The financial landscape witnessed a seismic shock on January 27th, when the market for American tech stocks crumbled dramatically, resulting in a collective loss of over a trillion dollars in market value among major tech companiesNVIDIA, a leading chip manufacturer, was estimated to see a staggering $300 billion decrease in its valuation, marking a record for the largest single-day decline ever recordedAlongside NVIDIA, other prominent U.Stechnology firms and European energy providers that support data centers also faced significant devaluation.

This turmoil in the capital markets was catalyzed by an unexpected player—DeepSeek, a nascent tech startup founded just over a year ago, which was hardly known even in China until nowThis small company has emerged as the first unexpected disruptor, or as it’s called in finance jargon, a “black swan,” not only affecting stock prices but shaking the very foundations of Wall Street.

Founded by Liang Wenfeng, who heads a quantitative fund called Huanfang, DeepSeek has launched its AI model, R1, which many are touting to rival the latest offerings from tech giants OpenAI and Meta

Advertisements

Esteemed investor Marc Andreessen described R1 as "one of the most stunning and impressive breakthroughs," dubbing it as part of an “Sputnik moment” in the field of artificial intelligence.

Nonetheless, there are voices of skepticismSome institutions have gone on record to warn investors that the hype around DeepSeek is exaggerated, arguing that attempting to challenge giants like OpenAI with a mere $5 million might be overly ambitious.

As of the time of reporting, DeepSeek experienced yet another instance of what seemed like an application outage, making it the third such incident in just two days following the rapid popularization of its appThe company has stated that it is investigating the ongoing issues.

So, who or what exactly is changing the game in this industry?

At present, DeepSeek's application is topping the charts in the Apple App Store, adopting an open-source model

Advertisements

According to app tracking firm App Figures, the app's downloads have surged to 1.6 million as of January 25.

To add to the excitement, shares of Daily Interactive (300766.SZ), a stakeholder of DeepSeek, surged to their limits on the A-share market on January 27thLiang Wenfeng, who played an instrumental role within Daily Interactive under its chairman Fang Yi, has successfully integrated DeepSeek models into the company’s marketing strategy.

Liang, born in 1985 and now at forty, recently represented the AI sector at a conference attended by the Premier and various business leadersHis prior experience in quantitative trading has provided him the foresight to invest heavily in AI projects, securing thousands of NVIDIA GPUs.

However, early business associates paint a different picture of LiangThey describe him as somewhat of a "bookworm" sporting a rather disheveled hairstyle

Advertisements

When he excitedly shared his ambitions of creating a large-scale chip cluster capable of training expansive models, few took him seriously.

Liang conveyed his ambitious vision to his colleagues, stating, "I aim to establish a vast cluster for training large models, which will significantly shift paradigms." Many were skeptical, believing only established titans like ByteDance or Alibaba could endeavor such a feat in China.

Previously, Liang had utilized artificial intelligence and algorithmic trading during his time with the quantitative fund, creating substantial wealth by identifying patterns that could sway stock pricesHis adeptness with NVIDIA’s chips in the stock trading arena became a hallmark of his success.

In 2023, he set the groundwork for DeepSeek with an investment of 10 million RMB, announcing plans to develop human-level artificial intelligenceTo realize this, he brought top talents from his quantitative fund to DeepSeek, forming a skilled team well-versed in AI infrastructures who are capable of innovating ways to maximize chip usage before competitors in China.

The emphasis on hiring elite technical talent cannot be overlooked

• US Stocks Continue to Rise in January
• Nasdaq Surges Over 2%
• US-Canada Trade Spat: Will Oil Prices Surge?
• Bank of Japan Resumes Interest Rate Hikes
• FOF Asset Allocation: Strategies for Steady Progress

Currently, DeepSeek remains a purely technical entity, not unlike the early days of Google’s DeepMindIt operates an open-source framework and has yet to monetize its offerings or seek external fundingLiang holds the perspective that increased investments do not always correlate with innovation, cautioning against larger corporations monopolizing creativity.

Fortunately for Liang, the vast wealth gathered from his previous endeavors allows him to attract the finest in AI talent with lucrative salariesReports suggest that salaries for deep learning researchers can reach up to 110,000 RMB per month, and key systems development engineers can earn as much as 90,000 RMB monthly, easily pushing annual income above a millionOperating teams are based in both Beijing and Hangzhou.

The size of the DeepSeek workforce is noteworthy: under 140 employees, equating to merely about 10% of the research and development staff at overseas rivals like OpenAI

Unlike many technology firms that vigorously seek talent from major American tech companies, DeepSeek prides itself on a fully localized talent pool, recruiting primarily from China’s elite universities, like Peking University, Tsinghua University, and Beihang University, instead of experts or "overseas returnees" from the U.S.

There lies a unique value in local talent, especially in understanding how to economize resources effectively for training and running modelsDeepSeek claims to have utilized just 2,048 NVIDIA H800 chips and about $5.6 million to train a model boasting 671 billion parametersIn contrast, the equivalent AI models developed by OpenAI and Google have cost approximately ten times more.

This suggests that the prevalent barriers to large model training could be shattered, fundamentally altering the AI industry landscape for years to come“There’s a significantly higher number of businesses with $6 million in funds than those with $100 million or $1 billion,” remarked a senior investor in the AI domain.

Ritwik Gupta, an AI policy researcher at UC Berkeley, added that, “The first person to train a model must invest substantial resources to achieve results, but subsequent developers can replicate this in a quicker and more cost-effective manner.” Furthermore, he noted that the pool of software systems engineers in China dwarfs that of the U.S.

So, is the “DeepSeek effect” being overstated? Opinions diverge

Some Wall Street analysts and AI researchers don’t share this optimismA report by Bernstein cautioned that the prospect of a Chinese company aspiring to replicate OpenAI’s success with $5 million sounds unrealistic and is not worthy of serious discussionMeanwhile, other researchers suggest that while DeepSeek may have a cost advantage in training, it only demonstrates an ability to quickly replicate the latest American technological advancements, but realistically, sustaining a competitive edge in the long run remains a challenge.

DeepSeek’s rise occurs amid heightened restrictions on high-performance AI chips imposed by the United StatesLiang has openly discussed the constraints faced by Chinese AI development not as a lack of funding, but rather due to limited access to superior American semiconductor technologiesThis underscores his belief that China must foster its homegrown ecosystems and that open-source initiatives are essential for cultivating a thriving software landscape.

An industry analyst remarked, “Open sourcing allows for the aggregation of users, leading to greater collaboration with top-tier talent across the globe

Current trends show that the best open-source technologies are emerging from China, enabling American developers to create products based on Chinese open-source systems.”

In stark contrast to DeepSeek’s low-cost approach, the U.Sis funneling substantial investments into its AI infrastructureLast week, the American government announced the 'Interstellar Gateway' AI investment initiative, setting aside $100 billion for initiation, with plans to potentially inject up to $500 billion over the next four years.

Furthermore, UBS data projects that by 2024, massive tech firms in the U.Swill collectively allocate $224 billion towards AI, with expectations of that figure climbing to $280 billion by 2025.

Some chip analysts opine that the plummet in U.Stech stock values was merely a short-term reactions to DeepSeek’s developments, which could ultimately positively influence manufacturers of AI chips like NVIDIA in the long run