In the ever-evolving world of artificial intelligence (AI), the emergence of new players and technologies can shift the landscape significantlyOne such entity that has recently made waves is DeepSeek, a company based in China that has developed a powerful large model that rivals the capabilities seen when ChatGPT burst onto the sceneWhile ChatGPT pushed AI to new heights, DeepSeek presents a refreshing alternative, illuminating a pathway for small and medium-sized enterprises (SMEs) to explore the realm of large model training.

Despite the excitement surrounding DeepSeek, speculation has arisen suggesting that it could lead to a catastrophic decline in American tech stocks, particularly those of giants like NVIDIASuch predictions, particularly following NVIDIA's staggering 17% drop on January 28, 2023, are not only unfounded but also reflect a lack of understanding of the broader technology ecosystem

Advertisements

In response to these alarming claims, I promptly posted on social media, countering the narrow-minded forecast that seemed to infiltrate public discourse.

To the surprise of naysayers, when the markets opened later that day, NVIDIA's stock surged over 2% and continued its upward trajectory, closing at a near-record high of $128.99 per share, marking an impressive 8.9% gainThis recovery is noteworthy considering that the day prior, NVIDIA had experienced a staggering loss, with over 40 billion yuan of market cap evaporatingHowever, through a mix of investor confidence and market dynamics, NVIDIA regained approximately 19 billion yuan in market value by the end of the next trading day.

While such a recovery might tempt some to hastily conclude that NVIDIA's troubles are over, it is essential to recognize that its stock price has appreciated eightfold over the past two yearsSuch substantial growth often invites corrections, driven by the cyclical nature of financial markets where bulls and bears constantly battle

Advertisements

In this environment, DeepSeek could be viewed as a tactical weapon in the hands of short-sellers, creating a momentary opportunity to bet against NVIDIA, yet it does not alter the overarching trend where demand for advanced computational chips remains robust.

What many observers fail to grasp is the fundamental fact concerning DeepSeek’s operations—they utilize NVIDIA’s GPUs in training their large modelsThere has been rampant speculation online regarding the extent of DeepSeek’s GPU acquisition, with figures ranging from 10,000 to 50,000, but one truth remains clear: their large model is indeed built on NVIDIA’s technology.

DeepSeek hails from the same lineage as Huanshan Quantitative, a name familiar to many in the trading world, particularly those who engage in stock speculationThis renowned quantitative hedge fund has drawn considerable ire from retail investors who have called for a ban on such private quantitative funds active in the A-stock market

Advertisements

As a top-tier player in quantitative finance, Huanshan had already begun stockpiling computational chips years ago, securing resources essential for developing quantitative trading modelsReports suggest that their GPU reserves far exceed those of newer AI startups, approaching the stockpiling levels of prominent Chinese Internet firms like Baidu, Alibaba, and Tencent (collectively known as BAT).

In 2021, a research paper co-authored by DeepSeek's founder, Liang Wenfeng, detailed their acquisition of 10,000 A100 GPUsAt that time, fewer than five companies across China possessed more than 10,000 GPUs, with Huanshan being the notable exception among Internet giantsThus, it is evident that DeepSeek’s large model is grounded firmly in NVIDIA’s computational technology, and any ambition to enhance their model further will inevitably require finding ways to circumvent U.Sregulations to acquire the latest GPUs.

Recognizing these dynamics, NVIDIA’s response to the initial stock drop was measured, emphasizing that DeepSeek's research exemplifies how to harness their technology using widely available models and compliant computational resources to develop new models

• Volatile Market Value of Leading AI Companies in the U.S.
• U.S. Stock Indices Decline This Week
• Surge in U.S. Inflation Expectations
• Good News! Foreign Capital: Bullish on Chinese Stocks!
• U.S. Interest Rates May Decline in 2025

They stated that the inference process demands a significant volume of high-performance NVIDIA GPUs and sophisticated networking capabilities.

NVIDIA's remarks reflect a pragmatism rather than self-promoting rhetoricFor DeepSeek to advance further, they will require an abundance of top-tier GPUs, and currently, NVIDIA stands alone as the primary producer of the most advanced AI computational chipsEven AMD, another American semiconductor company, lacks the capacity to rival NVIDIA in this domain.

This reality places computational power at the heart of the future of AI competitionThe nation or enterprise capable of mastering adequate infrastructure for computational resources will emerge as a leader in the fiercely contested AI arenaThe semiconductor industry’s prowess—an amalgamation of manufacturing and design capabilities—will dictate who thrives in this space

Despite considerable advancements made by Chinese companies in recent years, a gap still persists when contrasted with the West’s established supply chains, particularly that of the United States.

Critical to remember is that we must not overly pivot our strategies based solely on the remarkable performance of any single large model at a given timeInstead, there should be an increased emphasis on enhancing our investments in the semiconductor sectorEfforts should be made to cultivate and develop domestic champions akin to NVIDIA, ASML, or TSMC, which could significantly bolster competitive capabilities.

Nevertheless, the important takeaway from DeepSeek's journey is the innovative path they have forged for training large modelsTheir approach, while cost-efficient, has demonstrated that impressive outcomes are attainableSmall and medium-sized enterprises around the globe, with constrained financial capacity, can draw inspiration from DeepSeek’s exploratory path, leveraging the discoveries made by the company to train their own large models