Sora Concept Ignites AI Sector, Beware of Sora Concept Stocks
-
Sora has emerged suddenly, capturing global attention and investors' interest.
On the first day of the Year of the Dragon, AI-related stocks in the A-share market surged as expected, with over 100 companies seeing gains exceeding 10%. Will AI repeat the success story of the first half of 2023?
On February 15, Open AI unexpectedly announced Sora, causing a significant stir.
Open AI describes Sora as a "world simulator." Musk stated bluntly: Humans admit defeat.
Former Alibaba VP Jia Yangqing commented on Sora: It's truly amazing.
Many enthusiasts even exclaimed: The real world no longer exists.
What exactly is Sora? Simply put, it is a large model for generating videos from text. This might not seem like a big deal, as similar large model products have been emerging one after another even before OpenAI's launch, such as PIKA, Runway, SVD, Genmo, and Moonvalley.
However, Sora is capable of outperforming its competitors by a significant margin.
First, it extends video duration from the current mainstream few seconds to one minute, entering the time range suitable for short video creation. Second, it supports multiple camera angles within a single video while maintaining consistency. Furthermore, it can generate complex scenes with multiple characters and specific movements, demonstrating a certain level of understanding of the physical world. Its astonishing effectiveness stems from standing on the shoulders of giants, utilizing two mature technologies in the market (Transformer + diffusion models). Moreover, it integrates its own GPT model paradigm, where user-input prompts are first expanded in detail by GPT before being passed to Sora for frame-by-frame generation of more precise videos. This gives it unique advantages.
Sora's competitors, lacking GPT's underlying text expansion processing model, remain stuck at text-to-text and text-to-image generation, creating an insurmountable gap with Sora.
Sora's emergence marks a milestone advancement. It will profoundly transform content creation fields such as short videos, gaming, entertainment, film and television, animation, advertising, marketing, news, and social media in the future. The short few-minute videos we see from brands on TV or online are traditionally priced at millions of yuan by advertising agencies, which also require substantial investment in costs and time to produce. In the future, leveraging Sora could drastically reduce these expenses for brands. For advertisers, this might mean a complete reshuffling of their business models.
As a result, Sora has received high praise from the outside world.
Even so, the impressive Sora still has significant flaws. While Sora shows considerable improvement in understanding the real world, it remains fundamentally based on training and computation with massive datasets rather than possessing an intrinsic understanding of the world, let alone matching the imaginative capabilities of the human brain. Therefore, Sora-generated videos still have many limitations, even including some factual errors, and cannot understand causal relationships that require imagination. For example: a person biting a cookie leaves no bite marks; the direction of a person running on a treadmill is chaotic; a glass falling to the ground only deforms on one side.
In recent days, some enthusiasts have exclaimed that Sora surpasses Runway and Pika, becoming a "universal world model." Clearly, this overly mythologizes Sora—it still has a very long way to go before achieving AGI (Artificial General Intelligence) or becoming a world model.
The so-called "the real world no longer exists" is certainly not happening now. No one can deny the vast potential of artificial intelligence, but it's crucial to remain clear-headed at present. The terrifying Sora has ignited the AI concept in A-shares, bringing the major theme back into the spotlight. But how sustainable is it?
Looking back at history, let's first review the previous major AI market trends.
At the beginning of 2023, ChatGPT took the world by storm, marking the beginning of the artificial intelligence revolution. In response, the U.S. capital market was generous, granting high valuation premiums to the true beneficiaries. For example, NVIDIA's PE ratio reached as high as 95 times. After ChatGPT took the world by storm, the AI sector in China's A-share market seemed to mirror the frenzy seen in US stocks. As AI stocks soared in the US market, related AI concept stocks in A-shares also began to stir.
Especially in the first half of 2023, AI-related stocks in A-shares experienced a meteoric rise. In just a few months, 86 related concept stocks doubled in value. Among them, Hongbo Co., Ltd. surged over 500%, while Zhongke Information and Wanxing Technology rose more than 400%. Remarkably, these gains were achieved against the backdrop of a generally sluggish A-share market.
However, from the second half of 2023 to the present, the AI sector has retreated to its original state, not only erasing all previous gains but also falling significantly below pre-ChatGPT levels. From this perspective, the AI rally in A-shares appears to have been more about hype and speculative trading than substantial growth. Cambricon serves as a typical example. Since financial data became available in 2017, the company has never reported a profitable year. Over the past four years, it has accumulated losses of 4.5 billion yuan while generating only 2.499 billion yuan in revenue, with extremely sluggish growth. In the first three quarters of 2023, revenue plummeted by 44.8%.
With such dismal fundamentals, Cambricon's true colors quickly became apparent as its stock price nosedived, suffering a maximum decline exceeding 60%. However, on the first trading day of the Lunar New Year, its shares surged 16% amid the Sora-related news, delivering a remarkable performance.
Another case is Kunlun Wanwei, which soared 17% on February 19. The company faced widespread skepticism in 2023 over its founder Zhou Yahui's ex-wife Li Qiong's share sell-off, suspected to be "fake divorce for real减持". Since its 2015 ChiNext listing, Kunlun has aggressively pursued acquisitions and overseas expansion, attempting to sell a "gaming + platform + investment" capital story to the market, yet its stock performance has remained lukewarm. In 2023, stock prices soared on the back of AI hype, but the underlying fundamentals remained weak, with profits declining sharply for three consecutive years. Subsequently, stock prices plummeted by nearly 60%, trapping a large number of investors.
From February 19 to 20, Chinese Online saw consecutive 20CM limit-up surges due to its tangential connection with Sora. However, its financial performance remains dismal.
Therefore, the AI-driven market rally sparked by Sora, much like the previous one, appears to be more of a speculative bubble, with questionable sustainability. Moreover, while Sora represents a significant upgrade to ChatGPT, its market impact is unlikely to match the initial hype of ChatGPT. The speculative momentum may not last long, likely serving as a short-term valuation correction based on news-driven sentiment. After all, artificial intelligence has been hit too hard in recent months, with P/E ratios dropping from 72x to 40x. Some core leaders even saw their P/E ratios fall below 20x.
▲ AI sector P/E trend chart Source: Chocie
Of course, Sora's explosive popularity will still help some excellent A-share companies stand out, presenting considerable investment opportunities. The emergence of Sora marks a shift in the competition among large models from text and images to video, significantly increasing AI training demands and exponentially boosting the need for computing power. Therefore, companies that provide computing power—'selling shovels'—should be the first to benefit.
Previously, some institutions predicted that the global total computing power would grow from 615 EFlops to 56 ZFlops by 2030, an increase of nearly a thousandfold, with an annual compound growth rate of 65%.
This projection may not be overly optimistic. Just days before Sora's debut, OpenAI CEO Altman was reportedly in talks with investors, including the UAE government, to raise up to $7 trillion to build a global semiconductor chip supply chain. What does $7 trillion mean? Altman could acquire Nvidia, AMD, TSMC, Broadcom, ASML, Meta, Samsung, Intel, Qualcomm, and more in one go.
Whether the rumor is true or not, it reflects the potentially immeasurable demand for AI computing chips in the future.
Of course, China's domestic computing industry chain will also experience high growth. According to data from Southern Wealth Network, there are 52 computing-related companies listed on the A-share market, with total revenue reaching 1,043.58 billion yuan in 2022, an 8.5% year-on-year increase, marking five consecutive years of growth. Net profit attributable to shareholders was 43.064 billion yuan, down 2.6% year-on-year but still over 70% higher than in 2018. This occurred even before ChatGPT went viral. With Sora's outstanding generative capabilities potentially sparking a domestic text-to-video AI model boom, major domestic companies are expected to quickly follow suit, thereby increasing the demand for computing power.
The AI computing power race is inevitable, particularly impacting the demand for optical components. A recent securities research report pointed out that high-end computing power modules are in short supply. With Amazon, Google, and others gradually deploying self-developed AI chips, the demand for 400G and 800G optical modules is rising. In 2024, global AI investments will expand from North American cloud giants to Tier 2 cloud providers, large enterprises, and intelligent computing centers, driving the recovery of the global optical module market. In China's A-share market, Zhongji Innolight stands as the global leader in optical modules and one of the few companies in the computing power sector capable of delivering tangible results, with its net profit soaring 120-fold over six years. It holds the top market share in high-speed optical modules and significantly outpaces competitors like Eoptolink and Accelink in R&D investment. Leveraging its core product competitiveness, the company has secured major clients including Google, Amazon, Huawei, and ZTE, with overseas revenue accounting for approximately 70%.
On February 19, market rumors suggested that after testing 800G silicon photonics, Nvidia found Innolight's performance overwhelmingly superior and placed a small batch order.
▲Zhongji Innolight's Net Profit Attributable to Parent Company (Source: Choice) In the first half of 2023, Zhongji Innolight's stock price skyrocketed by over 500%. Although it later experienced a correction along with the industry to digest valuations, it far outperformed the industry average. After hitting bottom in November, the stock has rebounded more than 80% (with a 20% surge on February 19), now just one step away from refreshing its historical high.
The AI juggernaut continues to roll forward. Embracing this burgeoning industry may not guarantee the optimal choice, but it certainly isn't the worst. However, from a capital market perspective, few companies like Zhongji Innolight can deliver solid performance—most A-share AI companies are merely riding hype, and their stock prices will eventually return to dust.