Citi Sets MiniMax HKD 1,330 Target on AI Cost-Efficiency Trend

Citi has initiated coverage on MiniMax (00100.HK) with a "Buy/High Risk" rating, setting a HKD 1,330 price target that points to a broader shift in the artificial intelligence landscape. The bank's thesis suggests that while frontier models from OpenAI and Anthropic lead in performance, a new wave of cost-efficient models, exemplified by competitors like DeepSeek, is rapidly expanding the total addressable market for AI.

"Closed frontier models maintain a notable competitive advantage in long-term workflows, where reliable performance remains critical," Citi's research report said. However, the bank noted that for coding, agent workflows, and long-context applications, "the performance gap of open-weight models has narrowed, mainly benefiting from cost-competitive model architectures."

The valuation for MiniMax is based on a projected 30x price-to-sales ratio for 2028, underpinned by an aggressive compound annual growth rate forecast of 184% from 2025 to 2028. This compares to a 128% CAGR projected for the 2025 to 2030 period. The valuation reflects the potential for companies that can lower AI deployment costs to capture a significant share of the market.

The move highlights a growing polarization in the AI sector. While models like OpenAI's GPT-5.5 and Anthropic's Opus 4.7 push the boundaries of capability, companies like DeepSeek are focusing on efficiency. As competition among these cost-focused models intensifies, Citi expects the broader ecosystem to benefit from accelerated enterprise adoption and a larger overall market.

The Efficiency Play

DeepSeek's recently released V4 model is a case in point. The model performed on par with Claude's Sonnet 4.6 in some benchmarks but at a significantly lower price than GPT-5.5, according to Citi's report. This is achieved not by sacrificing features, but through architectural innovation.

According to its technical report, DeepSeek V4 uses a hybrid attention design combining Compressed Sparse Attention (CSA) and Heavily Compressed Attention (HCA). This allows the model to handle million-token contexts more practically by compressing parts of its memory, directly addressing one of the main cost drivers in large-scale AI. This focus on engineering solutions for cost makes large-context workflows more realistic for a wider range of developers and enterprises.

The strategy appears to be gaining traction. Reuters reported that DeepSeek V4 has been adapted to run on Huawei’s Ascend chips, signaling a move toward full-stack co-design where models and hardware evolve together to optimize for efficiency over raw power.

Valuation Hinges on 184% Growth

Citi's "High Risk" qualifier on its rating acknowledges the significant uncertainties in the rapidly evolving AI sector and MiniMax's own limited trading history. Achieving a 184% revenue CAGR through 2028 is a monumental task that depends on flawless execution and the market's continued appetite for cost-effective AI solutions.

The investment thesis is a direct bet on market expansion. By lowering the barriers to entry for experimentation and deployment, models from DeepSeek, MiniMax, Kimi, and GLM are enabling use cases that were previously too expensive. This could unlock new revenue streams and justify the high-growth multiples, but investors will be watching closely to see if the reasoning gap with frontier models can be closed without sacrificing the cost advantage.

This article is for informational purposes only and does not constitute investment advice.