DeepSeek is not one of China’s tech giants like Alibaba, Baidu, or Tencent. Instead, it is a relatively obscure division of High-Flyer Capital Management, a Chinese hedge fund specialising in quantitative trading. Yet, this little-known subsidiary has unexpectedly emerged at the forefront of AI research, challenging the notion that China can only imitate rather than innovate.
What has caught global attention is not just that DeepSeek may be outperforming Alibaba, Baidu, or Tencent—it is the possibility that it could rival OpenAI or LLaMA. But is that truly the case? The Chinese Communist Party (CCP) prioritises shaping narratives, and if the world accepts China’s version of events at face value, it only serves the party’s interests. While there are valuable insights to be gained, understanding them requires careful contextual analysis.
The key questions remain: How much of DeepSeek’s advancements are genuinely original, and how much is copy-pasted? And, more importantly, what does this reveal about the evolving global digital landscape—where the US fights to maintain its dominance while China pushes forward with innovations to close the gap?
The story
Liang Wenfeng, one of the co-founders of DeepSeek’s parent company High-Flyer Capital Management, launched the startup with the goal of developing broadly applicable generative AI models. The first model debuted on 2 November 2023. However, at the time, it barely made an impact. In fact, 2022 and 2023 were defined by the Biden administration’s intensified efforts to safeguard US dominance in advanced semiconductors—a topic I will revisit later when discussing export controls.
DeepSeek’s second attempt with V3 also failed to generate significant buzz. Then came its third and arguably most ambitious breakthrough—the reasoning model R1. While AI enthusiasts grasped its significance early on, what truly propelled DeepSeek into the global spotlight wasn’t just the innovation itself. Instead, it was the reaction on Wall Street—particularly the sharp decline in Nvidia’s stock, the leading US AI chipmaker. That’s when the conversation shifted and Donald Trump weighed in with his signature remarks. And that’s when DeepSeek officially arrived.
The timing was ironic, to say the least. Just days before, Trump had unveiled his $500 billion plan to secure America’s lead in AI. But as DeepSeek’s model demonstrated the ability to run efficiently on personal MacBooks, the real question emerged—why pour half a trillion dollars into an AI arms race when China appears to be achieving breakthroughs almost for free?
Also read: DeepSeek is the dawn of a new world order, say the Chinese
How did DeepSeek do it?
Allegations of outright cyber espionage may be unsubstantiated, but DeepSeek did engage in enough ‘copy-pasting’. The company reportedly found a way to interact with ChatGPT’s system at scale, logging vast amounts of interactions and using that data to train its own model—drastically cutting costs in the process. However, the story runs deeper. Evidence now suggests that DeepSeek may have accessed the ChatGPT API, which is officially unavailable in China but could have been accessed through external means. Not only did the company harvest this invaluable training data and integrate it into its own models, but it also went a step further—publishing a publicly accessible research paper detailing its findings.
Of course, true technological innovation requires sophisticated mathematical modelling and advanced algorithms. But at its core, DeepSeek’s model stands on a foundation largely built by ChatGPT—only optimised for efficiency and developed at a fraction of the cost. Whether this constitutes a violation of OpenAI’s service agreements or outright intellectual property theft remains to be seen, but major tech firms are already investigating.
Dismissing such innovation outright would be shortsighted. Understanding its true value is crucial. From a purely academic standpoint, openly published technological advancements contribute to human progress—a testament to the scientific method, where knowledge evolves through a cycle of conjecture and refutation.
However, this idealistic view of shared knowledge collides with the harsh realities of geopolitics. The pursuit of national interest often overrides the noble spirit of scientific collaboration. While AI experts can marvel at the technical breakthroughs, strategic analysts must examine the broader implications—particularly given that DeepSeek operates under Chinese law. This legal framework is deeply intertwined with a system notorious for secrecy, mass surveillance, and psychological control. The real conversation, then, is not just about AI innovation—it’s about the geopolitical and ethical consequences of technological power in the hands of an opaque regime.
Also read: Indian foreign secretary’s Beijing visit shows progress in ties. Can China be trusted?
Leaking export controls
To understand the geopolitical context, we need to rewind to the Biden administration’s October 2022 export controls on advanced chips. These restrictions were designed to regulate the sale of semiconductors based on their processing power, ensuring that cutting-edge AI chips wouldn’t fall into Chinese hands.
What the policy failed to anticipate, however, was the diversion of Nvidia chips to DeepSeek.
At the heart of this issue is Nvidia’s built-in ‘compartmentalisation’ system. To account for potential manufacturing defects, Nvidia designed its chips so that specific sections could be activated or deactivated as needed, rather than scrapping an entire chip. While this was meant to be a safeguard, it also became a loophole.
The feature allowed certain chips to be classified as lower-capability models and, therefore, eligible for export. As a result, China received the H800—technically an ‘older generation’ chip—while the US continued using the H100, its most advanced version. However, the distinction between the two is more of a regulatory workaround than a genuine technological gap. Nvidia’s H800 is strategically compartmentalised in a way that allows it to function almost identically to the H100—meaning China effectively gained access to high-performance AI chips despite the export controls.
Between October 2022 and October 2023, exporting H800 chips to China was entirely legal. Not only DeepSeek but many other Chinese firms seized the opportunity, stockpiling these chips in anticipation of future restrictions. By the time the fiscal year closed in early 2024, Nvidia had already sold $10 billion worth of these chips to China.
Now that the initial shockwave of DeepSeek’s rise is beginning to settle, these details are coming to light and being widely debated. Questions are mounting over the claim that DeepSeek developed its breakthrough AI model with a mere $5 million. And as OpenAI and Microsoft complete their investigations into possible intellectual property violations, even more loopholes may emerge.
The most immediate casualty of this unfolding saga has been Nvidia’s stock price, which plummeted in reaction to exaggerated fears. For investors, this dip presents a golden opportunity—given market indicators, Nvidia shares are likely to rebound and skyrocket once again.
For the Trump administration, these revelations provide tangible lessons for future export controls. The next set of restrictions will likely focus on closing technical loopholes and adopting a more comprehensive strategy to cut China off from critical AI hardware—one that anticipates potential workarounds.
It seems increasingly likely that DeepSeek or other Chinese firms won’t be able to pull off the same “innovative magic” under tightened restrictions—unless China successfully develops its version of the H100 chip. That, however, is only a matter of time. Meanwhile, if the US’ export controls work as intended, its technological lead will continue to widen, keeping China in catch-up mode for the foreseeable future.
Also read: Can we be a real alternative to China? Modi’s 2014 promise of Indian greatness is in tatters
What else?
In the long run, DeepSeek’s demonstration of innovative efficiency will serve as a case study for the broader AI community—offering insights into how to develop powerful models at lower costs, and perhaps in a more ethical manner.
But beyond technology, there’s the political dimension. This is where regulation comes into play. However, just as economic sanctions often backfire, overregulation risks stifling the very players it aims to protect—creating an environment where research becomes too expensive to remain competitive. The European model is a cautionary tale. Despite being a leader in technological innovation, it remains hamstrung by its own excessive regulations, preventing it from truly competing at scale.
Where does India stand in all of this? The short answer: nowhere.
But India’s relevance to this conversation lies in the ongoing brain drain to Silicon Valley. The issue isn’t a lack of talent—India produces some of the world’s best engineers and researchers—but the absence of an ecosystem that offers the same level of professional growth and financial rewards as the West.
It would be redundant to highlight the chronic lack of R&D investment, a problem that affects nearly every sector of India’s technological progress. Instead, the most Pareto-optimal solution for India is to actively collaborate with trusted partners and participate in shaping the emerging digital order rather than attempting to go solo.
As for the future? It remains fluid, unfolding in ways that are yet to be fully understood.
Swasti Rao is Consulting Editor, ThePrint and a foreign policy expert. She tweets @swasrao. Views are personal.
(Edited by Prasanna Bachchhav)
Excellent nuances. However, the author has somehow failed to take note of the emergence of a new generation of Chinese entrepreneurs who prioritise foundational research and long-term technological advancement over quick profits.