Advertisements
On January 27, a notable upheaval occurred in the American stock market, particularly affecting the Nasdaq 100 index, which represents large tech companiesJust as the market opened, it plummeted by 3%, and at one point, the drop neared 5% in pre-market trading.
NVIDIA, the flagship stock leading the current AI tech wave, saw its shares tumble by as much as 12%, with maximum losses circulating close to 15%. This raises the question: what has rattled the usually resilient US stock market and its perennial AI narrative?
The prevailing speculation points towards a fierce competitor emerging from China, DeepSeek, a large language model company that has seemingly disrupted established norms.
On January 24, a respected third-party benchmark revealed that DeepSeek's large model outperformed those of formidable competitors with multi-billion dollar investments, such as Meta's Llama 3.1, OpenAI's GPT-4o, and Anthropic's Claude Sonnet 3.5. To condense this, the DeepSeek-R1 benchmark testing placed the company within the top three globally across all categories of large models.
Widely recognized assumptions pigeonhole large models as mere power-intensive, costly endeavors dictated by OpenAI's templatesHowever, DeepSeek's team has taken an unconventional route by proposing a Multi-Head Potential Attention Mechanism (MLA) architecture that has cut memory usage down to just 5% to 13% of the past prevalent MHA configurations
Advertisements
Their innovative DeepSeek MoE Sparse structure minimizes computational loads significantly, leading to drastic cost reductions—achieving remarkable reasoning capabilities at less than 5% of the cost typically associated with international large models.
Consequently, the unfortunate casualty appears to be NVIDIA, not the cash-burning large model companies, since various enterprises no longer need to invest heavily in expensive graphics cards for computationDeepSeek has made its project open source, revealing its architecture, mechanisms, and algorithms, enabling users to adopt and apply them freelyThis paradigm shift raises doubts about the ongoing demand for NVIDIA chips, which were previously in high scarcity.
This narrative has not only impacted NVIDIA but sends tremors through the AI wave and the collective market valuation represented by the Nasdaq 100, which saw a staggering evaporation of $2 trillion in value.
However, the question arises: is this story truly reflective of reality?
The answer is a resounding no.
DeepSeek had already garnered significant attention by mid-2024, often being likened to the “Pinduoduo” of the AI sector due to its cost-effective large model training approachesIts reasoning capabilities have been anticipated to rival those of international giants, making the outstanding assessment results on January 24 less surprising.
In the vast realm of the US stock market, how could a product from a single Chinese company cause such a seismic shift overnight?
Moreover, financial markets typically react with great sensitivity, as nobody desires to disregard profitability
Advertisements
Given the assessment results from January 24, one would expect an immediate market reaction; what prompted the delay until the 27th—was it a strategic pause for short sellers to position themselves?
Why then did the US stock market undergo such a significant decline?
The answer is, in fact, quite straightforward—
overall valuations have reached the highest levels in over two decades, and market sentiment has escalated to the utmost optimism wherein risk tolerance has peakedFurthermore, President Biden’s hypothetical interstellar gateway, referred to as Stargate, has further ignited the AI narrative.
According to the latest data (as of September 30), American households currently hold a striking 43.4% of their financial assets directly and indirectly allocated in stocks—an unprecedented peak in historySimultaneously, allocation to savings and bonds is nearing historical lows, indicating an overwhelming bullish attitude towards US stocks.
Evaluating from a valuation lens, metrics such as the forward price-to-earnings ratio and median forward price-to-earnings ratio have risen to heights unseen, save for during pandemic bubbles over the past 22 yearsOne might ponder what more could be expected?
Even isolating the tech giants, often referred to as the “Seven Sisters,” their valuations stand at peaks comparable only to pandemic bubble times.
What justifies this valuation?
It is precisely due to these factors that I penned a piece last Friday entitled “No One Is Worth Four Trillion Dollars!” specifically targeting NVIDIA’s valuation with scrutiny.
Moreover, during late Thursday and Friday, data from the Chicago Board Options Exchange indicated that the Put/Call ratio was nearly 0.8, hovering near all-time lows over recent years
Advertisements
Advertisements
Advertisements
Leave a Reply