Written by Yang Jianyong

Taking copper as a mirror, one can straighten one's attire; taking history as a mirror, one can understand the rise and fall of dynasties.

Information technology has developed at an astonishing pace, and with each technological revolution, dazzling unicorn enterprises have emerged. In today's era of generative AI, OpenAI stands out as the brightest star globally, leading the trend of large model development and serving as a benchmark for this field. Its valuation has reached a staggering $157 billion, making it the world's most valuable AI large model startup, second only to ByteDance and SpaceX.

Driven by OpenAI, numerous large model enterprises have emerged in China, with over 200 large models registered. Among them, six startups – Baichuan AI, ZeroOne AI, Zhipu AI, MiniMax, Dark Side of the Moon, and Leapfrog AI – are known as the "Six Little Tigers" of large models. These startups are shining unicorns in China's large model scene, with valuations in the billions of dollars.

The title of "Six Little Tigers" reminds me of the former "Four Little Dragons of AI": SenseTime, Megvii, CloudWalk, and Yitu Technology. Unfortunately, they are no longer as prominent as they once were.

Back in the day, they shone brightly as the Four Little Dragons of AI, attracting immense attention and rounds of funding. However, to this day, they still struggle to achieve profitability, with sustained losses remaining one of the biggest points of contention. This highlights the challenges of commercializing AI technologies.

Although SenseTime and CloudWalk have successfully entered the capital market, they are no longer the darlings of investors as they once were. SenseTime, once the leader of the Four Little Dragons, debuted with record-breaking stock prices, peaking at HK$9.7. However, its share price has since plummeted to HK$1.62, and its market value has shrunk from a peak of over HK$320 billion to just HK$53.7 billion, representing an 80% decrease.

CloudWalk's market value, once as high as RMB 46 billion during the generative AI wave, has now shrunk to RMB 11.5 billion, a decrease of over RMB 30 billion or 70%. Megvii, on the other hand, has been relatively silent, with no substantial progress in its IPO process and remaining in the registration stage.

Overall, the glory days of the Four Little Dragons of AI are over. The so-called "Six Little Tigers" of large models may face a similar predicament in the coming years.

After all, large model development is a technology- and capital-intensive industry. It requires not only AI talent but also robust AI computing infrastructure. Consequently, significant investments are necessary, particularly in purchasing high-performance AI chips from companies like NVIDIA. Given the long-term financial commitment required, the most pressing priority for the "Six Little Tigers" is to commercialize their technologies and generate revenue to enhance their self-sustaining capabilities.

It is noteworthy that companies selling large models, including OpenAI, have yet to turn a profit from them. In contrast, NVIDIA, often referred to as the "shovel seller," has profited immensely from selling the underlying infrastructure necessary for large model development.

In fiscal year 2024, NVIDIA's net income reached a staggering $29.76 billion. In the most recent quarter (ending July 28, 2024), its net income was $16.6 billion, a year-over-year increase of 168%. In other words, selling computing power is currently the only AI business that can generate significant profits.

Given that the development of AI large models by both tech giants and startups relies heavily on NVIDIA's high-performance chips, the competition among AI large model providers ultimately boils down to who has access to the most NVIDIA cards. For deep-pocketed tech giants like Google and Meta, investing billions of dollars in purchasing and stockpiling NVIDIA chips is a strategic move to enhance their AI capabilities.

As tech giants engage in a buying spree, NVIDIA's chips are in high demand, leading to record-breaking sales and profits. Consequently, NVIDIA's stock price has soared, and its market capitalization has reached $3.39 trillion (approximately RMB 24 trillion), making it the second-largest company globally by market value, second only to Apple. In terms of both revenue and market capitalization, NVIDIA is the largest chip company in the world.

Large model startups, lacking the financial resources of tech giants, struggle to achieve revenue balance in the AI large model market. Nonetheless, they must still invest heavily to support their development.

Training large-scale AI models with trillions of parameters can cost millions or even tens of millions of dollars. For large model startups, continuous fundraising is crucial to enhance their AI capabilities. SenseTime, for example, boasts that its large model capabilities rival those of GPT-4 Turbo, underpinned by significant investments in AI infrastructure.

According to disclosed information, SenseTime's large model infrastructure, SenseCore, has achieved a breakthrough in total computing power, reaching 20,000 petaFLOPS, powered by 54,000 GPUs. These substantial investments in AI infrastructure and R&D have contributed to SenseTime's continued losses, amounting to RMB 2.477 billion in the first half of 2024.

AI infrastructure lies at the heart of large model capabilities. Billions of dollars are invested to purchase tens of thousands of GPUs to build AI infrastructure for large models or to procure computing services from cloud providers like Alibaba Cloud, Tencent Cloud, and Huawei Cloud.

Whether building AI infrastructure in-house or procuring computing services from cloud providers, the costs are substantial and require a steady stream of funding to sustain. This funding determines how far a large model startup can go.

Among them, Baichuan AI, founded by Wang Xiaochuan, has raised RMB 5 billion in funding, valuing the company at RMB 20 billion. Zhipu AI has also secured billions of yuan in funding, valuing it at up to RMB 20 billion. LeadOne AI, an AI large model unicorn founded by Kai-Fu Lee, has raised hundreds of millions of dollars in funding, valuing the company at over USD 1 billion, although the exact amount has not been disclosed.

It is evident that with funding support, large model startups have thrived in the wave of large model development, attracting capital and achieving impressive valuations.

However, the current large model market has become highly competitive, with prices plummeting to rock-bottom levels. In 2024, a wave of price reductions hit the market, with large companies aggressively competing for market share regardless of costs. Additionally, open-source large models like Llama pose further competitive pressures. As a result, monetizing large model startups has become increasingly challenging, leading to a never-ending cycle of fundraising with limited profitability prospects.

Overall, while AI large models are enjoying significant buzz and attracting round after round of funding, this speculative behavior has contributed to significant bubble formation. The current situation of the "Six Little Tigers of AI Large Models" may mirror that of the former "Four Little Dragons of AI," reminding us that history serves as a valuable mirror for understanding trends and avoiding past mistakes.

Finally, the generative AI wave has swept across the globe, with various sectors embracing this transformative technology. The potential market demand is gradually being unleashed. IDC previously predicted that global enterprises' investment in generative AI solutions will reach $143 billion by 2027.

The generative AI market holds immense potential. However, in the race to develop AI large models, the reality is often harsher than the ideal. Core to success lies in the practical application and commercialization of these models. Only by solidifying and steadily promoting their development and commercialization can the "Six Little Tigers of AI Large Models" find their way forward.

Yang Jianyong is a Forbes China contributor whose opinions are solely his own. He focuses on in-depth analysis of cutting-edge technologies such as the Internet of Things, cloud computing, artificial intelligence, and smart homes.