Introduction

On 20 January 2025, the same day as the inauguration of United States (US) President Donald Trump, China released DeepSeek-R1, a reasoning model that claimed to be on a par with industry-leading models from the US.^[1] Shortly after, downloads of the DeepSeek mobile app skyrocketed on the Apple App Store^[2] while sending the US stock market into a tumble with major technology stocks—particularly that of NVIDIA—taking a nosedive and losing nearly US$600 billion worth of its market value. It was the biggest fall recorded in US stock market history, with other technology companies like Alphabet and Microsoft losing US$100 billion and US$7 billion, respectively. In sum, the technology-heavy Nasdaq Composite Index lost over US$1 trillion over the course of a week.^[3]

The ensuing sensation and market disruption that DeepSeek created was on two fronts: compute economics and an open-weight model, combined with its achievement of industry-competitive performance in the face of stringent export controls aimed at slowing down China’s progress in Artificial Intelligence (AI).^[4] The combined effect of these innovations and their accessibility and adaptability changes the way foundational models spur downstream innovations across industries and geographies, leading to higher levels of diffusion and lower barriers to entry. Hyperscalers such as Google, Amazon Web Services (AWS), and Microsoft were quick to include DeepSeek within their suite of services despite claims of intellectual property rights violation by OpenAI.^[5]

Deepseek has seen exponential diffusion since its release, reaching 96.88 million monthly active users worldwide in April 2025. This reflected a 25.82-percent increase in just over a month.^[6] China, India, and Indonesia comprised its top three markets with over 51.24 percent of monthly active users worldwide within the month of its release.^[7] In addition, Deepseek dominates in markets like Russia, where ChatGPT cannot operate.^[8] Notably, India is the second-largest user base for DeepSeek worldwide, after China, constituting 20 percent of monthly active users.^[9] Its growing global diffusion has been a wake-up call for the US:^[10] its AI Action Plan, released on 23 July 2025, contains a section on the need to focus on encouraging open-source and open-weight models to promote higher diffusion and the adoption of US technology, combined with the need to “improve the financial market for compute.”^[11] The levels of market penetration exemplified by DeepSeek have made China a strong contender in the global AI race, challenging the US’s long-held dominant position in the domain.

To be sure, China’s DeepSeek moment unfolds within the wider context of its military might, manufacturing prowess, and robust network of regional and international institutions. The latter includes the Belt and Road Initiative (BRI), the Asian Infrastructure Investment Bank, and the Digital Silk Road initiative, which are working to expand the country’s influence in other parts of the world. This is combined with its supremacy in 5G technologies, which makes China a formidable tech leader on the world stage.^[12]

However, antithetical to private sector and consumer uptake, DeepSeek has faced bans and restrictions in several countries.^[13] Italy was the first European country to ban DeepSeek, citing a failure to cooperate regarding information sharing,^[14] while Australia expressed concerns about risks to national security and government systems.^[15] South Korea temporarily suspended its downloads in the middle of February, highlighting security concerns and requirements under the country’s data protection regime. The app became available to download again in April after “at least partially reflecting” recommendations of the national data protection agency.^[16] Taiwan has taken a step further by prohibiting the use of DeepSeek in all government agencies, framing the controversy within the broader context of Beijing's sovereignty claims.^[17] Taiwan's Ministry of Digital Affairs issued a warning that DeepSeek "endangers national information security" due to cross-border data transmission and information leakage.

In the US, Texas has taken the lead in prohibiting the use of the tool on government devices. Governor Greg Abbott specifically linked it to national security, stating, “Texas will not allow the Chinese Communist Party to infiltrate our state's critical infrastructure through data-harvesting AI and social media apps.”^[18] The Department of Defense, the US Navy, National Aeronautics and Space Administration (NASA), and the House of Representatives have also disallowed its use on official devices and networks.^[19]

AI innovation in different geographies has been increasingly inflected by the global distribution of talent, computing resources, foundational models, and finance. This spatial complexity has introduced dimensions of national and economic security into the cascading effects of competition that distort markets, intensifies rationales for technological sovereignty, and reinforces newer dimensions of dependency given the rapid diffusion of innovations like DeepSeek. Its accelerated market penetration raises questions about the architecture of future global innovation and these broader multiple implications due to the intensifying competition between the two AI superpowers (the US and China) and the structure of global economic and geopolitical dependencies.

This brief analyses how DeepSeek expands the adoption of Chinese technology and identifies broader implications for global innovation. It unpacks the narratives around DeepSeek’s technological capabilities and the inherent risk within the model itself. It then shows the strategic implications in terms of how DeepSeek has not only energised China’s AI ecosystem and localisation strategy but has also become an important offering within the context of its global technology expansion.

DeepSeek and Its Discontents

On 20 January 2025, DeepSeek released its flagship reasoning model, DeepSeek R1, an MIT-licensed^[1] open-weight model intended to rival those from OpenAI’s ChatGPT. DeepSeek R1, geared towards advanced reasoning, built on the foundations of the general purpose model, DeepSeek V3.^[20] In December 2024, the latter was achieving comparable performance with models like ChatGPT 4.0 and Claude 3.5 Sonnet.^[21] DeepSeek was founded by Liang Wenfeng and backed by his quant fund, High Flyer Capital Management^[22] which, though a private company, also has deep ties to the Chinese government and military establishment. The company’s researchers have worked on nearly 400 AI projects by the People’s Liberation Army (PLA), and have either past or current affiliations with 42 government talent recruitment programmes and 375 government entities—all signalling extensive government funding.^[23]

Since its release, DeepSeek has either been largely hailed as a breakthrough innovation or dismissed as an unsurprising development within the context of overall scientific progress and increasing algorithmic and compute efficiency. As is the case with most innovations, its abilities lie somewhere in between. It contains important innovations like the use of pure reinforcement learning without any human supervision, as well as architectural changes like the multi-token prediction and multi-latent attention as well as other hardware optimisations.^[24] However, DeepSeek is as much about an easily adoptable innovation as it is about the narrative, implications, and consequences it engenders for China, global AI innovation, and the geopolitics implicated therein.

The DeepSeek success story becomes more plausible given the substantial amount of time that High Flyer spent experimenting and applying foreign AI models to their businesses while investing in high-end NVIDIA chips to support its AI transformation strategy.^[25] This includes two AI supercomputing clusters comprised entirely of NVIDIA A100 chips, which were eventually subject to export controls in 2022. DeepSeek’s 10000 A100 cluster is said to be Asia’s first.^[26] The first cluster of 1100 chips cost 200 million yuan, while the second cluster of 10000 chips cost 1 billion yuan. It is unclear how much High Flyer has invested in DeepSeek, with Liang having 55 percent of the stake and 99 percent voting rights. In 2022, High Flyer’s researchers claimed to have maximised the second cluster’s efficiency when training AI models as presented at an NVIDIA conference.^[27]

While DeepSeek claimed to use the lower H800 in training its model, reports claim that it has far more computing power than is being disclosed due to its potential violation of export controls.^[28] Its existence has also been attributed to gaps in export controls and circumvention strategies to access the requisite chips despite the October 2022 ban, owing to slow regulatory updates and an inability to plug the flow of chips from manufacturers like TSMC.^[29] In October 2022, export control on advanced chips was based on performance thresholds across two metrics: total processing power and interconnect speed.^[30] At the time, such criteria would apply to NVIDIA’s A100 and H100 chips. However, as a result of post-manufacturing modification capabilities to downgrade the performance of their A100 chips, NVIDIA reduced interconnect speed but not processing power. This led to the creation of the A800 line, which could be exported to China. The H800s were a degraded version of the H100 chips made for the Chinese market. These modifications allowed the continued sale of advanced chips to China while remaining within the ambit of export control regulations. While this loophole was realised in December 2022, the ban on A800 and H800 only came into effect in October 2023.

Additionally, US officials have alleged that DeepSeek sought to use Southeast Asian shell companies to access advanced semiconductors that were subject to export control.^[31] However, the demand for smuggled chips began after the October 2023 tightening of export control rules, with evidence of smuggling transactions totalling over US$100 million.^[32] DeepSeek’s leadership has openly acknowledged export controls as a critical handicap as the compute gap continues to widen between the US and China amid more stringent export controls.^[33]

Yet, DeepSeek is not merely a consumer-range AI product but a strategic benchmark in China's long-term vision to become a technology and innovation-driven economic superpower. With export restrictions on AI chips, trade embargoes, and shifting geopolitical limitations, DeepSeek is serving to signal China’s undeterred prowess in AI. The timing of the release of DeepSeek R1, coinciding with the inauguration of the US president and subsequent US stock market disruption, could be seen as a response to the counter-disruptions that China can cause in retaliation to the US’s chokepoint strategy of throttling advanced chips supply to China. In addition to the techno-economic retaliation, combined with the narrative of breakthrough Chinese innovation, its release also has serious geopolitical and security implications.

A key part of DeepSeek’s narrative was the comparative economy of training the model at less than US$6 million—a fraction of the cost of LLMs available today. In comparison, Claude 3.5 Sonnet by Anthropic, for example, cost “$10s of millions to train.”^[34] However, industry analysis shows that the total server capital expenditure would be around ~US$1.6 billion, along with a US$944 million cost associated with operating GPU clusters.^[35] The ~US$6 million figure attributable to the “official training” costs does not take into account the cost of research and development, architectural innovation, and the recruitment of top talent at attractive salaries.^[36] According to estimates, hardware spend could be well over US$500 million over the life of the firm.^[37]

Another aspect of DeepSeek’s narrative has been the openness of the model, which has led to widespread adoption. It is open in the sense that it makes its model weights or trained parameters and architecture available under permissive licences like the MIT licence. However, it does not release the source code or training data to replicate the model from scratch, with limited clarity on compute and data trade-offs.^[38] This mode of openness allows users to download, run, fine-tune, and analyse the model but not replicate it from scratch. Efforts to replicate the model will likely involve curating new large-scale datasets for specific problems like maths, coding, and reasoning.^[39] Though it does not fall within the strictest definition of open-source,^[40] its accessibility and adaptability lead to high levels of adoption.

Its innovation breakthrough was challenged when OpenAI claimed that DeepSeek used a technique known as ‘distillation’ to build a competitive product.^[41] It is a common method of teaching a smaller model to replicate the capabilities of a bigger model by querying the model and learning from its output.^[42] However, OpenAI claims that DeepSeek misused the method by querying ChatGPT at scale to build and train a competitive product. It launched an investigation to understand whether DeepSeek used its proprietary models or obtained access to APIs that enabled it to query ChatGPT at scale. If this were true, it would constitute a breach of contract in the form of a violation of its terms of use, which expressly prohibit using its outputs to develop a competitive product.^[43]

On the security front, Cisco security researchers had highlighted that DeepSeek R1 failed to block any harmful prompts, demonstrating a 100-percent success rate in jailbreaking the model.^[44] Its innovations could have ironically compromised its safety features and did not block a single harmful prompt from the HarmBench benchmark.^[45] Though, notably, other models like “Meta’s Llama-3.1-405B and OpenAI’s GPT-4o also had high failure rates of 96 percent and 86 percent, respectively,”^[46] in a safety test run by Anthropic, it found that DeepSeek generated “rare information” on bioweapons with “absolutely no blocks whatsoever.”^[47]

Bias testing on DeepSeek showed high refusal rates and bias when queried on 12 geopolitical incidents, like the 1989 Tiananmen Square, India-Pakistan conflict, the Kargil war, as well as the semiconductor ecosystem in Taiwan.^[48] Across these incidents, the test included 300+ prompts with DeepSeek chat showing high censorship at 88 percent and pro-China responses to incidents like the Sino-Indian war and the one-child policy.^[49] DeepSeek R1 showed 91.2 percent pro-China bias with 114 out of 125 answers leaning overtly in favour of China.^[50]

The tool’s embeddedness with the Chinese government, the gap between hype versus reality, its innovation roadmap, and inherent risks highlight both its role in China’s AI sovereignty and the modalities of its global diffusion.

Strategic Technology and Leverage: Corporations, Nation-States, and Those In-Between

The implications of DeepSeek are not only commercial, as AI is at the centre of military and defence strategy, financial prediction, and healthcare innovation. It has since been deployed for research and development (R&D) for China’s most advanced warplanes, with BMW planning to incorporate DeepSeek into its vehicles in China.^[51] Its compute efficiency gains and open accessibility have spurred increased adoption. DeepSeek also caused Chinese stocks to rally, with MSCI China surging 21 percent from its low in January. A number of corporations across sectors have adopted the model because of its open architecture. The Chinese government accepts DeepSeek's breakthroughs as a direct response to US efforts to control the country’s technological progress through the limitation on the export of chips. In an article published in February this year, Defense One quoted Hu Xijin, former editor-in-chief of Global Times, as saying that DeepSeek's success was a vindication of Chinese technological progress against the US sanctions.^[52]

The same report said that by demonstrating the capacity to innovate, regardless of Western technologies, China indicates its strength in the AI industry and its determination to shift the global AI power balance. The development of DeepSeek aligns with Beijing's overall strategy of using AI to achieve technological independence and ideological leverage. The latter is a direct consequence of the model’s pro-China bias and adherence to state-mandated controls on content. Chinese Ministry of Industry and Information Technology specialist Pan Helin describes DeepSeek as a demonstration of China's capacity to develop independently, thereby lessening its dependency on US semiconductor value chains.^[53]

Indeed, DeepSeek presents a strategic advantage to the armed forces of China as it is founded on widely accessible consumer technology. The platform makes it possible to develop sophisticated AI systems regardless of expensive Western technologies, particularly under US semiconductor export restrictions. This aligns with China's vision to integrate AI into its defence system. This ability enhances combat analysis and decision-making in real time. Chinese defence analysts posit that the country’s AI capabilities may enhance military decision-making, streamline autonomous systems, and enable real-time combat monitoring.^[54] The integration of generative AI into China's defence system can result in previously unknown levels of cyber threat detection, intelligence gathering, and command-and-control capabilities.

Beijing is also determined to shape global regulatory standards and counter the sweeping Western dominance in AI technologies.^[55] DeepSeek has spurred adoption across the public and private sectors in China, with applications across government services, automobiles, smartphones, consumer electronics, and healthcare.^[56] The country’s financial sector has rapidly deployed DeepSeek's AI algorithms. Brokers like Tiger Brokers have adopted DeepSeek's R1 model in their chatbots to enhance value in trading insights and money analysis.^[57] Over 20 companies, like China Universal Asset Management and CICC Wealth Management, use DeepSeek's technology to transform client interactions, investment propositions, risk analysis, and research. The technology enables businesses to efficiently process large volumes of data, gain in-depth market insight, and make improved decisions, thereby enhancing competitiveness and profitability. DeepSeek has provided a way for Chinese firms to establish AI sovereignty by relying on its indigenous models, precluding the necessity to rely on Western ones.

China has managed to build an entire ecosystem to support the development of AI within the country—this spans access to financing, central and state government subsidies, and the establishment of 48 data exchanges and authorised marketplaces that enable AI companies to buy large datasets with regulatory oversight.^[58]

While DeepSeek has enabled China to raise its own sovereign capabilities, it could reinforce structures of dependency through its potential introduction into global infrastructure projects, which would extend China's strategic reach.^[59] With projects such as the BRI, China has extended its connectivity and infrastructure presence over 20,000 projects in 150 countries with combined public-private investment amounting to US$1.174 trillion between 2013-2024.^[60] The Digital Silk Road (DSR)—one of the BRI’s major sub-initiatives is structured to promote ICT (Information and Communication Technology) connectivity, digital cooperation and infrastructure development through agreements with over 40 countries.^[61] The DSR has helped China facilitate the global market expansion of its technology corporations, allowing it to open up markets and digital free trade zones.^[62] Further, if DeepSeek becomes entrenched in critical sectors like finance, logistics, healthcare, and governance, China may control AI standards and protocols and gather economic intelligence in ways that might not align with the sovereign interests of individual states.^[63]

This would not only increase China's control over the digital economy but also give Beijing more access to global data streams. Historically, American AI policy has promoted rapid innovation while exercising regulatory levers to contain China's technological reach. DeepSeek's success in the face of such limitations, however, points to a larger trend in AI development, one that prioritises resilience and adaptability over raw computational power. President Trump's declaration that DeepSeek's success was a wake-up call for American technology companies reflects the growing importance of AI in geopolitical competition, where AI leadership is directly linked to military progress, cyber capabilities, and economic reach.^[64]

Additionally, DeepSeek is vulnerable to censorship policies supporting China's strategic narratives. Unlike OpenAI, which is trained on the open internet, DeepSeek is trained on Chinese social media platforms and government-approved data sources.^[65] This enables it to extend its digital “walled garden” to other geographies. It thus raises the possibility of digital propaganda and the wider weaponisation of AI in influencing public opinion.

DeepSeek’s widespread diffusion, its bias potential, and inclusion within China’s global infrastructure projects has significant implications for global AI governance arrangements and the nature of norms, standards, and protocols that may determine how AI is used.^[66] DeepSeek, as an instrument of wider Chinese foreign policy of global technological expansion, has manifold implications for how global innovation can be structured, wherein the pursuit of AI sovereignty by one nation creates structures of dependency in adopting countries, particularly within the wider context of economic dependency and development.

Conclusion: Implications for Global Innovation

AI capabilities increasingly shape geopolitical strategy and economic security, with the ability to determine the “rise and fall of nations.”^[67] Individual AI capabilities, their potential for transformative change, and the infrastructural requirements essential to implementing them are increasingly becoming the determinants of global competition. This has led to the logic of regulating critical resources like advanced chips and critical minerals, which are arbiters of AI development. The global technology and innovation landscape is shaped by such rationales and responses. The regulation of critical resources within the context of global technological competition between two AI superpowers has prompted questions around sovereignty, economic security, and the fragility of the existing distribution of power.

As the great-power dynamics in AI play out between the US and China, the vulnerabilities within global economic and technological dependencies are getting revealed more clearly.^[68] It has highlighted an increasingly important need to navigate an ambiguous strategic environment characterised by abrupt political changes and cascading economic vulnerabilities and risks.^[69]

Global priorities in innovation will be determined by the trade-offs that countries attempt to make if they seek to reduce dependence on dominant allies such as the US and China. This requires careful calibration, as it could affect economic and strategic relationships within the short term, with key partners likely to retaliate by reducing cooperation and even inviting backlash and counteractions which could result in reduced access to resources or support in critical scenarios.^[70]

If countries are to cultivate strategic autonomy within the global AI world order, they would then require the development of robust national capabilities, derisking supply chains, and building strong AI ecosystems that can serve as the fulcrum of securing their position within the globally competitive landscape.^[71] Although some countries are pursuing the development of their indigenous models, DeepSeek’s efficiency, accessibility, and low cost have paved the way for rapid adoption across a range of organisations while reducing the barriers for building smaller and bespoke models.^[72] More than half (58 percent) of new startups cite DeepSeek as a part of their infrastructure stack, with reported wider enterprise adoption.^[73] Its increased market penetration would hinder the adoption of domestic models over time due to the increase in the switching costs of migrating workflows driven by DeepSeek to another platform.

While DeepSeek might reduce the costs of AI integration and adoption across different segments of the economy and society, it will eventually stymie countries’ domestic innovation. Given the data protection and cybersecurity risks, countries should carefully evaluate the concerns around bias, safety, and security.^[74] China’s tech and economic expansion, coupled with the economic dependency it creates among countries in the Global South, and the enormous volumes of information it is now able to collect, requires careful consideration of the ease of adoption versus implications for domestic innovation. Overall, DeepSeek serves to catalyse China’s global tech and market reach, affecting the structure of global innovation in profound ways.

---

Anulekha Nandi is Fellow, Centre for Security, Strategy and Technology, ORF.

Shreya Balasubramani is former Intern, ORF.

---

All views expressed in this publication are solely those of the authors, and do not represent the Observer Research Foundation, either in its entirety or its officials and personnel.

Endnotes

^[1] MIT- license is a permissive software license originating in the Massachusetts Institute of Technology (MIT) with conditions requiring the preservation of copyright and license notices (https://github.com/deepseek-ai/DeepSeek-V3/blob/main/LICENSE-CODE)

^[1] International Institute for Strategic Studies, DeepSeek’s Release of an Open-Weight Frontier AI Model, Strategic Comments, 2025, https://www.iiss.org/publications/strategic-comments/2025/04/DeepSeeks-release-of-an-open-weight-frontier-ai-model/

^[2] Maxwell Zeff, “DeepSeek Displaces ChatGPT as the App Store’s Top App,” TechCrunch, January 27, 2025, https://techcrunch.com/2025/01/27/DeepSeek-displaces-chatgpt-as-the-app-stores-top-app/

^[3] Dan Milmo, Amy Hawkins, Robert Booth and Julia Kollewe, “‘Sputnik Moment’: $1Tn Wiped Off US Stocks After Chinese Firm Unveils AI Chatbot,” The Guardian, https://www.theguardian.com/business/2025/jan/27/tech-shares-asia-europe-fall-china-ai-DeepSeek

^[4] Sinéad Carew, Amanda Cooper, and Ankur Banerjee, “DeepSeek Sparks AI Stock Selloff; Nvidia Posts Record Market-Cap Loss,” Reuters, January 28, 2025, https://www.reuters.com/technology/chinas-deepseek-sets-off-ai-market-rout-2025-01-27/; John Villasenor, “DeepSeek Shows the Limits of US Export Controls on AI Chips,” Brookings, January 29, 2025, https://www.brookings.edu/articles/deepseek-shows-the-limits-of-us-export-controls-on-ai-chips/

^[5] Georgia Butler, “US and Chinese Cloud Providers Offer Access to DeepSeek Models,” Data Centre Dynamics, February 4, 2025, https://www.datacenterdynamics.com/en/news/us-and-chinese-cloud-providers-offer-access-to-DeepSeek-models/; Kyle Wiggers, “Microsoft Brings a DeepSeek Model to its Cloud,” TechCrunch, January 29, 2025, https://techcrunch.com/2025/01/29/microsoft-brings-a-DeepSeek-model-to-its-cloud/

^[6] “DeepSeek AI Usage Stats,” Backlinko, May 27, 2025, https://backlinko.com/DeepSeek-stats

^[7] “DeepSeek AI Usage Stats”

^[8] Mary Meeker, Jay Simons, Daegwon Chae, and Alexander Krey, Trends – Artificial Intelligence (AI), Bond

Capital, 2025.

^[9] Barry Elad, “DeepSeek AI Statistics 2025: Users, Benchmarks & Enterprise Reach,” SQ Magazine, July 22, 2025, https://sqmagazine.co.uk/DeepSeek-ai-statistics/

^[10] David Ingram, “Trump Says China's DeepSeek AI 'Should be a Wake-Up Call' for American Tech Companies,” NBC News, January 28, 2025, https://www.nbcnews.com/tech/innovation/trump-china-DeepSeek-ai-wake-call-rcna189526

^[11] Executive Office of the President of the United States, Winning the Race: America’s AI Action Plan, 2025, https://www.whitehouse.gov/wp-content/uploads/2025/07/Americas-AI-Action-Plan.pdf

^[12] Hoang Le Thai Vu, Lan Di Ngo, and Thanh Tien Nguyen, “A Spectrum of Autonomy: Towards a Theoretical Framework of Strategic Autonomy,” International Journal, vol. 79, no. 2 (2024): 230-249

^[13] Kriti Barua, “DeepSeek Banned: List of Countries and Government Agencies That Have Imposed Restrictions on This Chinese AI Model.” Jagranjosh, February 6, 2025. https://www.jagranjosh.com/general-knowledge/list-of-countries-and-govt-agencies-which-banned-the-use-of-DeepSeek-ai-1738828041-1

^[14] “Italy Becomes First Country to Ban DeepSeek, Cites Privacy Concerns Over Personal Data Usage,” Financial Express, January 31, 2025, https://www.financialexpress.com/world-news/italy-becomes-first-country-to-ban-deepseek-country-cites-privacy-concerns-over-personaldatausage/3731852/

^[15] Tom Gerken, “Australia Bans DeepSeek on Government Devices Over Security Risk,” BBC, February 4, 2025, https://www.bbc.com/news/articles/c8d95v0nr1yo

^[16] Barua, “DeepSeek Banned”; “DeepSeek Available to Download Again in South Korea After Suspension,” Reuters, May 1, 2025, https://www.reuters.com/sustainability/boards-policy-regulation/DeepSeek-available-download-again-south-korea-after-suspension-2025-04-28/

^[17] “Taiwan Bars Govt Agencies, Service Providers from Using DeepSeek AI Services,” Hindustan Times, February 1, 2025, https://www.hindustantimes.com/world-news/taiwan-bars-govt-agencies-service-providers-from-using-DeepSeek-ai-services-101738400151844.html

^[18] “Texas Becomes the First State in the US to Ban Chinese AI Sensation DeepSeek After App Reaches Top of Apple Store,” The Economic Times, February 4, 2025,  https://economictimes.indiatimes.com/news/international/us/texas-becomes-the-first-state-in-the-u-s-to-ban-chinese-ai-sensation-DeepSeek-after-app-reaches-top-of-apple-store/articleshow/117895429.cms

^[19] Nooree Lee et al., “U.S. Federal and State Governments Moving Quickly to Restrict Use of DeepSeek,” Global Policy Watch.

^[20] Elie Bakouch, Leandro von Werra, and Lewis Tunstall, “Open-R1: A Fully Open Reproduction of DeepSeek-R1,” Hugging Face, January 28, 2025, https://huggingface.co/blog/open-r1

^[21] Supreeth Koundinya, “DeepSeek-V3 is Now The Best Open Source AI Model,” Analytics India Magazine, December 26, 2024, https://analyticsindiamag.com/ai-news-updates/DeepSeek-v3-is-the-best-open-source-ai-model/

^[22] Eduardo Baptista, “High-Flyer, the AI Quant Fund Behind China's DeepSeek,” Reuters, May 1, 2025,  https://www.reuters.com/technology/artificial-intelligence/high-flyer-ai-quant-fund-behind-chinas-DeepSeek-2025-01-29/

^[23] DeepSeek’s Deception: How the Chinese Military and Government Funded DeepSeek’s AI Research, Exiger, https://www.exiger.com/perspectives/DeepSeek-chinese-government-funded-ai/

^[24] Bakouch, von Werra, and Tunstall, “Open-R1: A Fully Open Reproduction of DeepSeek-R1”

^[25] Baptista, “High-Flyer, the AI Quant Fund Behind China's DeepSeek”

^[26] Lennart Heim, “The Rise of DeepSeek: What the Headlines Miss,” RAND, January 28, 2025, https://www.rand.org/pubs/commentary/2025/01/the-rise-of-DeepSeek-what-the-headlines-miss.html?ref=blog.heim.xyz#disruptions5; Baptista, “High-Flyer, the AI Quant Fund Behind China's DeepSeek”

^[27] Baptista, “High-Flyer, the AI Quant Fund Behind China's DeepSeek”

^[28] Karen Freifeld, “Exclusive: TSMC Told US of Chip in Huawei Product after TechInsights Finding, Source Says,” Reuters, October 23, 2024, https://www.reuters.com/technology/tsmc-told-us-chip-huawei-device-after-techinsights-finding-source-says-2024-10-22/

^[29] Ashley Lin and Lennart Heim, “DeepSeek's Lesson: America Needs Smarter Export Controls,” RAND, February 5, 2025, https://www.rand.org/pubs/commentary/2025/02/DeepSeeks-lesson-america-needs-smarter-export-controls.html

^[30] Gregory C. Allen, DeepSeek, Huawei, Export Controls, and the Future of the U.S.-China AI Race, Center for Strategic and International Studies, 2025, https://www.csis.org/analysis/DeepSeek-huawei-export-controls-and-future-us-china-ai-race

^[31] Michael Martina and Stephen Nellis, “Exclusive: DeepSeek Aids China's Military and Evaded Export Controls, US Official Says,” Reuters, June 23, 2025, https://www.reuters.com/world/china/DeepSeek-aids-chinas-military-evaded-export-controls-us-official-says-2025-06-23/

^[32] “DeepSeek, Huawei, Export Controls, and the Future of the U.S.-China AI Race”

^[33] Heim, “The Rise of DeepSeek: What the Headlines Miss”

^[34] Dylan Patel, AJ Kourabi, Doug Olaughlin, and Reyk Knuhtsen, “DeepSeek Debates: Chinese Leadership On Cost, True Training Cost, Closed Model Margin Impacts,” SemiAnalysis, January 31, 2025,  https://semianalysis.com/2025/01/31/DeepSeek-debates/

^[35] Patel, Kourabi, Olaughlin, and Knuhtsen, “DeepSeek Debates”

^[36] Patel, Kourabi, Olaughlin, and Knuhtsen, “DeepSeek Debates”; Hayden Field, “DeepSeek’s Hardware Spend Could be as High as $500 Million, New Report Estimates, CNBC, January 31, 2025, https://www.cnbc.com/2025/01/31/DeepSeeks-hardware-spend-could-be-as-high-as-500-million-report.html

^[37] Patel, Kourabi, Olaughlin, and Knuhtsen, “DeepSeek Debates”

^[38] Bakouch, von Werra, and Tunstall, “Open-R1: A Fully Open Reproduction of DeepSeek-R1”

^[39] Bakouch, von Werra, and Tunstall, “Open-R1: A Fully Open Reproduction of DeepSeek-R1”

^[40] Daniel Nest, “Just How Open Is "Open-Source" AI?,” Why Try AI, February 6, 2025, https://www.whytryai.com/p/open-source-vs-closed-source-ai

^[41] Maria Paulina Muñoz, “The Innovation Dilemma: AI Distillation in OpenAI v. DeepSeek,” UC Berkeley Law, March 30, 2025, https://sites.law.berkeley.edu/thenetwork/2025/03/30/the-innovation-dilemma-ai-distillation-in-openai-v-DeepSeek/

^[42] Muñoz, “The Innovation Dilemma,”

^[43] Meesha Reiisieh, “Code, Claims, and Consequences: The Legal Stakes in OpenAI’s Case Against DeepSeek,” Santa Clara Business Law Chronicle, February 12, 2025, https://www.scbc-law.org/post/code-claims-and-consequences-the-legal-stakes-in-openai-s-case-against-DeepSeek

^[44] Paul Kassianik and Amin Karbasi, “Evaluating Security Risk in DeepSeek and Other Frontier Reasoning Models,” CISCO, January 31, 2025, https://blogs.cisco.com/security/evaluating-security-risk-in-DeepSeek-and-other-frontier-reasoning-models

^[45] Kassianik and Karbasi, “Evaluating Security Risk in DeepSeek and Other Frontier Reasoning Models”

^[46] Charles Rollet, “Anthropic CEO Says DeepSeek Was ‘The Worst’ on a Critical Bioweapons Data Safety Test,” TechCrunch, February 7, 2025, https://techcrunch.com/2025/02/07/anthropic-ceo-says-DeepSeek-was-the-worst-on-a-critical-bioweapons-data-safety-test/

^[47] Rollet, “Anthropic CEO Says DeepSeek Was ‘The Worst’ on a Critical Bioweapons Data Safety Test”

^[48] “DeepSeek Under Fire: Uncovering Bias & Censorship from 300 Geopolitical Questions,” Enkrypt AI, January 30, 2025, https://www.enkryptai.com/blog/DeepSeek-under-fire-uncovering-bias-censorship-from-300-geopolitical-questions

^[49] “DeepSeek Under Fire: Uncovering Bias & Censorship from 300 Geopolitical Questions”

^[50] “DeepSeek Under Fire: Uncovering Bias & Censorship from 300 Geopolitical Questions”

^[51] Supreeth Koundinya, “New DeepSeek-R1 Is as Good as OpenAI o3 and Gemini 2.5 Pro,” Analytics India Magazine, May 29, 2025, https://analyticsindiamag.com/ai-news-updates/new-DeepSeek-r1-as-good-as-openai-o3-and-gemini-2-5-pro/

^[52] Tye Graham and Peter W, Singer, “To China, DeepSeek Is More Than an App-It’s a Strategic Turning Point,” Defense One, February 21, 2025.

^[53] Graham and Singer, “To China, DeepSeek Is More Than an App-It’s a Strategic Turning Point”

^[54] Graham and Singer, “To China, DeepSeek Is More Than an App-It’s a Strategic Turning Point”

^[55] Raphael Racicot and Kurtis H. Simpson, “China’s AI Governance Initiative and Its Geopolitical Ambitions,” Centre for International Governance Innovation, July 22, 2025, https://www.cigionline.org/articles/chinas-ai-governance-initiative-and-its-geopolitical-ambitions/; Matt Sheehan and Jacob Feldgoise, “What Washington Gets Wrong About China and Technical Standards,” Carnegie Endowment, February 27, 2023, https://carnegieendowment.org/research/2023/02/what-washington-gets-wrong-about-china-and-technical-standards?lang=en

^[56] Anniek Bao, “Beijing Embraces DeepSeek to Lead AI Adoption as it Looks for New Growth Drivers,” CNBC, February 21, 2025, https://www.cnbc.com/2025/02/21/deepseek-led-ai-adoption-offers-china-opportunity-to-boost-growth.html; Kinling Lo, “China’s AI Frenzy: DeepSeek is Already Everywhere — Cars, Phones, Even Hospitals,” Rest of World, March 13, 2025, https://restofworld.org/2025/china-embeds-deepseek-ai-in-everything/

^[57] Samuel Shen and Tom Westbrook, “Tiger Brokers Adopts DeepSeek Model as Chinese Brokerages, Funds Rush to Embrace AI,” Reuters, February 18, 2025, https://www.reuters.com/technology/artificial-intelligence/tiger-brokers-adopts-DeepSeek-model-chinese-brokerages-funds-rush-embrace-ai-2025-02-18/

^[58] Zou, “DeepSeek is Now a Global Force”

^[59] Adam Au and Felicia Feiran Chen, “China Expands AI Globally through the Digital Silk Road,” East Asia Forum, April 11, 2025, https://eastasiaforum.org/2025/04/11/china-expands-ai-globally-through-the-digital-silk-road/

^[60] Belt and Road Reboot: Beijing’s Bid to Reorient Globalisation, AidData, https://www.aiddata.org/publications/belt-and-road-reboot; Prithvi Gupta, “China’s Belt and Road After the 3rd Forum: Reform or Rebrand?,” Observer Research Foundation, https://www.orfonline.org/english/expert-speak/belt-and-road-reforms-sustainability-strategy-and-soft-power.

^[61] Sameer Patil and Prithvi Gupta, “The Digital Silk Road in the Indo-Pacific: Mapping China’s Vision

for Global Tech Expansion,” ORF Issue Brief No.683, Observer Research Foundation, January 2024, https://www.orfonline.org/research/the-digital-silk-road-in-the-indo-pacific-mapping-china-s-vision-for-global-tech-expansion

^[62] Patil and Gupta, “The Digital Silk Road in the Indo-Pacific”

^[63] Tobias Feakin, “DeepSeek’s Disruption: Geopolitics and the Battle for AI Supremacy,” RUSI, February 7, 2025, https://www.rusi.org/explore-our-research/publications/commentary/DeepSeeks-disruption-geopolitics-and-battle-ai-supremacy

^[64] Paulo Aguiar, “The Global AI Race: The Geopolitics of DeepSeek,” Geopolitical Monitor, February 12, 2025, https://www.geopoliticalmonitor.com/the-global-ai-race-the-geopolitics-of-DeepSeek/

^[65] Mimi Zou, “DeepSeek is Now a Global Force. But It’s Just One Player in China’s Booming AI Industry,” The Conversation, February 28, 2025, https://theconversation.com/DeepSeek-is-now-a-global-force-but-its-just-one-player-in-chinas-booming-ai-industry-250494

^[66] Feakin, “DeepSeek’s Disruption”

^[67] Barry Pavel et al., “AI and Geopolitics: How Might AI Affect the Rise and Fall of Nations?,” RAND, November 3, 2023, https://www.rand.org/pubs/perspectives/PEA3034-1.html

^[68] Vu, Ngo, and Nguyen, “A Spectrum of Autonomy”

^[69] Vu, Ngo, and Nguyen, “A Spectrum of Autonomy”

^[70] Vu, Ngo, and Nguyen, “A Spectrum of Autonomy”

^[71] Vu, Ngo, and Nguyen, “A Spectrum of Autonomy”

^[72] Naresh Mehta, Laksh Parthasarathy, and Arun Prasad, “Why DeepSeek AI Disruption Spells Opportunity For Businesses,” Tata Consultancy Services, https://www.tcs.com/what-we-do/services/artificial-intelligence/deep-seek

^[73] Elad, “DeepSeek AI Statistics 2025”

^[74] Rashmi Ramesh, “Security Researchers Warn of New Risks in DeepSeek AI App,” Bank Info Security, February 10, 2025, https://www.bankinfosecurity.com/security-researchers-warn-new-risks-in-DeepSeek-ai-app-a-27486; “A Deep Peek at DeepSeek,” Security Scorecard.

• Cyber Security
• Artificial Intelligence
• India
• deepseek
• deepseek llm
• deepseek r1
• digital silk road
• global ai innovation
• global ai race
• large language models

The views expressed above belong to the author(s). ORF research and analyses now available on Telegram! Click here to access our curated content — blogs, longforms and interviews.