization of technological innovation. Second, it shows the trend that technology is no longer monopolized by American tech giants. In the past, companies like OpenAI and Google DeepMind dominated the development of large models. In short, the rise of DeepSeek not only demonstrates the competitiveness of China's AI industry in technological innovation, model training, and global application but also promotes a profound transformation of the global tech landscape, heralding the emergence of more AI products and technical standards led by Chinese enterprises.
The technical approach of DeepSeek is essentially a hedge - fund - style "risk - reward ratio" gamble. More than 60% of its core team have math competition backgrounds, and founder Liang Wenfeng has infused quantitative trading thinking into AI research and development. Since Liang Wenfeng announced his entry into the AGI field in May 2023, DeepSeek has quickly risen to become a significant force in China's AI track. In July, he officially founded Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co., Ltd. (DeepSeek) and attracted industry attention with its groundbreaking organizational structure and resource allocation strategy. In talent recruitment, DeepSeek abandons the traditional experience - centered approach, favoring instead those with passion and curiosity, especially young people eager to explore the AI industry. This strategy injects fresh thinking into the team and ensures continuous technological innovation. DeepSeek's organizational management is also revolutionary, adopting a completely bottom - up work model without fixed positions, with divisions naturally forming through collaboration. When an idea shows real potential, the company quickly allocates resources top - down to fuel its rapid growth. This highly flexible mechanism ensures the agility and efficiency of technological exploration. In terms of resource supply, DeepSeek adopts an extremely open computing resource strategy, giving team members maximum freedom to create a relaxed and efficient frontier exploration environment. Liang Wenfeng emphasizes that the core goal of this strategy is to unleash innovation speed and break through technological bottlenecks. Facts have proven that this model has delivered astonishing results. For example, the pre - training of the DeepSeek V3 large model used only 2048 GPUs, took two months, and cost about $5.576 million, yet achieved performance comparable to OpenAI's GPT - 4. According to Foreign Media Semianalysis, the training cost of GPT - 4 is as high as $63 million, 11 times that of DeepSeek. This significant cost advantage not only highlights DeepSeek's extreme cost - effectiveness in the AI field but also consolidates its competitiveness on the technological frontier.
The impact of DeepSeek is reshaping the value distribution of the global AI industry chain. On the training side, Nvidia's data center business revenue forecast has been cut by 7%, with a $37.5 billion market size facing restructuring risks.
The essence of this transformation is the shift of AI value focus from "infrastructure investment" to "scenario penetration." Before this, especially in AGI and large - model fields, Chinese technology hadn't been so widely applied across global industries. Domestically, on February 28, the global launch of the Honor Magic 7 series featured an on - device AI assistant based on the DeepSeek - R1 - Distill model, enabling localized legal consultation and medical diagnosis. Overseas, Saudi Arabia's Public Investment Fund (PIF) invested $5 billion in DeepSeek technology to build a "de - Westernized" AI infrastructure. Silicon Valley finally realized: emerging markets are bypassing traditional IT architectures and entering the "AI - native" era directly. This means Chinese companies have established a "chip - algorithm - application" capability loop in consumer electronics.
Meanwhile, DeepSeek's open - source strategy is tearing apart Silicon Valley's closed - source empire. On January 28, the Hugging Face community launched the "Open R1" project to replicate its training process. In just one week, over 4,700 derivative code repositories appeared on GitHub, covering 23 vertical fields like medical diagnosis and industrial quality inspection. Behind this ecological is DeepSeek's "technical bait": its open - source code deliberately retains some unoptimized modules to attract developer improvements, creating a self - reinforcing flywheel effect.
EqualOcean analyses that through DeepSeek's low - cost, open - source model and global market breakthroughs, it may prompt US tech giants to adjust strategies, from accelerating innovation, increasing financing to reassessing open - source models. Meanwhile, the global AI industry's entry barriers are lowered, innovation speeds up, and more competitive AI products may emerge, further reshaping the global tech landscape. The New York Times commented: "Chinese AI companies have grasped the agenda - setting ability for the first time — they no longer respond to Silicon Valley's rules but set their own."
But undercurrents are always surging. For Chinese AI going global, a more real threat comes from the hardware end: although the AI inference performance of SMIC's 7nm process and Huawei Ascend chips reaches 60% of NVIDIA H100, the US ban on H800 chips still casts a shadow over DeepSeek's long - term computing power reserves. Given US export restrictions, DeepSeek and other Chinese AI model developers may focus more on software research. However, this situation's deep - seated change isn't over. The latest AI proliferation rules effective in January 2025 will further tighten export restrictions, locking China's hardware level at the 2022 technological generation, while US developers can still access the most advanced AI hardware.
Besides export controls, DeepSeek's rapid rise and market may trigger stricter reviews by regulatory bodies worldwide, posing short - term resistance to its global expansion. For instance, Italy and Ireland have banned DeepSeek downloads in their markets, with Italy's data protection authority issuing restriction orders to Hangzhou and Beijing DeepSeek, citing unclear personal information collection, data processing methods, and whether storage locations are in China. Meanwhile, the US Commerce Department is investigating whether DeepSeek uses export - controlled chips, and the US Navy has instructed service members to avoid its AI products. Given DeepSeek's influence and ongoing investigations, more countries may impose similar restrictions in the coming weeks.
Thus, DeepSeek's global expansion faces greater resistance. The European market bans and US security reviews may reduce its international market share. Also, against the high regulatory barriers of the US and Europe, Chinese AI companies may have to accelerate building a stronger independent and controllable tech ecosystem, including optimizing data governance, exploring new commercialization paths, and strengthening tech cooperation with non - Western markets, to mitigate the impact of globalization obstacles.
Compared to Western firms, Chinese AI and cloud providers have unique positions in global competition, backed by a large engineering talent pool, vast consumer market, massive data advantages, and closely - integrated internet ecosystems. However, core technology self - sufficiency remains a key challenge, especially with high - end chip restrictions, which may become a critical bottleneck for China's AI development.
On February 19, The Information reported that DeepSeek is considering its first external financing. In recent weeks, many well - funded investors, including Alibaba Group and state - owned funds, have expressed interest in financing DeepSeek's next - stage growth. As a representative of China's AI innovation, DeepSeek's expansion and upgrade may have just begun.
DeepSeek's overseas expansion reflects the unique survival strategy of Chinese tech companies in globalization — using algorithm optimization to offset computing power shortcomings and open - source ecosystems to break technological blockades. In the global AI market expansion, DeepSeek has not only secured a place in the overseas developer ecosystem but also triggered profound impacts on technical standards. When global developers start submitting code comments in Chinese, a deeper change is underway: the AI industry's "technical standard - setting power" is quietly shifting from Silicon Valley to the East. When Huawei Ascend chips and SMIC's 7nm production line form a domestic computing power chain, DeepSeek's global expansion and China's strong AI going - global are just beginning. This offers important insights for Chinese AI going global — by changing local technical standards and entry thresholds, Chinese AI can gain more market share globally.
But at the same time, cost reduction and standard changes only bring more market share. Besides opportunities, limitations must be recognized. EqualOcean considers that by cutting costs and lowering DeepSeek's usage barriers, China has just secured a place in international AI competition. In other words, despite its wide application, DeepSeek is still a Chinese replica of OpenAI's ChatGPT in terms of technical core and application scenarios. Just like Facebook to Weibo, Amazon to Taobao, and PayPal to Alipay — almost all Chinese apps have prototypes in the US and Europe. This means China isn't truly leading in technological innovation but is following in the footsteps of the West.
In short, only when China can create applications and core technologies that the West emulates will it truly lead in innovation and gain global technological dominance — and China's AI still has a long way to go to reach this groundbreaking scenario.
China's New Energy Intelligent Vehicle Industry Chain Enterprises Have Opened a New Era
Feb 27, 2025 03:42 PM
AIEV Leads the Way into a New Era of Automotive Intelligence
Feb 27, 2025 03:27 PM