More languages
More actions
v1 Tag: Visual edit |
→Bridging the technology gap in the Global South: can't really cite as it's moreso opinion/clear potential, lmk Tag: Visual edit |
||
| Line 37: | Line 37: | ||
A 2026 [[Microsoft]] study confirmed the effect: while the AI adoption gap between the Global North and South persists, DeepSeek’s cost-free, open model has "lowered the threshold for millions of users, especially in price-sensitive regions."<ref>{{Web citation|newspaper=新华网 (参考消息)|title=DeepSeek推动全球南方广泛使用人工智能|date=2026-01-12|url=http://www.news.cn/liangzi/20260112/30c68eccd04e4abf9337447f6bb88c49/c.html|retrieved=2026-05-09|quote=DeepSeek不收取费用，"为数百万用户降低了门槛，特别是在对价格敏感的地区"。报告补充说："这种开放性和可负担性的结合，使DeepSeek在西方人工智能平台服务不足的市场获得了关注。DeepSeek的崛起表明，全球人工智能的普及不仅取决于模型质量，也同样取决于获取途径与可用性。"(DeepSeek charges no fees, "lowering the threshold for millions of users, especially in price-sensitive regions." The report added: "This combination of openness and affordability has enabled DeepSeek to gain traction in markets underserved by Western AI platforms. The rise of DeepSeek shows that global AI adoption depends not only on model quality, but equally on access and availability.") (translated from Chinese)}}</ref> The study noted that in countries where Western platforms are blocked or struggle to operate — Russia, Iran, Cuba, Belarus — DeepSeek fills the gap, expanding Chinese influence in regions underserved by U.S. tech monopolies.<ref>{{Web citation|newspaper=新华网 (参考消息)|title=DeepSeek推动全球南方广泛使用人工智能|date=2026-01-12|url=http://www.news.cn/liangzi/20260112/30c68eccd04e4abf9337447f6bb88c49/c.html|retrieved=2026-05-09|quote=报告称："开源人工智能可以作为地缘政治工具发挥作用，在西方平台难以运营的地区扩大中国的影响力。"(The report stated: "Open-source AI can function as a geopolitical tool, expanding China's influence in regions where Western platforms struggle to operate.") (translated from Chinese)}}</ref> | A 2026 [[Microsoft]] study confirmed the effect: while the AI adoption gap between the Global North and South persists, DeepSeek’s cost-free, open model has "lowered the threshold for millions of users, especially in price-sensitive regions."<ref>{{Web citation|newspaper=新华网 (参考消息)|title=DeepSeek推动全球南方广泛使用人工智能|date=2026-01-12|url=http://www.news.cn/liangzi/20260112/30c68eccd04e4abf9337447f6bb88c49/c.html|retrieved=2026-05-09|quote=DeepSeek不收取费用，"为数百万用户降低了门槛，特别是在对价格敏感的地区"。报告补充说："这种开放性和可负担性的结合，使DeepSeek在西方人工智能平台服务不足的市场获得了关注。DeepSeek的崛起表明，全球人工智能的普及不仅取决于模型质量，也同样取决于获取途径与可用性。"(DeepSeek charges no fees, "lowering the threshold for millions of users, especially in price-sensitive regions." The report added: "This combination of openness and affordability has enabled DeepSeek to gain traction in markets underserved by Western AI platforms. The rise of DeepSeek shows that global AI adoption depends not only on model quality, but equally on access and availability.") (translated from Chinese)}}</ref> The study noted that in countries where Western platforms are blocked or struggle to operate — Russia, Iran, Cuba, Belarus — DeepSeek fills the gap, expanding Chinese influence in regions underserved by U.S. tech monopolies.<ref>{{Web citation|newspaper=新华网 (参考消息)|title=DeepSeek推动全球南方广泛使用人工智能|date=2026-01-12|url=http://www.news.cn/liangzi/20260112/30c68eccd04e4abf9337447f6bb88c49/c.html|retrieved=2026-05-09|quote=报告称："开源人工智能可以作为地缘政治工具发挥作用，在西方平台难以运营的地区扩大中国的影响力。"(The report stated: "Open-source AI can function as a geopolitical tool, expanding China's influence in regions where Western platforms struggle to operate.") (translated from Chinese)}}</ref> | ||
Moreover, open-source models such as Deepseek V4 can easily be downloaded and hosted locally by any country, bypassing potential sanctions, with potential uses in an intranet network. This means that such LLMs are untouchable by the US, whereas they could easily force Anthropic or OpenAI to stop providing API service to certain geographical locations. Instead with open-source, state-of-the-art models such as Deepseek, any country can rehost it for their state or economic needs, as well as further train the model or fine-tune it. In a world where AI is becoming increasingly applied to science and warfare, this further bridges the gap between the strength of the imperial hegemony and periphery countries. | |||
=== Western pushback and bans === | === Western pushback and bans === | ||
Latest revision as of 22:54, 9 May 2026
DeepSeek (深度求索) is a Chinese artificial intelligence startup founded in 2023 by Liang Wenfeng, a former hedge fund manager and Zhejiang University graduate. The company develops large language models (LLMs) that have repeatedly matched or exceeded the performance of the most advanced Western AI systems, while operating under severe U.S. semiconductor sanctions. DeepSeek’s models are released as fully open-source, free to use, and designed to run with extreme efficiency, making frontier AI accessible to the Global South thus closing the technological gap between the imperial core and the periphery.[1]
Origins and early development[edit | edit source]
DeepSeek was launched in May 2023 by Liang Wenfeng, who had previously founded the quantitative hedge fund High-Flyer in 2015.[2] High-Flyer’s early stockpiling of Nvidia A100 chips, amassed before U.S. export bans, provided the computational foundation for DeepSeek’s work.[3] Estimates suggest the firm retained between 10,000 and 50,000 units, enabling it to train foundation models despite escalating restrictions.[3]
Unlike most Chinese AI firms, which are backed by tech giants like Alibaba or ByteDance, DeepSeek is independently financed and has not sought outside funding. A former employee described the working environment as “a luxury that few fresh graduates would get at any company” in terms of compute access and research autonomy.[3]
Technical breakthroughs under U.S. sanctions[edit | edit source]
The United States has imposed a series of escalating chip export controls on China since 2019, aimed at crippling the country’s AI capabilities.[4] However, under sanctions that forced reliance on less powerful Nvidia H800 GPUs (with half the interconnect speed of the top-tier H100), DeepSeek still managed to train and deliver state-of-the-art models that rival the frontier Western models.[5]
In December 2024, DeepSeek released its first model, V3, a 671-billion-parameter model that matched GPT-4o on key benchmarks at a fraction of the training cost.[2] The 2025 R1 reasoning model achieved parity with OpenAI’s o1 at a fraction of the operating cost, and the V3.2-Speciale (December 2025) scored 96.0% on the AIME math exam, beating GPT-5-High’s 94.6%.[6]
In April 2026, the V4-Pro model was released as open-source, with performance rivaling the world’s best closed-source models across math, coding, and agentic tasks.[7]
These results demonstrated the limits of sanctions as a means of maintaining U.S. technological dominance. As one analyst put it, “DeepSeek disproved the assumption that only the U.S., with its immense compute power and capital, could dominate AI.”[8]
Cost efficiency[edit | edit source]
DeepSeek’s efficiency has challenged prevailing assumptions about the economics of AI. V3 was trained on 2,048 H800 GPUs for 2.788 million GPU hours at a total cost of $5.6 million — roughly one-twentieth the estimated cost of GPT-4o.[2][9] The R1 reasoning model’s training cost was reported at just $294,000 — a figure so low it made the cover of Nature.[10] By 2026, DeepSeek’s pricing undercut GPT-5 by a factor of 10: $0.028 per million input tokens versus $1.25 for GPT-5.[11] This makes deepseek highly cost-efficient for sanctioned populations of the Global South and hobbyist projects, as complete coding projects can be completed for as little as 1 USD.[12]
Open source philosophy[edit | edit source]
Where Western AI companies like OpenAI, Google, and Anthropic treat their most powerful models as proprietary “black boxes” and charge steep API fees with rate limits, DeepSeek has released every major model under permissive open-source licenses such as Apache 2.0 and MIT. This allows anyone to view, modify, and commercialize the code for free, or even fine-tune the model and train it further for their own needs.[13] GitHub repositories have accumulated more than 120,000 clones, and a community of over 23,000 developers has built more than 200 vertical-domain models in fields such as medicine, education, and agriculture.[14]
This approach differs from the “closed-source monopoly + capital extraction” model of the West, where, under monopoly capitalism, AI is designed to maximize corporate profit and concentrate wealth.[15] DeepSeek thus embodies an alternative: AI as a public good, where PRC state policy encourages decentralized innovation across the industrial chain rather than enriching a handful of monopolists.[16]
State-directed AI: the “AI Plus” initiative[edit | edit source]
DeepSeek’s development is embedded within a broader national strategy. In August 2025, the State Council issued the "AI Plus" Initiative, outlining a three-step timeline: 70% AI penetration by 2027, 90% by 2030, and a mature "intelligent society" by 2035.[17] The policy prioritises six sectors: science, industry, consumption, public welfare, governance, and global cooperation and explicitly names healthcare, education, and agriculture as core application domains.[18]
By early 2025, China had 439 registered large models covering over 30 industries.[19] Concrete deployments include an AI-assisted doctor project operating in 30 provinces, 377 districts, and 30,000 primary medical institutions, serving 100 million residents, and a smart agriculture platform reaching 15 million farmers across 2,000 counties.[20] The "AI Plus" framework explicitly frames AI as a tool for "people's well-being," aiming to "promote equitable distribution of high-quality education resources" and "provide solutions for issues concerning people's livelihood."[21] The initiative also stresses "technology for good" and international cooperation to bridge the global AI divide.[22]
This state-directed, public-good approach marks a sharp departure from the Silicon Valley venture-capital model, where R&D priorities are dictated by shareholder returns. In China, AI is treated as "civilisational infrastructure," designed to deliver broad-based productivity gains across healthcare, agriculture, and education rather than concentrating wealth in a handful of private firms.[23]
Global impact and geopolitical response[edit | edit source]
Bridging the technology gap in the Global South[edit | edit source]
DeepSeek’s free-to-use chatbot[24] and open-weight models have significantly lowered barriers to AI adoption in developing countries. By treating AI as a public good rather than a proprietary product, China positions itself as a technology partner for the Global South — an alternative to the Silicon Valley model that locks countries into dependency.[25]
A 2026 Microsoft study confirmed the effect: while the AI adoption gap between the Global North and South persists, DeepSeek’s cost-free, open model has "lowered the threshold for millions of users, especially in price-sensitive regions."[26] The study noted that in countries where Western platforms are blocked or struggle to operate — Russia, Iran, Cuba, Belarus — DeepSeek fills the gap, expanding Chinese influence in regions underserved by U.S. tech monopolies.[27]
Moreover, open-source models such as Deepseek V4 can easily be downloaded and hosted locally by any country, bypassing potential sanctions, with potential uses in an intranet network. This means that such LLMs are untouchable by the US, whereas they could easily force Anthropic or OpenAI to stop providing API service to certain geographical locations. Instead with open-source, state-of-the-art models such as Deepseek, any country can rehost it for their state or economic needs, as well as further train the model or fine-tune it. In a world where AI is becoming increasingly applied to science and warfare, this further bridges the gap between the strength of the imperial hegemony and periphery countries.
Western pushback and bans[edit | edit source]
Western governments have responded to the surge of Chinese AI labs and models with restrictions. The U.S. Department of Justice was reportedly investigating whether DeepSeek’s training on Nvidia Blackwell chips violated export controls, while analysts noted that sanctions had stimulated domestic innovation in China.[28] Multiple Western countries — Australia, Germany, and the United States — moved to ban or restrict DeepSeek in official settings, and Microsoft prohibited its own employees from using the service.[29] The bans echo similar suppressions of Chinese tech companies like Huawei, and reflect a deeper conflict: imperialist powers’ resistance to a socialist-oriented, open-source model of development that outcompetes their profit-driven monopolies.
See also[edit | edit source]
References[edit | edit source]
- ↑ Jeffrey Ding (2025-01-15). "Chinese firm’s faster, cheaper AI language model makes a splash" Science. Retrieved 2026-05-09.
- ↑ 2.0 2.1 2.2 "Chinese firm’s faster, cheaper AI language model makes a splash" (2025-01-15). Science. Retrieved 2026-05-09.
- ↑ 3.0 3.1 3.2 "Sanction-Proofing AI: Sophgo, DeepSeek Boost China's Domestic Tech" (2025-07-17). Astute Group. Retrieved 2026-05-09.
- ↑ Jaya Wen (2026-03-12). "How US Trade Sanctions Fueled China's Innovation Surge" Harvard Business School Working Knowledge. Retrieved 2026-05-09.
- ↑ “制裁非但没有削弱中国的人工智能能力，反而似乎推动了像 DeepSeek 这样的初创公司以优先考虑效率、资源共享和协作的方式进行创新。(Rather than weakening China's AI capabilities, the sanctions appear to have driven startups like DeepSeek to innovate in ways that prioritize efficiency, resource sharing, and collaboration.) (translated from Chinese)”
MIT Technology Review (2025-01-27). "中国顶级 AI 模型如何克服美国制裁" Sohu. Retrieved 2026-05-09. - ↑ Blake Crosley (2025-12-10). "DeepSeek V3.2 Beats GPT-5 on Elite Benchmarks" Introl. Retrieved 2026-05-09.
- ↑ "DeepSeek V4 Preview Release" (2026-04-24). DeepSeek API Docs. Retrieved 2026-05-09.
- ↑ “DeepSeek's rise shattered the notion that only the United States, with its immense computational power and capital, could dominate the global AI market. DeepSeek has proved that innovation and efficiency can level the global tech playing field.”
"DeepSeek reflects China's vision for inclusive growth" (2025-05-28). China Daily. Retrieved 2026-05-09. - ↑ "DeepSeek-V3.2 Matches GPT-5 at 10x Lower Cost" (2025-12-02). Introl. Retrieved 2026-05-09.
- ↑ "DeepSeek-R1登Nature封面：29.4万美元训练成本引关注" (2025-09-18). DoNews. Retrieved 2026-05-09.
- ↑ "DeepSeek-V3.2 Matches GPT-5 at 10x Lower Cost" (2025-12-02). Introl. Retrieved 2026-05-09.
- ↑ Considering current pricing policies of Deepseek and that most small projects would rarely require more than 1 million output tokens.
- ↑ “Unlike most AI companies that keep their models behind pay walls, DeepSeek released its R1 model under the MIT License — free to use, modify and even commercialize.”
"DeepSeek reflects China's vision for inclusive growth" (2025-05-28). China Daily. Retrieved 2026-05-09. - ↑ “DeepSeek将核心框架代码完全开源，开发者可基于其架构定制私有化模型。数据显示，开源后社区贡献代码量月均增长120%，形成了一个由2.3万名开发者组成的创新网络。这种"技术普惠+生态共建"的模式，彻底打破了AI技术的垄断格局。(DeepSeek has fully open-sourced its core framework code, allowing developers to customize private models based on its architecture. Data shows that after open-sourcing, community code contributions grew at an average monthly rate of 120%, forming an innovation network of 23,000 developers. This model of "technology inclusion + ecosystem co-construction" has thoroughly broken the monopoly structure of AI.) (translated from Chinese)”
"DeepSeek全民免费时代：国家战略下的AI普惠革命" (2025-09-26). Baidu Developer. Retrieved 2026-05-09. - ↑ “西方以閉源壟斷收割資本...中國的人工智能發展，從頂層設計到技術路線，都和西方完全不同。我們既能抓住科技紅利，又能避免出現富者愈富、貧者愈貧的畸形局面。(The West reaps capital through closed-source monopolies... China's AI development, from top-level design to technical路线, is completely different from the West. We can capture technological dividends while avoiding the畸形 pattern of the rich getting richer and the poor getting poorer.) (translated from Chinese)”
张俊狮 (2026-05-07). "【第三眼觀世界】AI發展之路 中西走向迥異" 香港文匯網. Retrieved 2026-05-09. - ↑ “西方的財權至上思維是：靠閉源築高牆，然後靠壟斷收取高額費，將利潤集中在少數科技大財團之手；而中國的思維是靠開源降低門檻，然後壯大生態，再帶動整條產業鏈，讓更多企業、更多就業崗位受益。(The Western wealth-power-first mindset is: build high walls through closed source, then charge high fees through monopoly, concentrating profits in the hands of a few tech conglomerates; China's approach is to lower barriers through open source, then strengthen the ecosystem and drive the entire industrial chain, benefiting more enterprises and more jobs.) (translated from Chinese)”
"【第三眼觀世界】AI發展之路 中西走向迥異" (2026-05-07). 香港文匯網. Retrieved 2026-05-09. - ↑ “The essence of China's AI blueprint is to define AI as a new generation of "civilizational infrastructure." Its goal is to build symbiosis and balance between AI and technologies, organizations and society, ensuring its development adheres to the principle of "technology for good."”
"China's AI blueprint" (2025-09-22). China's AI blueprint. Retrieved 2026-05-09. - ↑ “The policy document states that China will promote the use of AI in science and technology, industrial development, consumption, people's well-being, governance capability and global cooperation. [...] The policy document also promotes to advance ecological governance through AI-enabled monitoring of air, water, soil, and biodiversity. [...] Globally, it encourages open-source collaboration, ethical AI governance, and international cooperation to bridge the global AI divide.”
"State Council Opinions on Advancing the Comprehensive Implementation of the ‘AI Plus’ Initiative (Guo Fa [2025 No. 11)"]. State Council Opinions on Advancing the Comprehensive Implementation of the ‘AI Plus’ Initiative (Guo Fa [2025] No. 11). Retrieved 2026-05-09. - ↑ “截至2025年上半年，我国大模型备案总量达439款，覆盖医疗健康、农业、教育、智能制造、金融科技等30余个行业，有力推动人工智能创新和应用生态成熟。在医疗领域，智能医生助理项目应用覆盖全国30个省（自治区、直辖市）、377个区县、3万家基层医疗机构、1亿居民；在农业领域，智慧农业平台项目覆盖31省市2千多县农民1500万人，累计服务60亿次。(As of the first half of 2025, China had 439 registered large models covering over 30 industries including healthcare, agriculture, education, smart manufacturing, and financial technology, vigorously推动 the maturation of AI innovation and application ecosystems. In the medical field, the AI-assisted doctor project covers 30 provinces, 377 districts, 30,000 primary medical institutions, and 100 million residents nationwide; in agriculture, the smart agriculture platform reaches 15 million farmers across 2,000 counties in 31 provinces, with a cumulative 6 billion services.) (translated from Chinese)”
"以技术持续创新支撑“人工智能+”行动加速落地" (2025-08-27). 国家发展改革委. Retrieved 2026-05-09. - ↑ "以技术持续创新支撑“人工智能+”行动加速落地" (2025-08-27). 国家发展改革委. Retrieved 2026-05-09.
- ↑ "State Council Opinions on Advancing the Comprehensive Implementation of the ‘AI Plus’ Initiative (Guo Fa [2025 No. 11)"]. State Council Opinions on Advancing the Comprehensive Implementation of the ‘AI Plus’ Initiative (Guo Fa [2025] No. 11). Retrieved 2026-05-09.
- ↑ "China's AI blueprint" (2025-09-22). China's AI blueprint. Retrieved 2026-05-09.
- ↑ “The essence of China's AI blueprint is to define AI as a new generation of "civilizational infrastructure." Its goal is to build symbiosis and balance between AI and technologies, organizations and society, ensuring its development adheres to the principle of "technology for good." The AI Plus initiative aims to seize the opportunities presented by AI and ensure they are channeled into satisfying people's needs for a better quality of life. It will do so by promoting the in-depth application of AI to people's daily life, including in the healthcare, education, transportation and elderly care sectors.”
"China's AI blueprint" (2025-09-22). China's AI blueprint. Retrieved 2026-05-09. - ↑ On https://chat.deepseek.com
- ↑ “By treating AI as a public good rather than a proprietary product, China is establishing itself as the technology partner of choice for Global South nations, offering an alternative to Silicon Valley's profit-driven model that locks countries into dependency. China's AI journey underscores a vital lesson: technology should be a bridge, not a barrier.”
"China's AI approach offers the Global South partnership over dependency" (2026-01-09). chinadiplomacy.org.cn. Retrieved 2026-05-09. - ↑ “DeepSeek不收取费用，"为数百万用户降低了门槛，特别是在对价格敏感的地区"。报告补充说："这种开放性和可负担性的结合，使DeepSeek在西方人工智能平台服务不足的市场获得了关注。DeepSeek的崛起表明，全球人工智能的普及不仅取决于模型质量，也同样取决于获取途径与可用性。"(DeepSeek charges no fees, "lowering the threshold for millions of users, especially in price-sensitive regions." The report added: "This combination of openness and affordability has enabled DeepSeek to gain traction in markets underserved by Western AI platforms. The rise of DeepSeek shows that global AI adoption depends not only on model quality, but equally on access and availability.") (translated from Chinese)”
"DeepSeek推动全球南方广泛使用人工智能" (2026-01-12). 新华网 (参考消息). Retrieved 2026-05-09. - ↑ “报告称："开源人工智能可以作为地缘政治工具发挥作用，在西方平台难以运营的地区扩大中国的影响力。"(The report stated: "Open-source AI can function as a geopolitical tool, expanding China's influence in regions where Western platforms struggle to operate.") (translated from Chinese)”
"DeepSeek推动全球南方广泛使用人工智能" (2026-01-12). 新华网 (参考消息). Retrieved 2026-05-09. - ↑ "China's DeepSeek trained AI model on Nvidia's best chip despite US ban, official says" (2026-02-24). The Telegraph India. Retrieved 2026-05-09.
- ↑ "DeepSeek推动全球南方广泛使用人工智能" (2026-01-12). 新华网 (参考消息). Retrieved 2026-05-09.
