Arthor: Kris Main

Artificial Intelligence (AI) is a hot technology and investment concept in recent years. Semiconductor and high-end equipment manufacturing industries are booming. Private equity funds, hedge funds, venture capital, investment banks, seed round funds and other investment industries are also strongly sought after.

China and the United States are undoubtedly the first echelon of the global AI industry.As a pioneer country, the United States has advantages in capital investment, hardware advancement and concept innovation;China has obvious advantages in industrialization, engineer reserve and government efficiency.

The nine-year Sino-US trade war and technological war are particularly evident in the AI arms race. In addition to industrial competition, the United States is also suppressing China's AI industry in politics and public opinion. In the field of AI, the United States has been in an offensive position, while China is constantly seeking breakthroughs in its defensive position.

The seesaw battle often has a tipping point. The AI industry competition between China and the United States is also in line with this rule. It can even be said that the arrival of this singularity is inevitable.

Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co., Ltd. (Hereinafter referred to as DeepSeek) launched the DeepSeek R1 model on January 20, 2025.This is undoubtedly the singularity. Liang Wenfeng, born in 1985, and High-Flyer, a private equity fund he set up.Is undoubtedly the biggest contributor to this singularity.

At the same time, the DeepSeek - R1 model blew a super hurricane in the US capital market. The Nasdaq Index's loss of nearly $1 trillion in market value in a matter of days is certainly a case in point. Is the high valuation of American technology stocks reasonable, and does the high investment of AI industry necessarily mean high returns? It has become the focus of suspicion among global investors.

The subsequent suppression and slander of DeepSeek in the US political arena and media is quite absurd. In addition to the trumped-up accusation that DeepSeek plagiarized, the focus of public opinion is on the ideological field. The helplessness of "long-arm jurisdiction" is more like a kind of psychological comfort.

Faced with the impact of DeepSeek model, the US capital market continues to choose to increase its bets. With the advent of xAI's Grok-3 model, US tech stocks have regained their previous heat. However, the stock market bubble and the financial bubble are rapidly overlapping at a speed visible to the naked eye. The undoing of the Ponzi scheme looms.

However, it is clear that American investors and the technology community are not all irrationals. The Chinese AI white paper of QUALCOMM (Qualcomm): "AI Change is Driving Terminal Side Reasoning Innovation" is more explicit. The general trend of AI development is the transformation from AI training (Pre-training and post-training) to large-scale reasoning models (Inference Models).

At the same time, the Chinese government and industry are integrating into the DeepSeek model ecosystem. From traditional industries to science and technology industries, from government management to children's winter vacation homework. Society as a whole is struggling to adapt to the new changes brought about by the DeepSeek model. The special action of "AI +" of the central enterprises of the State-owned Assets Management Committee of the State Council tells the world. China has a strong national will to seriously develop the AI industry.

This article will explore the different logic of AI industry development between China and the United States through DeepSeek. The actions of China and the United States can better show the essential difference between the virtual economy and the real economy in the face of revolutionary technology. All kinds of insolence in the United States are also the best footnotes of the economic system. We can even say that the future development path of China and the United States has been divided in response to the different attitudes of DeepSeek model.

First, DeepSeek came out of nowhere and surprised everyone.

The story begins with the debut of the DeepSeek- R1 model.

On January 20, 2025, DeepSeek published an article on its official website entitled DeepSeek-R1. Press Release of Performance Benchmarking OpenAI O1 Official Release. Manuscripts represent DeepSeek- R1 open-source model weights. DeepSeek also says that 1) the DeepSeek- R1 model follows the MIT License, Allow users to train other models with DeepSeek- R1 through distillation techniques;2) The online API opens the thinking chain output to the user, which can be called by setting model = 'DeepSeek-reasoner'; 3) DeepSeek official website and App will be updated and launched simultaneously from now on.

At the same time, DeepSeek emphasizes three points on the performance of DeepSeek- R1 model:

DeepSeek-R1 Model Performance Alignment in Various Tests OpenAI O1 Official Version Source: DeepSeek Official Website, January 20, 2025

First, the performance is equivalent to the official version of OpenAI O1. Reinforcement learning techniques are used in a large scale in the post-training phase of the DeepSeek-R1 model. In the case of only a few labeled data, the reasoning ability has been greatly improved in mathematics, code, natural language reasoning and other tasks.

DeepSeek- R1 distillation mini model surpasses OpenAI o1-mini-Source: DeepSeek official website, January 20, 2025

Second, the distillation mini model goes beyond OpenAI o1-mini. At the same time of opening two 660B models of DeepSek-R1-Zero and DeepSek-R1, through the output of DeepSek-R1, distilled 6 small models open source to the community. Among them, the 32B and 70B models are outstanding.

API service price source: DeepSeek official website, January 20, 2025

Finally, low service prices. DeepSeek- R1 model API call is priced at 1 (cache hit)/4 (cache misses) per million tokens entered, 16 yuan per million output tokens.

DeepSeek-R1 is also highly praised by the American science and technology community. QUALCOMM (Qualcomm) in its AI white paper released on February 20, 2025: "AI changes are driving innovation in terminal-side reasoning. "In the article, the DeepSeek-R1 model is highly appraised: "Once the DeepSeek-R1 model, a cutting-edge AI reasoning model, comes out, it has caused waves throughout the technology industry. Because its performance can be comparable to or even surpass advanced similar models, it subverts the traditional understanding of AI development.

Such outstanding performance has naturally made DeepSeek the focus of public opinion, and founder Liang Wenfeng has become an adviser to the Chinese government. According to Xinhua News Agency on January 20, 2025, Li Qiang, member of the Standing Committee of the Political Bureau of the CPC Central Committee and Premier of the State Council, chaired a meeting of experts, entrepreneurs and representatives in the fields of education, science, culture, health and sports on the afternoon of January 20. At the forum, we listened to the opinions and suggestions on the Report on the Work of the Government (draft for comments). Liang Wenfeng also attended the premier's forum.

Screenshot of CCTV13 report of Liang Wenfeng attending the Premier's Forum on January 20 Source: Thepaper.cn, January 22, 2025

In a report by thepaper.cn on January 22, 2025, the development history of DeepSeek was detailed. When DeepSeek entered the field of general intelligence in July 2023, it had only 139 engineers and researchers, including founder Liang Wenfeng. OpenAI has 1200 researchers and Anthropic has more than 500 researchers. The DeepSeek- V2 model was successfully developed in May 2024. API calls are priced at 1 yuan per million tokens input and 2 yuan per million tokens output, which is only one percent of GPT-4 Turbo.

The DeepSeek-V3 model was successfully developed on December 27, 2024. The entire training cost for the DeepSeek- v3 model was only $5.576 million. A total of 2048 Nvidia H800 GPU clusters for the Chinese market were used and completed in 55 days. The cost is 10% of the OpenAI GPT-4o model training. Its evaluation results surpass the top open-source models such as Ali's Qwen2.5-72B model and Meta's Llama 3.1-405B model. Even some indicators can be compared with top-level closed-source models such as GPT-4o and Anthropic's Claude 3.5-Sonnet model.

Such an advanced large model with low cost and high efficiency has obviously caused a huge impact on the other side of the ocean. The subsequent performance of the domestic capital market and the public opinion field in the United States can better explain the impact of the DeepSeek-R1 model.

Two, the destructive power of $1 trillion to US stocks.

The day after the DeepSeek- R1 model was released, US President Donald Trump made a big news. On January 21, 2025, under the headline "U.S. President Donald Trump on Tuesday announced a private sector investment of up to $500 billion to fund infrastructure for artificial intelligence, speaking at a press conference at the White House on Tuesday (January 21), Trump said that ChatGPT’s creator OpenAI, SoftBank, opens new tab and Oracle, opens new tab are planning a joint venture called Stargate

US President Donald Trump speaks about artificial intelligence infrastructure at the White House. From left to right are Softbank CEO Masayoshi Son, Oracle co-founder Larry Ellison, and OpenAI CEO Sam O. Terman. Source: Reuters/Carlos Barria, January 21, 2025

Trump said at the meeting that the new enterprise will build data centers in the United States and create more than 100000 jobs in the United States. Masayoshi Son, CEO of Softbank, Sam Altman, CEO of OpenAI Altman and Oracle Chairman Larry Ellison. Trump said These companies, along with other equity backers of Stargate, have committed $100 billion for immediate deployment, with the remaining investment expected to occur over the next four years.

The three executives did not forget to kiss Trump's ass. Oracle's Ellison said the project's first data center is already under construction in Texas. Twenty data centers of 500,000 square feet each will be built in the future. Softbank's Masayoshi Son said, "Unless you win, we will not decide to do this. "This refers to Trump's victory in the presidential election. OpenAI's Altman put it even more bluntly, "To build AGI here," "Without you, Mr. President, We wouldn’t have decided to do this. AGI refers to the more powerful Artificial General Intelligence technology. Such flattery in public is a good indication of attitude.

However, Reuters is not optimistic about the plan. The report points out that the biggest obstacle to Stargate is the high energy consumption of data centers. The announcement on Trump’s second day in office follows the rolling back of former President Joe Biden’s executive order on AI, that was intended to reduce the risks that AI poses to consumers, workers and national security.

On the issue of electricity consumption in data centers, Trump said, "They have to produce a lot of electricity, and we'll make it possible for them to get that production done very easily at their own plants if they want." Unfortunately, however, The North American Electric Reliability Corporation, founded in 1968, disagreed. As U.S. power consumption rises from AI data centers and the electrification of buildings and transportation, about half of the country is at increased risk of power supply shortfalls in the next decade, the North American Electric Reliability Corporation said in December.

Trump is really a mouthpiece. As a candidate in 2016, Trump promised to push a $1 trillion infrastructure bill through Congress but did not. He talked about the topic often during his first term as president from 2017 to 2021, but never delivered on a large investment, and "Infrastructure Week" became a punchline.

Of course, for the US capital market, the intended investment of $500 billion is also a major positive. Shares of Oracle were up 7% on initial report of the project earlier in the day. Nvidia, opens new tab, Arm Holdings and Dell, opens new tab shares also rose. It can be said that the AI myth of the US capital market is still fermenting.

However, as the performance of DeepSeek-R1 model is understood by the market, technology stocks in the US capital market, a week later, on January 27, 2025, there was a super storm. Bloomberg wrote an article on the day of the Nasdaq's big drop titled "AI-Fueled Stock Rally Dealt $1 Trillion Blow by Chinese Upstart. The relevant information is introduced in detail.

Nasdaq futures down on high volume Source: Bloomberg, Jan 28, 2025

According to the article, for the better part of two years, artificial intelligence technologies have been a nonstop gift for US equities investors. The Nasdaq 100 Index rose 92% from the start of 2023, adding more than $14 trillion in value. The performance of DeepSeek-R1 model look as good as anything on the market at a fraction of the price and a sliver of the energy consumption. So, on Jan. 27, the Nasdaq 100 fell 3%, its biggest drop in six weeks. Its market value fell by nearly $1 trillion from last Friday (January 24). Nvidia suffered the most in the fall. Its market value evaporated nearly $600 billion, the largest market value evaporation in history. Shares of energy companies that provide power to the US artificial intelligence sector have suffered particularly. Wherein the Constellation Energy Corp. The biggest drop was 21%.

This article raises a very sharp question. And whether the hundreds of billions in AI investments would ever lead to profits large enough to justify the rich valuations afforded to Megacap stocks. By weight, the group makes up 30% of the S&P 500’s by weighting, more than at any time in history.

Franklin Templeton senior vice president, Franklin Templeton Investment Solutions, Investment Solutions "When valuations stretch to the sky it’s easier for small trembles to make the entire market rumble" says Max Gokhman, president. Vey-Sern Ling, managing director of UBS Private Bank (Union Bancaire Privee), said, "DeepSeek shows that it is possible to develop powerful AI models that cost less, it can potentially derail the investment case for the entire AI supply chain, which is driven by high spending from a small handful of hyper scalers."

Nvidia, for its part, downplayed the threat from DeepSeek, calling that an “excellent AI advancement” that complies with US technology export controls. Interestingly, most of Nvidia's advanced products cannot be sold to China. But even the Inference model, it said in its statement, "requires a lot of NVIDIA GPUs and high-performance networking."

Trump Advisor Compliments DeepSeek- R1 Source: Marc Andreessen's X Account, January 24, 2025

In contrast to Nvidia's evasiveness, Marc Andreessen, a Silicon Valley investor who has been advising President Trump on his own X on January 24, he praised the DeepSeek- R1 model as, "Is one of the most amazing and impressive breakthroughs I've ever seen -- as open-source software, it is a profound gift to the world." Paul Nolte, market strategist and senior wealth manager at Murphy & Sylvest Wealth Management, said: "We don’t know whether this is the ‘Sputnik Moment’ for stocks, but this is certainly a wakeup call that we are not the only game in town."

The words of Charu Chanana, chief investment strategist at Saxo Markets, are even more meaningful. "AI dominance cannot be taken for granted," she told Bloomberg. "Future competitors will evolve faster and challenge the established companies more quickly."

Faced with the DeepSeeK-R1 model, Nvidia's supply chain has also suffered a huge impact, such as Japanese related companies. According to Bloomberg on January 27, 2025, on the same day that technology stocks in the United States fell, Advantest, Nvidia's test equipment supplier, fell 8.2%, while Disco (Disco Corp.) fell by 2.9 percent. Softbank shares, which soared 16% last week (January 24), fell 5.4%. Furuwa Electric Industry Co., Ltd. (Furukawa Electric Co.) and Fujikura Corporation (Fujikura Ltd.), which make cables for data centers, weakened by at least 8%.

Sumitomo Mutual Life Insurance Co., Ltd. (Sumitomo Life Insurance Co.) Masayuki Murata, general manager of Balanced Portfolio"Big tech companies have spent a fortune on AI, almost to the scale of a government budget, but there’s been a question around whether there will be returns that match their size of investments." Andrew Jackson, head of Japan equity strategy at Ortus Advisors Pte, said: "It’s very clear that anything that’s overheated like Fujikura and Furukawa are getting the wind knocked out of them."

Faced with a major blow to technology stocks, a public opinion counterattack against DeepSeek is already on the way. The subsequent actions are precisely the Ponzi scheme woven around the AI industry in the US capital market, which is being ruthlessly exposed.

3. The sharp turn of DeepSeek Wind Review

In just a few weeks, there has been a huge change in the wind direction of DeepSeek in the US public opinion field, which can be described as rising first and then declining.

When the DeepSeek - R1 model was first released, On January 27, the Wall Street Journal published "Silicon Valley Is Raving About a Made in-China AI Model", which gave positive coverage to DeepSeek. According to the article, a Chinese artificial-intelligence company has Silicon Valley marveling at how its programmers nearly matched American rivals despite using inferior chips. Although the DeepSeek- R1 model has fewer and worse chips, but it is a close rival despite using fewer and less-advanced chips, and in some cases skipping steps that U.S. developers considered essential. The article quoted Anthropic's chief executive, Dario Amodei, as saying last year: DeepSeek said training one of its latest models cost $5.6 million, compared with the $100 million to $1 billion range as the cost of building a model. As of Saturday (January 25), the two models were ranked in the top 10 on Chatbot Arena, a platform hosted by University of California, Berkeley...A Google Gemini model was in the top spot, while DeepSeek bested Anthropic’s Claude and Grok from Elon Musk’s xAI (the Grok model will be mentioned later).

The article also quoted the high evaluation of the leading enterprises in the United States. Jim Fan, a senior research scientist at Nvidia, hailed as a breakthrough the DeepSeek paper reporting the results. According to Zack Kass, a former OpenAI executive, said that DeepSeek’s advances despite American restrictions “underscore a broader lesson: Resource constraints often fuel creativity."

Bloomberg, January 28, 2025, "DeepSeek Shows China Playbook for Even Bigger US Shock on Chips", China Playbook for Even Bigger US Shock on Chips), It seriously discussed the measures that China may take in response to the escalation of the US chip blockade. The author believes that China's future technological breakthrough is to produce its own high-end chips. The article also acknowledges its success highlights some of China’s key advantages. The country has a deep pool of highly skilled software engineers, a vast domestic market and government support in the form of subsidies as well as funding for research institutes.

Of course, this article also focuses on strengthening export control and training engineers. The article quoted Purdue University's Kratch Institute for Technology, Michelle Giuda, Chief Executive of Diplomacy, said that the United States should strengthen export controls to maintain a "truly strong defense". She believes that it is necessary to catch up with China in training engineers, otherwise it will not be able to maintain its leading position in the scientific and technological race. Because in 2020, China will graduate more than twice as many engineers as the United States. "Unless we have engineers to build and design data centers and design more advanced artificial intelligence," she said. Otherwise, we will not be able to become the capital of artificial intelligence in the world.

These more positive reports on DeepSeek were quickly overshadowed by public criticism, and even the political pursuit of American politics began. After the huge impact on the US capital market, in order to maintain the value of the US stock market, it also forced the stock market to pull up.

Given that DeepSeek is open source, powerful, and cheap, public opinion attacks are first launched in the ideological arena. For example, Catherine Thorbecke, a Bloomberg columnist, criticized it in a report on January 28, 2025. DeepSeek allows users to be more and more secretly exposed to Chinese ideology. At the same time, because DeepSeek is a Chinese enterprise that abides by Chinese law, it will not wander in the ideological field (ensure its AI models' political ideology doesn't stray). Although she herself admits that the model developed by Musk's xAI is against "wake" or that Google can generate Nazi photos. It also has the characteristics of not being politically correct.

Of course, labeling Liang Wenfeng, the founder of DeepSeek, is also indispensable. On January 28, 2025, the headline of the Wall Street Journal about Liang Wenfeng was directly "DeepSeek CEO from Math Geek (Weird)". A journey from a mother to a global disruptor (DeepSeek Chief's Journey from Math Geek to Global Disruptor). Anyone who knows a little English knows that Geek and Global Disruptor are already racist when they are put together to describe a person.

Allegations of patent plagiarism against DeepSeek soon began in order to stink it out.

First up are Microsoft and Open AI. According to Bloomberg on January 29, 2025, the so-called source told the media that Microsoft and Open AI are investigating whether data output from OpenAI technology was obtained in an unauthorized manner by organizations associated with DeepSeek. Although the matter is confidential, several anonymous people said that Microsoft's security researchers began to suspect in the fall of 2024. The group used the OpenAI application programming interface (API) to steal data. If this is true, then Microsoft's security system was found to be malfeasance after six months; if this is false, then it is obviously a malicious slander against DeepSeek.

David Sacks Photographer: Anna Moneymaker/Getty Images: Bloomberg. January 29, 2025

Interestingly, the Trump administration was quick to make use of the topic. David Sacks, President Trump's director of artificial intelligence, said on Tuesday (January 28) that in an interview with Fox News, Mr. Sachs said there was "substantial evidence" that Mr. DeepSeek ripped off the OpenAI model through distillation. He did not elaborate on the evidence. OpenAI is naturally elated about this. In a statement responding to Mr. Sachs's remarks, the division said, Chinese organizations "are actively working to use methods including so-called distillation," trying to replicate an advanced American AI model.

OpenAI Says Chinese Rivals Try to "Distill" U.S. Companies' Models Source: Dado Ruvic/Reuters Financial Times Transfer, January 29, 2025

Open AI further told the Financial Times on January 29, 2025. They found it. They found evidence of some distillation at DeepSeek. The division believes DeepSeek will use distillation technology to build its own competitive model. It violates OpenAI's terms of service, just like President Trump's director of artificial intelligence, Sachs. Open AI declined to comment further or provide evidence.

At the same time, Open AI seems to have a "perspective eye" and discovered the "dark curtain" of China's artificial intelligence industry. According to The New York Times on February 21, 2025, Open AI said China is using AI-powered surveillance tools to collect anti-China posts on social media services in Western countries. Ben Nimmo, OpenAI's lead researcher, said this is the first time the division has discovered such a monitoring tool.

Now that Open AI has denounced DeepSeek on technology, anti-China politicians are not far behind. The notorious US House Select Committee on Strategic Competition between the United States and the Chinese Communists CCP President John Moolenaar and member Raja Krishnamoorthi, On January 30, 2025, he wrote to Mike Waltz, the US National Security Adviser. They called for a review of export controls against DeepSeek. In the letter, they said that the DeepSeek- R1 model uses a large number of H20 chips that are not under US export control. Therefore, it is required to implement export control on H20 or similar chips that can be exported. At the same time, they are very arrogant to say, "We must not allow Chinese AI systems to gain significant market share in the United States." Get US user data at the same time. At the end of the article, they stressed the need to strengthen the control of goods in third countries. Because in Nvidia's most recent quarterly report, Singapore accounted for 22% of the company's revenue. At the same time, most of these goods end up going to users outside Singapore. It can be said that in order to restrict the development of China's AI industry, some American politicians have shown no mercy to their allies.

So-called security companies are also involved in planting evidence for DeepSeek. According to an "exclusive" report published by ABC on February 5, 2025, code was hidden in the DeepSeek program. It has the built-in function of sending user data directly to the Chinese government. The report interviewed Feroot Security, a network security company based in Ontario, Canada. Ivan Tsarynny, the division's chief executive, said in an interview that DeepSeek could be in the code. Hiding a direct route to China.

The gentleman said that they decrypted part of DeepSeek's code through artificial intelligence software (in fact, they used special tools to crack the code). Programs found to be "intentionally hidden" are sent to the CMPassport.Com, the online registration website of state-owned China Mobile. It has to be said that this gentleman's ability to lie about China's threat is extremely limited. Because it is very common for online services to use mobile phone numbers to register users. As one of the three major telecommunications providers in China, it is normal for China Mobile to provide such services. As a compliant Chinese enterprise, DeepSeek's use of China Mobile's services is a normal business practice.

However, with the appearance of such "experts", it is convenient for the American media to put labels on them. John Cohen, an ABC contributor and former acting undersecretary for intelligence and analysis at the Department of Homeland Security, DeepSeek is the most obvious example of the Chinese government's surveillance of user. The former official went on to say, "More and more Chinese technology companies are determined to pose a threat to the national security of the United States." China Mobile is one of them. "It should be reminded that China Mobile was banned from operating in the United States in 2019, naturally on the grounds of national security issues."

Comments on DeepSeek's rise and fall are a good indication that some people not only know that DeepSeek is okay, and there is no decent argument against DeepSeek. The reasons for pan-ideology and pan-national security seem very funny, but they ignore the image. Obviously, it is an effort to maintain the narrative logic of AI high-tech stocks and forcibly raise the stock price.

Four, the political and business Pond's fraud of AI shares in the United States.

The US capital market hit brought by DeepSeek has gradually subsided in three weeks. This is not DeepSeek's problem, but the result of the huge amount of money spent by the US capital market and a lot of public opinion hedging. Those who still want to relive the Ponzi scheme are still making up AI myths.

A Ponzi scheme, as defined by Merriam-Webster, is an investment scheme. Some of these early investors are paid off by later funds to encourage more and greater risk. After the American AI myth was broken by the DeepSeek-1 model, it was continued by the Grok-3 model of xAI. It can interpret this operation very well.

Bloomberg published an article by columnist Dave Lee on January 19, 2025, the article is entitled "Musk Helps Nvidia Revive." Ending the DeepSeek Panic (Nvidia Revival Aided by Musk Marks End of Great DeepSeek Panic), the main tone is that there is no problem with the American AI myth.

At the beginning of the article, it is said that "the deep search panic in January 2025 is officially over." The reason why he is so confident is also very interesting. First, he said big US tech companies were dismissive of DeepSeek's breakthrough in big model efficiency and would not scale back spending. Second, his confidence stems from the fact that Nvidia's share price has recovered from a nearly $500 billion decline that began three weeks ago, on January 28. Shares of Nvidia rose above $143 per share in trading on Tuesday (February 18). Higher than stock price before the big sell-off brought by DeepSeek. Finally, the latest Grok-3 model from Elon Musk's xAI company marks another Renaissance for US tech stocks.

The author goes on to brag about the Grok-3 big model, while emphasizing Nvidia's big role. Grok-3, for example, is making headway just 19 months after launch, thanks to xAI's massive stockpile of Nvidia chips. XAI's Memphis data center opened with 100,000 GPU modules and could expand to as many as 1 million modules. It is worth mentioning that xAI said on its official website on February 19 that the training cluster of Grok-3 has used 200,000 modules. Meanwhile, xAI is about to sign a $5 billion contract with Dell. To produce servers that hold more NVidia chips.

The authors go on to say that NVIDIA investors should be very encouraged that its technology path will continue for the foreseeable future. DeepSeek can't shake Nvidia's path in high-investment mode. Some of the current problems are simply the result of sneaky workarounds by the DeepSeek team. So, it's nothing to Nvidia.

In another Bloomberg article on Feb. 22, the author escalated his attack on DeepSeek to the national level. In an article titled "DeepSeek Will Never Escape China's Dubious reputation". In the article, the author said that it was "historically futile" for China to convince the world that its AI products were credible. The author concluded by saying that there is no sign that China has turned a new page for the era of artificial intelligence. The above two articles are very representative of the fixed mindset of some cold war dinosaurs.

Obviously, this idea is not unique. Karoline Leavitt, the White House press secretary, said that because of intellectual property issues, the National Security Council is investigating DeepSeek. Ken Paxton, the Texas attorney general, said: DeepSeek aims to "undermine the dominance of American artificial intelligence and steal the data of our citizens.".

Now that the media and the political circles have spoken together, Nvidia, as the core of this public opinion, has to express its attitude. On February 20, in the pre-recorded program of Nvidia Partner DDN, Nvidia CEO Huang Renxun said that. Such a strong market reaction to DeepSeek stems from a misunderstanding among investors. He went on to say in the interview, "Reasoning AI is that when you ask a question, it will give you an answer immediately." "I don't know whose fault it is, but obviously the pattern is wrong."

In the context of the public relations of the US government and public opinion, Nvidia later released its expectations for the first quarter of 2025. According to Reuters on February 27, 2025, Nvidia released its strong first quarter forecast for 2025 on Wednesday (February 26). It also said that the market demand for AI chips was strong, and its Blackwell chips were "amazing" "We have successfully mass-produced the Blackwell AI supercomputer," said Jen-Hsun Huang. Sales in the first quarter were in the billions. According to the London Stock Exchange Group (LSEG), Nvidia expects total revenue of $43 billion in the first quarter. Plus, or minus 2%, compared with the average analyst estimate of $41.78 billion.

Chart represents Nvidia's revenue growth slowing Source: Reuters, February 27, 2025

However, the market did not buy it, and investors showed their attitude with their money. Shares of Nvidia closed up 3.7% in regular trading before falling slightly in volatile after-hours trading. In a Bloomberg report on the same day, Investors' reaction to the good-but-not-great earnings report was muted. After all, investors are used to the company's stunning performance. Logan Perk, an analyst at Edward Jones Purk said in a note: "Guidance is a bit flat." "We believe it will be challenging for management to continue to significantly exceed growth expectations in the future," he added. According to data compiled by Bloomberg, the company's fourth-quarter sales rose the least since February 2023; at the same time, it is also the narrowest quarter of earnings since November 2022.

However, Nvidia and the American technology industry are still creating gods. "We will have strong growth in 2025," Jen-Hsun Huang said in a conference call with analysts on the day of the earnings release. During the conference call, Jen-Hsun Huang did not forget to suppress DeepSeek. He believes that DeepSeek stimulates people's interest in AI but strengthens the market demand for Nvidia products. And that DeepSeek model, in his view, rely on fine-tuning, The number of calculations required far exceeds the "one-time" training of other software. He believes that this may require millions of times more computing power than today. His conclusion: "Future reasoning models can consume more computing resources." "And CFO Kress, the chief financial officer of Nvidia, released the news that "The Stargate data center project mentioned above will use Nvidia's "Spectrum X" Ethernet for network connectivity

DeepSeek announced on February 26, 2025, that the price of API calls will be subject to off-peak preferential pricing from 00:30 to 08:30 Beijing time. Source: DeepSeek website, February 26, 2025

Interestingly, on the same day that Nvidia released its earnings, DeepSeek issued a price reduction statement. According to Hong Kong East Net (on.CC) reported on February 26, 2025, that DeepSeek has launched an API call price off-peak discount. The daily discount period is 00: 30-08: 30 Beijing time. During this period, DeepSeek- V3 was reduced to 50% of the original price, and DeepSeek- R1 was as low as 25%. DeepSeek said the move was to encourage users to make full use of the time slot and enjoy more economical and smooth services.

Compared with the high valuation pursued by Nvidia and the universality pursued by DeepSeek, it is easy to see that the AI Ponzi scheme in US technology stocks has reached a point where it is struggling to survive.

Five, the people who vote with their feet in the American investment and technology industry.

In this context, American investors and entrepreneurs who are skeptical of AI's mythological narrative have taken a stand with their actions.

According to a report by the Securities Times Network on February 19, 2025, Bank of America (BofA, Bank of America), from February 7 to 13, 2025, fund managers from 168 asset management companies around the world were surveyed. The survey shows that the confidence of global funds in the Hong Kong stock market in China has increased significantly, and the confidence level is close to the Nasdaq Composite Index. On February 18, US Eastern Time, Chief investment strategist at BofA Global, Chief Investment Strategist, Bank of America Global Research) Michael Hartnett's team said in the report, in the survey, 18% of respondents predicted that the Hang Seng Index would perform best in the world in 2025, ranking second with the Nasdaq. This is the first time that the Hang Seng Index has entered the top three since it first appeared in the monthly survey in November 2024. "Many of the investors we interviewed continued to see China last year as a 'trading market' – a place to make quick money," the report added. In the past, it chased a trading rebound, while global long-term capital still did not believe it and seldom participated in it. But it is encouraging that we now believe that China's basic investment theory has been improving.

The report also quoted Goldman Sachs strategists as saying in a recent report that the DeepSeek model and other Chinese AI models. Has changed the narrative of Chinese technology. Optimism about confidence and market prospects may raise the valuation of China's stock market by 15% to 20%. And more than $200 billion in portfolio inflows.

According to the report of Securities Times Network on February 20, Morgan Stanley finally abandoned its bearish view on China's stock market. He upgraded his MSCI China Index rating from "underweight" to "balanced". At the same time, the MSCI China Index is expected to increase by 22% to 77 points. At the same time, the Hang Seng China Enterprise Index is expected to rise from 6970 to 8600, and the Hang Seng Index is expected to rise from 19400 to 24000.Laura, strategist at Morgan Stanley "China's stock market, especially the offshore market, has finally undergone a structural qualitative change," Wang and her colleagues wrote in the report. "They further indicated that the Trump administration did not levy taxes on China alone as the war between Russia and Ukraine eased." With AI companies such as DeepSeek maintaining their leading position, the P/E ratio of MSCI China Index can rise to 10 to 12 times in the next 12 months.

JPMorgan Chase also said that thanks to the development of DeepSeek's generative artificial intelligence, the rise of Chinese technology stocks is more sustainable. The agency believes that the average return of China's stock market in the next 10 to 15 years is 7.8%. J.P. Morgan believes that even if there is a massive rebound, Chinese stocks have attractive valuations, and its view on Chinese stocks is "overweight".

China's capital market has always welcomed long-term foreign investment. According to Radio Television Hong Kong on February 17, 2025, Pei Huiqi, deputy director of the International Cooperation Department of the Shenzhen Stock Exchange, at the Annual Conference of the Asian Securities Industry and Financial Markets Association (ASIFMA) According to the seminar, an international investor conference is expected to be held in Shenzhen between May and June 2025.The aim is to open up a smoother way for international investors to invest in China. Many of last year's initiatives have received positive feedback from international investors.

Foreign investors have naturally received signals from the Chinese side. A columnist for the Financial Times, Ruchir Sharma, also chairman of Rockefeller International Sharma published an article in the Financial Times on February 24, 2025, entitled "Is China investable again?"

The authors say China has always been worth investing in. He made it clear that "investors are wrong to think that the world's second largest market is uninvestable." "Shamal also mentioned the substitution of DeepSeek model for similar products in the United States, which has brought about a huge change in market sentiment." He also believes that geopolitical risks are receding instead, which many people exaggerate.

The Chinese stock market has become unreasonably cheap, he said. Although the stock market rose sharply in 2024, the valuation of China's stock market was only half of the average valuation of the United States in 2025. Shamal then listed a data: there are more than 250 Chinese stocks with a market value of more than $1 billion and a free cash flow yield of more than 10%; by comparison, there are less than 150 in the United States. Of the more than 250 stocks, only about 20 are technology stocks. The rest are industrial and consumer discretionary companies, so China's investment opportunities go beyond the Internet and artificial intelligence. Of the 11 leading sectors, 7 in China are less concentrated than United States. That is to say, the top five companies in the sector have a lower share of the market value. That's why startups like DeepSeek can rise. As long as we do a good job of risk control, China can invest at the right price.

QUALCOMM also represents that American technology and industry are not all obsessed with the mythological narrative of AI technology stocks. Cristiano Ammon, President and CEO, QUALCOMM, in an interview with Xinhua on February 21 about the white paper, the recently released DeepSeek-R1 model and other similar models show that AI models can be smaller, stronger and more efficient. And can be run directly on the terminal side. In the future, AI will become more convenient, customizable and efficient.

In the Chinese AI white paper mentioned above: "AI change is driving innovation in terminal-side reasoning," QUALCOMM said, after entering the era of AI reasoning, reasoning will run more and more on the terminal side. DeepSeek model represents the innovation of high-quality small language model and multi-modal reasoning model. It can prepare for the commercial use of AI and the landing of terminal-side reasoning. At present, there are four major trends in AI models that can run on the terminal side: 1) the current advanced AI small models have excellent performance; 2) The scale of parameters is shrinking rapidly; 3) Developers can build richer applications on the edge; 4) AI is becoming the new UI. The disruptive change and reassessment of DeepSeek - R1 model also verifies the trend that AI pattern is about to change to large-scale reasoning.

Comparison of LiveBench AI benchmark average results of Meta Llama 70 billion parameter model and DeepSeek corresponding distillation model. Source: Live Bench.AI, February 2025, "AI Change is Driving Terminal Side Reasoning"

The white paper further explains the distillation technology used by DeepSeek. As a key technology for developing efficient small models, distillation technology enables large models to "teach" small models. Transfer knowledge while maintaining accuracy. The report cites the average test results of Llama's 3370 million parameter model and similar DeepSeek - R distillation Live Bench. To prove the power of distillation. The results show that distillation can significantly improve performance in reasoning, programming and mathematical tasks at the same parameter scale.

As a result, large models can be reduced to achieve faster inference, a smaller memory footprint, and lower power consumption. While maintaining a high level of performance. AI models can also be deployed on terminals such as smartphones, PCs and cars. Distillation of small models can often be put into terminal operation within a few weeks or days. According to the white paper, for example, the DeepSeek - R1 distillation model was completed in less than a week. It will run on PCs and smartphones powered by Qualcomm's Snapdragon platform.

The QUALCOMM is promoting AI reasoning model in the edge terminal industry. This is a huge market for billions of smartphones, cars, XR headsets and glasses, PCs and Industrial Internet of Thing terminals. Major manufacturers such as Samsung, Asus, Millet, OPPO, vivo and Glory. The AI-driven function has been applied to the latest flagship smartphone with the Miaolong platform. This means that other technology companies in the United States, represented by Qualcomm, have begun to deploy models efficiently in practical applications on the edge side.

Six, the combination of AI model and entity can create real value.

Deng Xiaoping, the leader of the Communist Party of China, has a famous saying: "Science and technology are the primary productive forces." "After China's reform and opening up, this sentence is often used by Chinese governments at all levels to guide national construction and scientific and technological development." After the launch of DeepSeek model, all walks of life in China with industry as the core are accelerating their landing.

According to the report of Shanghai Securities News on February 18, 2025, many government systems in China are connected to DeepSeek model. Urban computing networks in Shenzhen, Guangzhou, Wuxi, Nanjing, Suzhou, Zhengzhou, Hohhot and other places. The large models of DeepSeek series (DeepSeek- R1 (671B), DeepSeek - V3 (671B)) were successively announced to be launched, and the localization deployment was completed. In the future, DeepSeek model will not only play the role of assistant but also rely on massive data and powerful computing power. To analyze, judge and determine the nature of various events.

Among them, Shenzhen is more advanced in promoting DeepSeek. According to the report of China Youth Daily on February 20, 2025, there are 70 "digital intelligent employees" in 11 categories in Futian District of Shenzhen. These AI "Digital Intelligence Employees" based on DeepSeek have clear support and framework. "Digital Intelligence Staff" is the assistant of civil servants, serving as an assistant. They do not have the ability to make decisions alone, and the department appoints the guardian to guide the work and is responsible for the whole process management during the service period.

The AI "Digital Intelligence Employee" launched by Futian District is based on DeepSeek, which is the 2.0 version of Futian Government Model. The technical base is a full-size DeepSeek- R1 model. "Digital Intelligence Employees" can provide services in 240 government work scenarios, such as official documents, people's livelihood, emergency response, business and other fields. Each "Digital Intelligence Employee" is an independent account, including ID number and name. It has the right of name and personality and has a series of personification characteristics such as personification characteristics, emotional intelligence and cultural adaptability.

"Digital Intelligence Staff" has achieved remarkable results in improving efficiency. A law enforcement document that takes 4 hours to complete can complete the first draft in 4 minutes. With the help of DeepSeek model, the accuracy of document correction is over 95%, and the audit time is shortened by 90%. The accuracy rate of people's livelihood appeal allocation has increased from 70% to 95%, and the first draft of the People's Daily Weekly has been generated by one key. This has reduced the burden of grass-roots units with fewer people and more repetitive affairs.

At the same time, DeepSeek also plays a role in the optimization of computing power centers. According to the Securities Times Network on February 21, 2025, the DeepSeek model has accelerated the transformation of the Intelligent Computing Center. Transform from computing power provision to professional services and improve the utilization efficiency of computing power resources.

Gu Licheng, solution architect of Zhonghao Xinying, said in an interview that after the DeepSeek model became popular, many customers are asking about the adaptation of the company's intelligent computing center to DeepSeek. Its intelligent computing center has been able to run various distillation versions of DeepSeek model, and the performance is excellent. PatSnap, a technology innovation intelligence SaaS service provider, also uses the DeepSeek-R1 model. The company's technical director told reporters that open-source models such as DeepSeek can effectively reduce the cost of pre-development. In particular, it saves the high cost of trial and error in infrastructure. Bai Xu, vice president of Boda Data, said that Chinese computing chip manufacturers cooperated with OEM (original equipment manufacturer) manufacturers to launch DeepSeek integrated machine. Major cloud platforms in China have also launched DeepSeek model services, which greatly reduces the threshold for the use of large model applications.

Before the DeepSeek model emerged, the industry had doubts about the proliferation of intelligent computing centers. According to the Assessment Report on the Development of Artificial Intelligence Computing Power in China recently released by International Data Corporation (IDC) and Wave Information, in 2024, the scale of China's intelligent computing power increased by 74.1% year on year. Such a large construction has also resulted in the low utilization rate of many intelligent computing centers and the mismatch between supply and demand. The government encourages "moderate advance" to lead to the construction of large-scale intelligent computing centers, but the demand has not blowout.

The emergence of DeepSeek helps to solve the problem of idle computing power. Shen Jiaqing, vice president of Shanghai Jingyi Industrial Digital Intelligence Research Institute, said, "By providing professional services covering the whole life cycle of computing power." Cooperate with different subjects in the industrial chain, it provides a package of solutions including computing power scheduling, measurement, optimization, fine-tuning, packaging, maintenance, secondary development, etc. It will effectively promote the penetration and application of computing services in all walks of life, thus expanding the market demand of the whole intelligent computing center.

The central government has also taken positive action in the face of the wave brought by DeepSeek. According to a report by the SASAC of the State Council on February 21, 2025, on February 19, The SASAC of the State Council held a meeting on deepening the deployment of the "AI +" special action for central enterprises. Zhang Yuzhuo, secretary and director of the SASAC Party Committee of the State Council, said at the meeting that we should accelerate the high-quality development of the artificial intelligence industry. Achieve more landmark achievements and breakthroughs. The meeting emphasized that the central enterprises should seize the strategic window period to tackle key problems, participate in the construction of open ecology, and promote original innovation. Promote the transformation and industrialization of achievements. It has high strategic significance and economic benefits, but it is related to people's livelihood.

The meeting asked the central enterprises to focus on the development of artificial intelligence in the "15th Five-Year Plan" and create more leading enterprises in science and technology. Incubate and cultivate a number of start-ups. Long-term capital, strategic capital and patient capital to support the development of artificial intelligence industry. At the same time, we should provide a better environment for the development of talents, so that talents can concentrate on scientific research.

Seven, the dream is fragile and the future is right.

China and the United States have completely different development paths in the AI industry. The two have completely different ideas.

The American model is the game of a few big capitals, aiming at pursuing returns and attracting more funds. The assumption of this logic is based on the absolute technological superiority of the United States. The emergence of disruptive companies like DeepSeek makes their narrative logic difficult to reconcile. Because of massive capital investment, the market is extremely sensitive to negative signals. The narrative of AI myth, in essence, cannot withstand any impact. In this case, the use of public opinion and political resources for crisis public relations has become inevitable. In order to ensure that there is no avalanche scenario in the capital market, we can only invest more money and introduce new concepts to attract investment. It can even be said that its purpose has been simplified to maintain the situation.

However, the US capital market has never been a dominant one. Those investors or institutions who are coerced to participate in raising the price of capital will not give up any opportunity to spread the risk. This internal divergence will aggravate market instability and lead to the need for more capital investment to maintain the situation. In the AI industry, a new Ponzi scheme has been formed, and there is no visible solution at present.

The Chinese model is a combination of industrial landing. DeepSeek model is a good tool for government and enterprise departments to improve efficiency. From the country to the enterprise, the purpose is to improve productivity and life, so that the landing of DeepSeek model becomes more operable. As the AI model is still in the early stage of deployment, it combines China's huge population and perfect real industry. There are quite a number of fine application scenarios that can become the blue ocean market.

In the future, all kinds of refined and customized distillation models will also bring powerful new growth points to economic development. Under the strong promotion of the central government, the local government and enterprises will also accelerate the integration of industries with AI.

The greatest value of DeepSeek-R1 is the narrative logic of destroying the financial hegemony of the United States. And that's the most important thing to remember.

If you need to communicate with the author after reading, you can write an email to [email protected]