The global artificial intelligence market is reaching a critical inflection point where raw computational power is being superseded by strategic efficiency. According to a recent report by Citigroup (Citi), the emergence of DeepSeek V4 and the impressive trajectory of MiniMax (00100.HK) highlight a fundamental shift: success in AI is no longer measured solely by parameter count, but by the cost per token. Recognizing this trend, Citi has set a price target for MiniMax shares at HKD 1,330, underscoring its confidence in Chinese firms that deliver top-tier performance at a fraction of the cost of their Western counterparts.
The DeepSeek V4 Revolution and the MoE Model
DeepSeek V4 is not just another large language model (LLM). It represents the pinnacle of Mixture-of-Experts (MoE) architecture, which allows the model to activate only the necessary parts of its neural network for any given task. This leads to a drastic reduction in energy consumption and computational resources during both training and, more importantly, inference. Citi points out that DeepSeek has successfully democratized access to high-level AI, forcing American giants like OpenAI and Google to re-evaluate their pricing structures.
The bank's analysis emphasizes that DeepSeek V4 shows improved capabilities in coding and mathematics—areas considered benchmarks for a model's true intelligence. The company's ability to produce these results despite export restrictions on advanced chips (such as NVIDIA’s H100) to China serves as a loud message regarding the resilience and innovation of the Chinese tech ecosystem.
MiniMax: The New Unicorn on the Hong Kong Exchange
While DeepSeek captures attention in the open-source and efficiency sectors, MiniMax is establishing itself as the preferred choice for enterprise applications. Citi’s valuation of HKD 1,330 per share reflects MiniMax’s strong penetration into the consumer application and corporate solution markets. MiniMax has managed to build a "sticky" user base through its digital assistants, which are noted for their high emotional intelligence and ability to maintain context over long dialogues.
Citi estimates that MiniMax will benefit immensely from the growing demand for AI applications in Asia, as local businesses seek solutions that are culturally and linguistically tailored while avoiding over-reliance on U.S. cloud infrastructure. MiniMax’s strategy to focus on User Experience (UX) combined with the technical excellence of its models makes it one of the strongest players in the industry.
Geopolitics and the Economics of Intelligence
Citi’s report goes beyond financial metrics, touching upon the broader geopolitical landscape. The rise of DeepSeek and MiniMax comes during a period of intense U.S.-China competition for AI supremacy. The shift toward cost-efficiency is, in many ways, a response to resource constraints. When you cannot buy more chips, you are forced to write better code.
This "forced" innovation appears to be paying off. Citi notes that the training costs for Chinese models are now significantly lower than those of Silicon Valley models, which could lead to a massive shift of investment capital toward the East, provided the Return on Investment (ROI) remains competitive. The conclusion is clear: the era of profligacy in AI is ending, and the era of smart resource management is beginning.
