Text/Yang Jianyong

Information technology is progressing at a breathtaking speed, and with each technological revolution, a host of dazzling tech companies emerge.

In the age of generative AI, OpenAI and Anthropic undoubtedly stand as the two most valuable AI unicorns globally, boasting valuations of $850 billion (approximately RMB 5.9 trillion, equivalent to the scale of three Moutai companies) and $380 billion (approximately RMB 2.6 trillion), respectively.

Spurred on by OpenAI, domestic large model services have surged forward like a tidal wave, with 748 generative AI services completing their registration. Among the plethora of large model services, Doubao, DeepSeek, Qianwen, Yuanbao, Wenxin, and others have risen to prominence, profoundly reshaping how people access information and posing a threat to replace traditional search engines as the new gateway for users to obtain information.

Meanwhile, six large model startups—Zhipu AI, MiniMax, Baichuan Intelligence, 01.AI, Moonshot AI, and Stepfun—have garnered significant attention and are hailed as the 'Six Little Tigers' of large models, drawing immense interest from the market.

Among them, Moonshot AI secured two rounds of financing in just two months, with financing amounts and valuations that are truly remarkable. The financing amounts were $500 million and $700 million, respectively, totaling $1.2 billion across the two rounds, propelling its valuation to around $10 billion.

As the first global AI large model stock, Zhipu has become a market focal point, highly favored by investors. On the first trading day of the Year of the Horse, its stock price surged by 42.7%.

Since its listing in January of this year, Zhipu's cumulative increase has reached an astonishing 524%, with its market value soaring five-fold to HK$323.2 billion. This demonstrates investors' optimism about the commercial prospects of AI in the face of the opportunities brought by large models.

Of course, despite being a hot market target, the risks behind Zhipu's soaring market value cannot be overlooked. Its market value of HK$320 billion far exceeds the support range provided by its revenue scale and profitability.

In the first half of 2025, Zhipu's revenue was only RMB 191 million, while its losses amounted to RMB 2.35 billion. From 2022 to the first half of 2025, its cumulative losses totaled RMB 6.23 billion. It's important to note that companies selling large models have not yet turned a profit from them. Therefore, it is particularly crucial to view the long-term development potential of AI large models rationally.

Most crucially, developing AI large models requires substantial financial support to build powerful AI computing infrastructure. Training AI large models with hundreds of billions or even trillions of parameters costs at least millions of dollars, if not tens of millions, with research and development and AI computing costs remaining high.

Data disclosed by Zhipu shows that 70% of its R&D investment is spent on purchasing computing services, highlighting that large models are a capital-intensive industry. To this end, it raised HK$4.348 billion at an issue price of HK$116.2 to fund R&D in general-purpose AI large models, further consolidating Zhipu's competitiveness in general-purpose foundation models.

Notably, as AI large model technology continues to iterate, commercialization is accelerating, and revenue for large model vendors is growing rapidly.

Zhipu's revenue surged from RMB 57.4 million in 2022 to RMB 312 million in 2024, achieving a five-fold expansion in just two years. Growth momentum remained strong in 2025, with first-half revenue reaching RMB 191 million, up 325% year-on-year. This confirms the immense potential of commercializing AI large model technology.

In its commercialization path, as a general-purpose large model company, Zhipu launched Model-as-a-Service (MaaS) and a commercialization platform, providing general-purpose large model services to institutional clients and individual users. As large models are applied, revenue has grown exponentially.

It's worth mentioning that Zhipu AI launched a new generation flagship foundation model, GLM-5, which has achieved significant leaps in agent and programming capabilities. Notably, its programming capabilities have aligned with Claude Opus 4.5, surpassing Gemini 3 Pro in performance.

Meanwhile, GLM-5 has completed deep inference adaptation with domestic computing platforms such as Huawei Ascend, Moore Threads, Cambrian, MetaX, Suiyuan, and Hygon.

It should be noted that due to sustained strong demand for the GLM Coding Plan in the market, with rapid growth in user scale and usage volume, Zhipu announced a structural adjustment to the pricing system of the GLM Coding Plan package, eliminating first-purchase discounts and raising overall package prices by at least 30%.

In an increasingly competitive environment for large models, Zhipu took the lead in significantly raising prices, sending a positive signal for healthy competition among other large model services. This move breaks the vicious cycle of low-price competition and instead enhances competitiveness through differentiated services and technological upgrades.

Finally, in the era of generative AI, AI large model services are injecting innovative vitality into various industries, which are actively embracing the transformative capabilities brought by large models, sparking a new wave of industrial intelligence.

Against this trend, the commercialization of large models is accelerating, leading to strong market favor for large model companies. Once again, it is emphasized that despite the current hype around AI large models, fueled by capital, companies with only a few hundred million in revenue and massive losses are propping up market values in the hundreds of billions, which is bound to contribute to the formation of an AI large model bubble.

Only solid implementation can navigate through cycles, and long-term value must ultimately be achieved through 'technological differentiation + commercial sustainability.'

Yang Jianyong, a Forbes China contributor, expresses views that represent his personal opinions only. He is dedicated to providing in-depth interpretations of cutting-edge technologies such as AI large models, artificial intelligence, the Internet of Things, cloud computing, and smart hardware.