China’s AI at the crossroad: dream less, profit more!
Not just about scientific and technological progress; it's a story of global rivalry, national identity, and a potential cornerstone for future economic growth
Since late 2022, the global AI frenzy has sparked widespread anxiety among nations, regions, and businesses. In China, this landscape brims with external pressures and internal strife. While the U.S. tightens its hold on China with supply chain constraints and capital controls, domestic internet giants scramble for chips and computing power.
Emotions run high, swinging between fear and complacency. Take Kimi, a Chinese AI platform; its text processing prowess, claimed ten times that of GPT 4.5, has sent its valuation soaring to $2.5 billion, igniting a surge in related stocks this March.
Yet, this isn't just about scientific and technological progress; it's a story of global rivalry, national identity, and a potential cornerstone for future economic growth. While domestic venture capitalists push for Chinese AI breakthroughs, a sober analysis warns against blind optimism. It highlights the United States' advantages in talent, capital, and hardware, points out a thoughtful approach to China's AI and high-tech industries development, and sensibly addresses why Chinese companies should be less anxious, if not optimistic, about their AI future.
Below is Baiguan's translation of the original article (some paragraphs are abbreviated or redacted):
Sheer anxiety is what many express when discussing AI this year, with investors and publicly listed companies alike feeling the fear of missing out (FOMO) and losing sleep over the vast technological gap between China and the United States.
Yet, this is precisely the time for optimism.
When drafting feasibility reports for projects in the finance sector, we typically explore three scenarios in the future: pessimistic, neutral, and optimistic.
The pessimistic forecast likens the current AI boom to the metaverse—a bubble waiting to burst, with Nvidia potentially meeting the same fate as Cisco. Or, the situation where America's AI advancements mirror China's in 5G—undeniably ahead, yet “burning” through cash without a blockbuster application in sight.
The neutral perspective sees this AI wave akin to the Internet's rise: a steep technology curve in the early days that eventually flattens, with China catching up over time. This scenario reiterates the notion that while the U.S. excels in foundational technology, China shines in application and implementation.
The most optimistic outlook suggests that this era of AI will lead to AGI (Artificial General Intelligence), with scaling laws ushering in super-intelligent AI, reminiscent of scenarios from "The Matrix," where humanity is outpaced and enslaved. However, even this should not incite panic but rather celebration. Historically, only in facing existential threats have China and the U.S. come together, whether it was against the formidable Soviet Union or 9/11 terrorism. AGI may be the next catalyst for cooperation.
Throughout history, humanity rarely solves its problems outright; instead, we tend to outlast them until a newer, more significant issue overshadows the old.
America's edge: talent, capital bubbles, and technological foundations fuel disruptive innovations
It is often said that China's systemic advantage lies in its ability to act with a unified national strategy. Similarly, we must acknowledge that the United States holds its own systemic strengths, notably in fostering disruptive innovation.
The brightest minds from the globe to the U.S. pursue unimaginable ideas. From integrated circuits to operating systems, from the internet to machine learning, disruptive innovations have consistently originated in the United States, supported by the absolute protection of private property, a liberal environment, and a culture that encourages capital to chase profits.
The current AI wave highlights the gap between China and the U.S. from three perspectives: talent, capital, and technology.
From a talent perspective, James Ding, the managing director of GSR Ventures, once remarked that whereas previously, over 70% of AI graduate students from Silicon Valley would return to China, now virtually none do.
From a capital viewpoint, the U.S. is currently experiencing a capital bubble. With the economy strong and an abundance of freshly printed dollars seeking investment opportunities, the narrative of AI has conveniently become Silicon Valley's next big gamble.
Technologically, Jensen Huang's Nvidia stands out as a formidable leader, setting the standard with its overwhelming lead.
In the short term, it seems improbable for these dynamics to reverse. This dictates that China's approach to AI should be to leverage its own strengths and forge its own path.
Interviews with Allen Zhu, managing partner at GSR Venture, have gone viral for his candid views on focusing not on large models but on AI projects that can generate profit, noting his investments in companies that operate with a single GPU card. Though such comments have drawn criticism for lacking idealism or conviction, Zhu's blunt truths resonate more today than they might have in the past. A closer look at China's entrepreneurial landscape, investment and financing, macroeconomics, and the themes of our times reveals a scene vastly different from six or seven years ago.
Dollar funds retreat, national teams advance: China’s high-tech sector at a crossroads
The U.S. containment of China began with supply chains, covering well-known elements like the Entity List, advanced semiconductors, and photolithography equipment. The next phase was a comprehensive decoupling, extending beyond supply chain restrictions to higher and lower market levels.
At the lower end, the market itself was targeted. The ban on TikTok sent a clear message: the U.S., the world's largest consumer market, is closing its doors to tech products with Chinese affiliations. At the higher end, the capital markets were affected. Leading U.S. dollar funds such as Sequoia Capital, GGV Capital, and Walden International have been scrutinized multiple times, effectively severing the flow of U.S. dollar investments into China's high-tech industries.
U.S. dollar funds had the advantage of sharing risks with founders. They were professional, tolerant of mistakes, and willing to experiment. This is what we call “long-term quality investing.“
U.S. dollar funds were the mainstream of Chinese venture capital during the internet and mobile internet waves. Now, they are a rarity. The primary market in China is currently dominated by CNY, with government-guided funds and state-owned capital leading the pack. This shift from Texas Hold'em to Guandan in the social activities within investment communities is notable.
Still on its path to professionalization, state-owned capital has a low loss tolerance. Investments from government-guided funds often prioritize project reinvestment()*, measuring success by local employment increases and GDP growth.
*Baiguan Note: “Project reinvestment" involves a private equity fund investing in a project with the agreement that the project will establish a presence in a specific administrative region, typically where the fund is based, after receiving the investment. This aims to attract high-quality projects to the local area, fostering industry transformation and development.
Among the dozen-plus funds I have recently met, only one insists on market-based fundraising, steadfastly avoiding state funds. The rest have adapted to the times. One PE firm shared with me that they feel their role has shifted from investing to business recruitment since accepting government-guided funding.
In the interview, Allen Zhu mentioned he only invests in profitable ventures, eschewing cash burn. It is not just his conservative stance; the entire Chinese investment community shares this trend. Avoid risks, avoid losses, and avoid complicated explanations.
A decade ago, the capital was willing to burn through cash, fueling the battles between group buying services and the subsidies war between Didi and Kuaidi Dache (快的打车; meaning "Fast Taxi"). New ventures were supported with money first, growing big while figuring things out. That era has passed. Unlike five or six years ago, when Alibaba's valuation surpassed Amazon's and Tencent's exceeded Facebook's, encouraging everyone to chase new ventures, with Alibaba and Tencent as potential acquirers. Post-antitrust regulations, that is no longer the case. Several past experiences have led to a new understanding among investors, summed up by a seasoned professional: initiatives related to the lives of the masses and vulnerable groups are not welcoming to capital.
However, state-owned capital can also serve as “long-term quality investing.” In China, this privilege was afforded to 5G. The country took the lead in the national effort with Huawei's endorsement and a grand narrative. This contrasts with AI, where U.S. public opinion is anxious about falling behind China in 5G and high-speed rail. Then, U.S. industry and capital explained that different national conditions dictate focusing only on profitable ventures. If 5G and high-speed rail aren't profitable, sticking with existing infrastructure is acceptable. Interestingly, the AI and 5G situations between China and the U.S. are reversed, reflecting different national missions for each generation.
Beyond brute force: the future of China's AI lies in 'application'!
In the artificial intelligence (AI) arena, it’s not the martial arts that matter, but the ideas: China's future in AI lies in "application"!
The current state of play with large AI models is such that proprietary models, epitomized by OpenAI, are a year or two ahead of open-source counterparts, leading Chinese models by a similar margin. In the hardware domain, Nvidia is in a league of its own, its dominance unchallenged. With over three decades dedicated to maximizing performance and accelerating computation, Jensen Huang's latest unveiling at GTC pushes the worship of computing power to new heights.
A seasoned CTO with over a decade of dealings with Nvidia offered a vivid analogy: using Qualcomm chips in mobile phones vastly reduces the optimization work compared to MediaTek chips, suggesting Nvidia's ecosystem in AI chips far exceeds any cost competition. Nvidia's profit margins, surpassing those of luxury brands like Louis Vuitton, underscore its stronghold. Despite the hype around domestic alternatives to Nvidia’s GPUs, the reality falls short of expectations.
The conclusion is stark: from hardware to the ecosystem, Nvidia is set to remain peerless in the AI arena for the next three to five years. AI represents a formidable scientific contest and a ruthless financial struggle. The challenges for China's AI ambitions include: firstly, a lack of long-term investment akin to U.S. dollar funds and, secondly, the inaccessibility of Nvidia's top-tier GPUs.
This reality consolidates the mainstream view, as represented by Allen Zhu, that for most aspiring AI entrepreneurs in China, practical application is key, whereas the idealistic path espoused by Zhilin Yang is for the few(Baiguan: Zhilin Yang is the founder of Moonshot AI and creator of the Kimi chatbot model).
In an interview, Allen Zhu shared his concerns about ventures like Yang's commercial viability.
We invested in his previous company. He is very impressive, and large models really suit him well. He is capable of conducting research, but I'm not sure how he will commercialize it. Their company (Moonshot AI) is leading in the field of large models domestically, but in the long term, they still need to prove their value, at least catch up with the open-source projects in the United States. If they can surpass open-source, then his team is truly valuable.
However, Zhilin Yang and Allen Zhu are not fundamentally opposed. Significant model development is viable in China, grounded on two realities: firstly, cloud computing giants like Alibaba are willing to invest heavily, leveraging their computing power for infrastructure. Secondly, the necessity for China to achieve autonomy in critical technologies, regardless of their initial effectiveness, is undeniable.
The opportunities belong to exceptional talents like Zhilin Yang, who has contributed significantly to the field with Yann Le Cun. Yet, for most, the advice of Allen Zhu rings true: focus on applications and practical monetization strategies to sustain ventures.
Acknowledging these gaps allows for a proper mindset adjustment, fostering confidence. By focusing on catching up and applying technology, even discussing AI applications becomes a high threshold globally.
Echoing the sentiment of "The Grandmaster," today, we compete not in martial arts but in ideas.
Conversations with the old money about cutting-edge technologies like AI and GPUs often reveal underlying anxieties. Yet, when it comes to their traditional industries, there is an inherent confidence. China has a peculiar knack for transforming groundbreaking American technologies into mainstream commodities within three to five years. Recent interactions with listed companies in sectors like LiDAR, electric vehicles, robotics, and image recognition sensors—innovated initially in the U.S.—illustrate how Chinese companies rapidly commoditize these technologies. Despite international efforts to suppress Chinese goods, the trend of domestic replacement and global market penetration persists and accelerates.
In the realm of autonomous driving, Wang Chuanfu's initial skepticism has been counterbalanced by BYD's quiet ascendance as a manufacturing leader. BYD's eventual pivot to smarter solutions reflects a broader pattern that could apply to AI's evolution in China.
While disruptive innovation might seem out of reach in the short term, my confidence in China's potential is buoyed by the diligence of Chinese entrepreneurs and the steadfastness of its manufacturing sector. This relentless progress, step by step, is seemingly unstoppable.
From the ground up: AI's role in catalyzing industry-wide revolutions and jumpstarting economic expansion
Delving into AI has led me to ponder: from AI's perspective, what might it make of humanity? Likely, it would find our arrogance hard to endure.
AI has always been exciting since its inception at the Dartmouth Conference. With every minor advancement, the industry reaches a climax, proclaiming that AI dominance is just around the corner and humanity's end is nigh, each time insisting, "This time it's different." Yet, invariably, expectations soar too high and unmet, and the industry plunges into winter, only to repeat the cycle.
This time around, from ChatGPT to Sora, the surprises keep coming. Yet, fundamentally, they are all based on Transformer technology, which leverages parallel computing to accelerate the training of deep learning models vastly. From chatbots to generative video, it all boils down to computers predicting the next word/pixel after extensive data training. However, our understanding of the human brain's workings hasn't advanced in the slightest, nor have we found new ways to imbue semiconductors with consciousness and intelligence.
To put it soberly, the rise of AI still leaves us with much more to do than just boasting and selling courses. Despite the emergence of large models, a solid, applicable, ground-level application has yet to appear. Given the inexplicability of large models and the uncontrollability of their outputs, how new technology can empower productivity, enhance production efficiency, transform countless industries, and reboot economic growth remains a colossal task. We have only taken the first step on a long march.
After every due diligence round, over meals and drinks, I emphasize two things.
First, do not peddle anxiety. I have grown to despise the phrase "an era has ended." A quick search through my WeChat moments last week revealed the end of seven such "eras." Amidst significant upheaval, there is no room for such sentimentality. Nor do I care for the term "the second half of XX." Anxiety won't help; your anxiety will never match that of Cathie Wood, who missed out on Nvidia.
Second, we must commit to the game. Before this AI wave, the acknowledged dividend was mobile internet. Recalling early investments in ByteDance, Pinduoduo, and Kuaishou, the tales shared over drinks won't be recounted here, but media coverage reflects the initial skepticism towards Yiming Zhang and Colin Huang. The evolution of new technology is unpredictable, even with extensive knowledge. But entering the fray is essential. From Zhang and Huang, we see that the ultimate beneficiaries of a boom are not always apparent from the start but are certainly those who entered the arena early, enduring the front lines of industry progress. I boldly predict that the heroes made by AI in twenty years won't be the highly-regarded Jensen Huang and Sam Altman of today but someone else, unknown to us now, yet undoubtedly already in the game.
Exaltation in success and despondence in downturns is not our way, my friends. Let's not be those whom AI would look down upon.
Our take
In the crescendo of technological tides,
Amidst the clamor of bytes and bits,
Prophecies whispered in silicon dreams,
And emotions dance to digital rhythms.
Yet beneath this tumultuous symphony,
How much truth lies veiled,
By the shadows of power and capital,
In the echo chambers of national pride?
How many feats, like stars in the night,
Are etched in history's boundless scroll,
While others, like secrets whispered in the wind,
Remain hidden in the annals of time?
Above is a poem crafted by ChatGPT when I input the key elements from the articles: technology, prophecy, public opinion, emotions, national power, capital competition etc.
Technologically speaking, AI's prowess lies not in grand gestures, but in the subtle prediction of the next character, born from the essence of human language and knowledge. It's as if AI, drawing from our collective wisdom, whispers of our fleeting existence amidst the boundless truth. Indeed, we know so little about ourselves - humanity must first study itself, or risk falling into internal strife and mutual attacks.
Back to China's story, China's AI journey is winding and long, facing a significant gap with the U.S. Yet, amidst these challenges, coupled with the hurdles in China's venture capital landscape, meaningful progress in this industry may take time. This prevailing sentiment suggests caution for investors seeking opportunities in this field. A wait-and-see approach might be prudent.
However, China's AI trajectory may diverge from that of the U.S. Regions with advantages in talent, capital, and knowledge (like the U.S.) are hubs for innovation and disruption. Conversely, regions facing technological blockades, navigating cautiously (like China), yet possessing vast and homogenous markets, diligent entrepreneurs, labor forces, and a strong-force and dedicated government, hold the potential for grounding and disseminating new technologies for the common good.
For an external observer, staying abreast of the latest developments is crucial.