By Yang Jianyong

Since the global boom in generative AI, AI large-scale models and AI computing chips have unquestionably become the most sought-after sectors. The demand for AI chips continues to surge, driven by model training and compounded by factors such as the push for domestic alternatives. This has propelled Cambricon’s performance to multiply rapidly, with its market value repeatedly hitting new highs—at one point surpassing the one trillion yuan mark—making it the standout winner in China’s AI computing chip sector.

From 1.73 Trillion to 810 Billion: A Post-Hype Reality Check

In the large-scale model arena, MiniMax and Zhipu stand out as leading domestic manufacturers and the only large-scale model listings on the Hong Kong stock exchange. At their peak, their combined market value exceeded 1.73 trillion Hong Kong dollars. However, their current combined market value has plummeted to just 810 billion, a staggering reduction of 920 billion.

Currently, Zhipu’s market value stands at 731.2 billion Hong Kong dollars, experiencing a correction of over 45% from its peak of 1.32 trillion, a decline of 588.8 billion.

MiniMax, after a prolonged period of volatility (a volatile decline), now has a market value of only 84.2 billion Hong Kong dollars, a sharp drop of 330 billion from its peak of 417.6 billion, representing an 80% decline. This is primarily due to pressure from regulatory changes, signaling a shift from euphoria to a phase of deflating hype and prompting a reassessment of implementation and profit expectations.

Escalating Losses: Large Models as Resource-Intensive Behemoths

While AI large-scale models are flourishing and the market remains optimistic about their commercialization potential, generating profits from large-scale model services remains challenging—even for OpenAI, which is also grappling with financial strain. As a global benchmark for large-scale models, OpenAI’s data reveals that its 2025 revenue is approximately 13 billion USD, with losses soaring to as high as 38.5 billion USD.

The primary reason for these substantial losses lies in the high dependency of large-scale models on AI infrastructure, necessitating vast computational resources for model training. This is one of the key factors making it difficult for AI large-scale models to achieve profitability in the near term.

In 2025, OpenAI’s total expenditures reached 34 billion USD, with 19 billion USD allocated to R&D. The bulk of this investment is directed towards large-scale model development and AI infrastructure construction, causing costs to skyrocket. It is projected that computing investments will reach 600 billion USD by 2030. It can be said that from R&D to AI computing, AI large-scale model unicorns are veritable resource hogs.

The two leading domestic companies are also mired in significant losses, with the scale of losses continuing to widen. Financial reports indicate that MINIMAX incurred losses exceeding 12.8 billion yuan last year, with adjusted losses also reaching as high as 1.7 billion.

Zhipu’s losses in 2025 amounted to 4.718 billion yuan, compared to 2.958 billion yuan in the same period the previous year, representing a 59.5% increase in losses. Among these, R&D expenditures were 3.18 billion yuan, a 44.9% year-on-year increase. Seventy percent of this was allocated to computing resources, reflecting increased investments to meet demand and resulting in expanded losses last year. This underscores that large-scale models belong to technology and capital-intensive industries.

Continuously rising computing costs, along with multiple factors such as industry competition, have directly led to a decline in gross margins. Zhipu’s gross margin decreased from 56.3% in 2024 to 41% in 2025. With the rapid development of Agentic AI and the subsequent explosion in Token consumption, computing resources are undergoing significant changes. Zhipu claims that its focus is not on profitability but on supporting the ever-increasing exponential curve of high-quality Token consumption.

The Two AI Giants Demonstrate Commercialization Potential

However, it is commendable that, despite massive R&D investments, AI large-scale model technologies are rapidly evolving, and the commercialization of large-scale models is showing growth, leading to multiplied revenue increases.

Zhipu’s revenue grew from 57.4 million yuan in 2022 to 724 million yuan in 2025, expanding more than 12-fold in three years. MiniMax’s revenue jumped from 3.46 million USD in 2023 to 79 million USD in 2025, expanding 22-fold in two years, confirming the enormous potential for commercializing AI large-scale model technologies.

Overall, in the era of generative AI, various sectors are seizing the opportunity to accelerate intelligent upgrades. With the rapid development of AI large-scale models, both consumer (C-end) users and business (B-end) enterprises are demonstrating significant commercial potential, not only reshaping information acquisition methods but also injecting vitality into innovation across various industries.

Against this trend, the commercialization of large-scale models is rapidly advancing, with AI becoming ubiquitous from the cloud to the edge. However, it remains crucial to maintain a rational perspective on the long-term development potential of AI large-scale models.

For large-scale model companies, it is essential to promote the steady development and commercialization of large-scale models, thereby positively impacting performance. There is also anticipation for achieving profitability in the future, realizing a transition from scale growth to profit growth.

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