Citi Maintains MiniMax at Buy/High Risk; Hailuo 3 and the 3tn-parameter model are key catalysts
AI summary card
Citi Maintains MiniMax at Buy/High Risk; Hailuo 3 and the 3tn-parameter model are key catalysts
Citi believes MiniMax has cost and capability advantages thanks to its self-built GPU cluster, optimized training pipeline, and multimodal strategy, but share price recovery still depends on whether new models can demonstrate stronger capabilities.
- MiniMax management emphasized “ultimate intelligence, ultimate cost efficiency,” aiming to deliver higher performance per dollar amid explosive token growth.
- The company has built its own data centers and GPU clusters; management said cluster utilization exceeds 90%, creating a structural cost advantage versus the cloud-rental model.
- The next-generation model of around 3tn parameters is expected to focus on programming and agentic tasks, with improved training outcomes through KV cache compression, expert data construction, and reinforcement learning.
- Citi believes Hailuo 3 is about to be launched, and if performance exceeds expectations, it could ease negative market sentiment regarding model capabilities and the share price.
- The report maintains a target price of HK$533, based on 12x 2028E P/S, implying 167.4% expected share price return.
Report interpretation
Overview
This report summarizes Citi’s main views after attending MiniMax’s senior management meeting in Shanghai. Management believes the LLM industry is entering a phase of explosive growth and intense competition, with computing power still scarce, while application demand—especially in programming, autonomous agents, and multimodal content generation—is set to grow rapidly. Citi maintains its Buy/High Risk rating on MiniMax, mainly based on the company’s potential advantages in computing infrastructure, training optimization, multimodal capabilities, and commercialization pathways.
Core views
Citi’s core view is that MiniMax has cost and full-stack integration advantages from its self-built computing infrastructure, and management has a clearer optimization path for next-generation model training, improving the outlook for the next base model iteration. However, market sentiment may remain cautious until the company proves the capabilities of its new models; the launch of Hailuo 3 and the future roughly 3tn-parameter model are key near-term catalysts.
Analysis framework
The report mainly uses management meeting notes, industry competition assessment, technology roadmap analysis, financial forecasts, and a relative valuation framework. For valuation, Citi uses 2028E revenue as the basis and applies a 12x two-year forward P/S to derive the HK$533 target price.
Methodology notes
Values MiniMax based on 2028E revenue and 12x P/S.
Because MiniMax lacks a long history of listed comparable companies and the AI model industry remains in an early stage, the report uses the company’s own relatively short trading history as a reference and uses two-year forward P/S to reflect 128% revenue CAGR for 2025-2030E and 184% revenue CAGR for 2025-2028E.
Assesses the company’s competitiveness through management’s statements on the industry, computing power, training, multimodality, and commercialization.
The report focuses on MiniMax management’s views on computing scarcity, cost efficiency, training processes, programming and agent tasks, multimodal models, and the launch timeline of Hailuo 3.
Focuses on near-term model launches and share price sentiment recovery.
The report notes that the previous Downside 30-Day Short-Term View has been closed, because the share price had already fallen sharply after IPO lock-up pressure, and if feedback on Hailuo 3 and the future 3tn-parameter model exceeds expectations, it may offset negative sentiment.
Asset mapping & comparison
Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).
- MiniMax Group Inc / 0100.HKCore covered name
- Strengths
- Self-built GPU cluster, high computing utilization, clear training optimization path, clear multimodal strategy, parallel ToC and ToB commercialization.
- Weaknesses
- Short listed trading history, still in a loss-making phase, and investors have doubts about next-generation model capabilities.
- Comparison
- The report states that among independent Chinese AI companies, MiniMax has a relatively large GPU cluster; compared with competitors relying on cloud rental, self-built data centers may create a cost advantage.
- Risks
- Rapid model iteration, intensifying competition, subsequent post-lock-up pressure, ongoing capital needs, FX volatility, and risks related to R&D talent retention and intellectual property.
Key data
- RatingBuy/High RiskCiti maintains its buy rating and high-risk rating on MiniMax shares.
- Target PriceHK$533.00Based on 12x 2028E P/S valuation.
- Current PriceHK$199.30Price as of 16:10 on July 23, 2026.
- Expected Share Price Return167.4%The report shows both expected share price return and expected total return at 167.4%.
- Market CapitalizationHK$69,603M / US$8,877MMarket capitalization as shown on the front page of the report.
- 2026E RevenueUS$232mIncome statement forecast.
- 2027E RevenueUS$726mIncome statement forecast.
- 2028E Revenue Growth149.2%The report shows the 2028E sales revenue growth rate.
- 2027E Adjusted EBITUS$-571mThe company remains in a phase of expanding losses.
- User and Customer ScaleOver 236m users, 214,000 enterprise customers and developersAs of December 31, 2025, covering more than 200 countries and regions, with enterprise customers and developers from more than 100 countries and regions.
Impact & implications
The report’s investment implication for MiniMax is positive: if the company can lower inference and training costs through its self-built computing power and demonstrate stronger programming, agent, and multimodal capabilities in Hailuo 3 and the next-generation 3tn-parameter model, the market may reassess its growth prospects and valuation upside. However, due to intense industry competition, high funding needs, and rapid model iteration, the stock’s high-risk profile remains prominent.
Risks
- Rapid model evolution and intense industry competition may weaken MiniMax’s technological leadership.
- Selling pressure from the expiration of the six-month post-IPO lock-up may continue to affect the share price.
- AI model training and inference require sustained capital investment, and the company may face long-term financing needs.
- The company reports in U.S. dollars and faces FX volatility risk.
- R&D talent retention and intellectual property infringement risks may affect long-term competitiveness.
- If Hailuo 3 or the next-generation 3tn-parameter model falls short of expectations, investor sentiment may remain under pressure.
What to watch
- User feedback, product traction, and multimodal generation quality after the launch of Hailuo 3.
- Capability performance of the roughly 3tn-parameter model expected to be launched in the coming months in programming and agentic tasks.
- Whether MiniMax’s self-built GPU cluster utilization, chip procurement pace, and computing cost advantages remain sustainable.
- ARR growth, ToC and ToB commercialization progress, and opportunities in software-plus-hardware robotics AI scenarios.
- Whether lock-up pressure continues to be absorbed after the stock’s earlier decline of more than 40%.