DeepSeek strikes the global AI market, triggering panic selling on Wall Street overnight, with the market value of European and American tech stocks potentially evaporating by $1.2 trillion, among which NVIDIA saw a decline of over 18%, marking the largest single-day market value loss in U.S. stock market history, approximately $600 billion.
In response to the shock, tech giants quickly reacted. Microsoft CEO Satya Nadella posted on social media platform X, referencing the concept of Jevons paradox, pointing out that as artificial intelligence becomes more efficient and user-friendly, its usage will surge and become &#34;a commodity we can never be satisfied with.&#34;
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/a39f4d6b-7aa6-4bda-b398-c399f3b08912.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/a39f4d6b-7aa6-4bda-b398-c399f3b08912.png"/>
<h2>Jevons Paradox: Efficiency Gains May Lead to Increased Demand</h2>
Jevons paradox is an economic concept proposed by British economist William Stanley Jevons in 1865. DeepSeek explains that it refers to the idea that improving resource usage efficiency may actually increase its total consumption. Jevons discovered in &#34;The Coal Question&#34; that as the efficiency of steam engines improved, coal consumption did not decrease but rather increased. The core idea is:
<blockquote>
Technological advancements improve resource usage efficiency, increased efficiency lowers resource usage costs, and reduced costs stimulate growth in resource demand. The growth in demand may exceed the savings brought about by efficiency improvements, ultimately leading to an increase in total resource consumption.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/704f9192-5a91-437e-b773-8961c8e5b067.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/704f9192-5a91-437e-b773-8961c8e5b067.png"/>
</blockquote>
Analysts from several investment banks, including Cantor Fitzgerald, have applied this theory to the DeepSeek R1 model and the trend of democratization in the field of artificial intelligence. Cantor Fitzgerald noted in an investment report:
<blockquote>
&#34;We believe that the concern about GPU spending reaching a peak is far from the truth. DeepSeek is actually very beneficial for computing and NVIDIA, as general artificial intelligence (AGI) seems closer to reality, and Jevons paradox will almost certainly lead the AI industry to require more, not less, computing resources.&#34;
</blockquote>
<h2>DeepSeek's Four Innovations Trigger Market &#34;Shock,&#34; Wall Street Analysts Speak Out</h2>
The DeepSeek R1 model has shocked the global tech community with its astonishing cost efficiency, completing training for about $6 million, roughly 1/50th the cost of similar large language models in the U.S. and EU. In some respects, this model is far superior to OpenAI's o1 model. More importantly, the operational cost of R1 is only 3% of what OpenAI typically charges for compute-intensive outputs.
Analysis indicates that DeepSeek's ability to achieve such high efficiency hinges on several innovative technologies:
<blockquote>
<ol>
<li>
Use 8-bit floating-point numbers to reduce memory usage by about 75%
</li>
<li>
Able to handle multiple tokens simultaneously
</li>
<li>
Only a small portion of parameters are active at any given time
</li>
<li>
Employ a rule-based reward system for reinforcement learning to teach the model to &#34;think&#34; about problems gradually
</li>
</ol>
</blockquote>
This breakthrough has raised concerns in the market about the valuation bubble of artificial intelligence concept stocks, with investors beginning to question the demand prospects for massive-scale computing, especially considering:
<blockquote>
&#34;The DeepSeek R1 model was trained using only 2,000 H800 GPUs. Is there still a need for hundreds of thousands of top GPUs from NVIDIA?&#34;
</blockquote>
Notably, the decline in stock prices before NVIDIA's fourth-quarter earnings report has added pressure to the group. Wall Street has already anticipated that, with a renewed focus on capital expenditures, the group's profit growth in the fourth quarter will drop to 22%, the lowest level in nearly two years.
Axios business editor Dan Primack pointed out, this could impact other startups building AI models, &#34;For venture capital firms fully invested in foundational model companies, this could be an extinction-level event, especially if these companies have not yet achieved widespread distribution of their products.&#34;
However, despite the panic selling in the market, some analysts believe this reaction may be exaggerated. Bernstein analyst Stacy Rasgon stated, the development of DeepSeek does not mean the 'end of AI infrastructure':
<blockquote>
&#34;I don't think we are close to the upper limit of AI computing demand. I believe that if you unleash computing power, it is likely to be absorbed... If we want things to continue to evolve, we need innovations like this.&#34;
</blockquote>
Futurum Chief Strategist Daniel Newman explained, &#34;The market has completely misunderstood this. If we can use computing resources more efficiently, those companies we think are not generating enough revenue will be able to build models at a lower cost. They will be able to create solutions with lower overhead, thus driving higher earnings per share.&#34;
Wedbush analyst Dan Ives called it a buying opportunity for tech stocks, &#34;Now is not the time to panic, as short sellers are trying to control today's narrative.&#34;
Although analysts from Raymond James believe this development is unfavorable for &#34;large GPU clusters,&#34; analysts from Citigroup and Bernstein have also taken a similarly optimistic view on NVIDIA.
Principal Asset Management Chief Global Strategist Seema Shah also stated that if DeepSeek is indeed as they claim, it will ultimately have a positive impact on productivity across various industries worldwide

DPSK

NVDA

Some analysts claim that the rise of DeepSeek could have an "extinction-level" impact on other startups building AI models, but Bernstein analysts believe that the development of DeepSeek does not signify the "end of AI infrastructure." Cantor Fitzgerald analysts state that, as proposed by the Jevons Paradox, improving resource efficiency may lead to an increase in the consumption of that resource. DeepSeek will result in the AI industry needing more, not less, computing resources

- DeepSeek's R1 model has caused a significant sell-off in the global AI market, with Wall Street analysts predicting a potential loss of $1.2 trillion in tech stock value, particularly affecting Nvidia.  
- The model's cost efficiency, achieved through innovative techniques, raises concerns about the future demand for large-scale computing resources, as it completed training with only 2,000 GPUs.  
- Despite the panic, some analysts argue that the advancements in AI infrastructure could lead to increased productivity across various industries, suggesting that the market's reaction may be exaggerated.

With the invention of the steam engine, did the use of coal decrease? The night of NVIDIA's plunge, this is Wall Street's hottest discussion