By Yang Jianyong

Large models, currently one of the hottest tech sectors, are attracting significant capital interest. Among them, OpenAI is pursuing a new round of financing at a valuation of $1.4 trillion, while Anthropic’s valuation soars even higher to $2 trillion, with its anticipated IPO financing possibly reaching $100 billion, setting a new global record for IPO fundraising.

Zhipu and MiniMax Experience Sharp Decline in Market Value, Marking End of Capital Extravaganza in Large Models

OpenAI and Anthropic are globally recognized as leading large model enterprises, serving as benchmarks for others in the sector, especially in markets where such benchmarks carry significant weight. Domestic companies like Zhipu, MiniMax, DeepSeek, and Moonshot AI have garnered considerable attention.

Among these, Zhipu and MiniMax stand out as rare large model prospects in the Hong Kong stock market. They previously fueled market speculation, with Zhipu’s peak price reaching HK$2,980, a 24-fold increase from its issue price, and its peak market value hitting HK$1.33 trillion. MiniMax’s market value also surged, peaking at HK$417.6 billion, a seven-fold increase.

Unfortunately, following the initial hype, Zhipu’s market value has diminished by over HK$1 trillion, while MiniMax’s market value now stands at only HK$74 billion, having shrunk by over HK$340 billion, marking an astonishing 82% decline.

This decline can be attributed to pressure from share lock-up expirations and a market reassessment of large model companies’ valuations, transitioning from a phase of exuberance to one of deflating the bubble, with a renewed focus on the practical implementation and profitability prospects of large models.

Crucially, large models represent capital-intensive industries that rely heavily on extensive AI infrastructure. Consequently, substantial computing resources are invested in model training to enhance competitiveness.

Three-Year, $1.2 Billion Cloud Services: Persistent High Costs for Large Models

As large models continue to evolve, they have transformed from initial knowledge Q&A systems to Agent AI entities capable of executing tasks, further driving up Token usage and amplifying computing costs for large model companies.

OpenAI’s computing investment is projected to reach $600 billion by 2030, while Anthropic’s total computing expenditure, including $100 billion for cloud services from Amazon AWS and Google, totals $518 billion.

MiniMax’s research and development (R&D) and computing costs have also escalated continuously. In the first half of 2026, R&D expenditure increased by 138.8% year-on-year to $297 million, primarily due to ongoing development and refinement of foundational models and multimodal capabilities, leading to a significant rise in training-related cloud services.

Notably, to meet the computing demands of model training, MiniMax recently revised its three-year, $1.2 billion computing service agreement with Alibaba Cloud, raising the procurement caps for 2026, 2027, and 2028 from $115 million, $125 million, and $135 million to $300 million, $400 million, and $500 million, respectively, indicating exponential growth in computing costs.

Substantial computing resources are invested in model training, driving up computing costs continuously. However, selling large models makes it challenging to achieve revenue balance, leading to widening losses. In the first half of 2026, MINIMAX incurred a loss of $358 million, with an adjusted net loss of $293 million, a 111% increase year-on-year, surpassing its full-year loss in 2025.

Revenue Surges 283%, Large Model Commercialization Accelerates Towards B-end

Fortunately, in the era of generative AI, there is robust demand for large models across various sectors, driving rapid revenue growth for large model companies. In the first half of 2026, MINIMAX’s revenue reached $116 million, a 283% increase year-on-year, with its models and product services reaching over 300 million users across more than 230 countries and regions. Over 60% of its revenue comes from international markets.

Notably, as Agentic AI gradually evolves from assisting with individual tasks to undertaking complex, multi-step tasks, intelligent agents are reshaping workflows, with applications in the Agent office sector exceeding expectations. MINIMAX’s business model has also shifted accordingly. Previously reliant on C-end products, it now derives its core revenue from the B-end.

In the first half of 2026, revenue from open platforms and other AI-based enterprise services reached $73.92 million, a 703% increase year-on-year. The B-end revenue share rose from 30% in the same period last year to 63.4%, underscoring the immense potential for commercializing AI large model technologies.

In Conclusion

The large model wave is sweeping across various sectors, with active adoption of the intelligent transformation brought by generative AI, gradually unlocking market demand potential. However, despite the current hype surrounding AI large models, R&D and computing costs continue to escalate, making it difficult for large model companies to achieve profitability in the short term. They still require substantial funding to support their development.

Anthropic expects to raise $100 billion through its IPO, while OpenAI is in talks for a new $30 billion financing round. Zhipu and MiniMax have opted to raise funds through market placements and bond issuances.

Zhipu has raised a total of HK$75.6 billion in the market this year to develop models and meet demand by investing in more computing resources. MiniMax has raised over HK$20 billion this year for AI infrastructure investment, model R&D, and global commercial expansion.

For large model unicorns, the most urgent task is to implement technology and monetize it to enhance their self-sustaining capabilities. In this emerging sector, sustained capital favor ultimately depends on product commercialization capabilities. The long-term development potential of AI large models should be viewed with rationality.

The views expressed herein represent personal opinions only. The author focuses on in-depth observations of technology sectors such as AI large models, artificial intelligence, industrial robots, IoT, cloud computing, and smart hardware.