Will DeepSeek Bring Down Nvidia?
China has long been recognized for its ability to produce goods more cheaply and efficiently, earning its place as the manufacturing hub of the world. However, for years, it lagged behind in the realm of innovation and cutting-edge technology. That narrative began shifting in recent years, prompting concern from the United States. Feeling the pressure of China’s rapid progress, particularly in artificial intelligence (AI), the U.S. imposed sweeping semiconductor restrictions during Biden's administration. These measures were aimed at slowing China's technological advancement, particularly in AI, under the justification of potential military applications.
Ironically, China responded in a way that aligns perfectly with its strengths: by making AI cheaper. Constrained by a limited supply of Nvidia GPUs—crucial for AI computing—and a lack of access to top-tier infrastructure, Chinese innovators pivoted from competing on raw power to focusing on efficiency. Enter DeepSeek, a Chinese-developed large language model (LLM), which has achieved performance comparable to OpenAI’s ChatGPT but at just one-tenth of the cost.
While OpenAI reportedly uses thousands of Nvidia chips—over 10,000 GPUs—to train large models like GPT-4, DeepSeek achieves comparable performance with only hundreds of GPUs. This remarkable efficiency stems from its innovative software design, which maximizes computing power while minimizing hardware requirements. As a result, DeepSeek delivers top-tier AI capabilities without the significant hardware and energy expenses that drive up costs for U.S. companies.
Surprisingly, DeepSeek wasn’t developed by a major Chinese tech giant like Tencent, Alibaba, ByteDance, Baidu, or SenseTime—companies long associated with AI advancements. Instead, it came from a quant hedge fund, High-Flyer, best known for using AI in financial trading. The fact that a hedge fund, not a technology company, was behind this innovation highlights how breakthroughs can emerge from unexpected sources.
The rise of DeepSeek has not gone unnoticed in the West. OpenAI’s CEO, Sam Altman, even appeared to take a veiled swipe at DeepSeek with a post on X, subtly acknowledging the competition posed by this new player.
Sam Altman is certainly correct that OpenAI introduced large language models (LLMs) to the mainstream with ChatGPT, but Altman perhaps should also acknowledge Google for its groundbreaking research paper, “Attention Is All You Need.” This paper introduced the Transformer architecture, which has become the bedrock of every major AI model, including GPT-4 and DeepSeek.
Altman is also right in suggesting that DeepSeek didn’t create something entirely new—it didn’t go from “0 to 1,” as Peter Thiel might put it. Instead, it refined and scaled existing technologies, moving from “1 to 2.” But that leap from 1 to 2 has far-reaching implications. By making AI significantly cheaper and more efficient, DeepSeek has the potential to disrupt the entire AI industry in the West. It challenges the status quo of relying on massive infrastructure and expensive GPUs, forcing companies to rethink how AI systems are developed, deployed, and monetized.
At the time of writing, Nvidia holds the title of the most valuable company in the world by market capitalization. This meteoric rise is nothing short of remarkable—it achieved this status in less than three years, with its share price surging an astonishing 11 times during this period.
Nvidia's extraordinary rise is largely due to its chokehold on the supply of its high-end GPUs. AI computing demands massive processing power, and no company makes GPUs as powerful and efficient as Nvidia’s. The wild success of ChatGPT was a catalyst that caught the attention of Big Tech, triggering a race to invest heavily in AI infrastructure. Companies like Microsoft, Meta, Amazon, and Alphabet poured billions into acquiring GPUs to ensure they wouldn’t be left behind in the AI revolution.
In this scenario, Nvidia effectively became the shovel seller to the gold miners, capitalizing on the AI gold rush. This dynamic is clearly reflected in the surge in capital expenditure by Big Tech from the latter half of 2023 onward. Their spending skyrocketed as they scrambled to secure the hardware needed to build cutting-edge AI capabilities, further cementing Nvidia’s dominant position in the industry.
Nvidia’s staggering profits are largely driven by a small group of Big Tech companies, including Microsoft, Meta, Amazon, and Alphabet. These firms have invested heavily in AI infrastructure to support their own AI programs while also offering cloud-based AI services to companies like OpenAI, Anthropic, Perplexity, Claude, Grok, and others. This concentration of demand has created a lucrative pipeline for Nvidia, as its GPUs are essential for powering the enormous computing needs of these AI models.
However, there’s a lingering uncertainty about the true end-user demand for AI applications. While the initial rush to build AI infrastructure has been explosive, there’s a possibility that it could lead to overbuilt capacity in the long run. If demand for AI services doesn’t scale as expected or shifts toward more cost-efficient solutions like DeepSeek’s models, these massive investments in AI infrastructure could become underutilized.
What makes DeepSeek even more disruptive is its open-source nature, meaning anyone can adopt and build upon its technology. It’s only a matter of time before cloud giants like Amazon, Microsoft, and Alphabet integrate DeepSeek’s models into their platforms. Doing so would allow them to meet customer demand for efficient, cost-effective AI solutions while also slashing their own operational expenses, thereby improving profitability.
The ripple effect doesn’t stop there. DeepSeek’s breakthrough could spur other AI players to rethink their designs, pushing for greater efficiency and lower costs. This shift would not only make AI more accessible but also challenge the high-cost infrastructure model that has dominated the industry, potentially reshaping the competitive landscape for years to come.
This shift toward cost-efficient models like DeepSeek’s would also mean that cloud giants may no longer need to buy as many Nvidia GPUs. As AI infrastructure growth moves toward a more sustainable and efficient pace, the demand for Nvidia’s GPUs is likely to decline. This could directly impact Nvidia’s growth trajectory, potentially slowing down its meteoric rise and causing a revaluation of its stock, which could lead to a drop in its share price. Furthermore, reduced GPU demand would cascade down the semiconductor supply chain, lowering orders for Nvidia’s suppliers, such as TSMC, and impacting the broader industry.
While the full impact of DeepSeek’s rise remains uncertain, the growing media coverage is undoubtedly increasing awareness and adoption potential. However, the effects won’t be evenly distributed. We believe Nvidia will face the hardest hit from this disruption, as its dominance is closely tied to high GPU demand. In contrast, cloud giants like Amazon, Microsoft, and Alphabet are likely to benefit from DeepSeek’s advancements. By leveraging its cost-efficient models, they can reduce expenses, improve profitability, and maintain their competitive edge, turning what might seem like a threat into an opportunity.