UBS is bullish on MiniMax's multimodal capabilities and broader commercialization scenarios

Overview

This report focuses on the potential undervaluation of MiniMax Group in AI foundation models, multimodal capabilities, and commercial applications. UBS believes the company will benefit from growing industry token demand, diversified revenue sources, and overseas expansion potential, while also having opportunities to further scale its model capabilities across text, video, and audio models, consumer AI applications, and open platforms for enterprises and developers.

Core views

The core view is that the market may be overly focused on already-validated scenarios such as AI coding, while underestimating the improvement in model intelligence enabled by MiniMax's cross-modal synergy between text and video models, training and inference efficiency, and next-generation M3/M4 model parameter scaling. UBS also believes the company can convert stronger model capabilities into broader productivity scenarios such as creativity and workspace applications.

Analysis framework

The report analyzes MiniMax's investment value from the perspectives of model capabilities, computing power supply, token efficiency, inference gross margin, the harness engineering layer, peer model parameter and pricing comparisons, catalysts, and valuation risks, and uses Price/ARR as the valuation method.

Methodology notes

• Valuation methodPrice/ARR

  Measure company valuation based on price relative to annual recurring revenue

  UBS uses Price/ARR as the valuation method for MiniMax, which is suitable for an AI foundation model company still in a stage of high growth and commercialization validation.

• AI Agent frameworkModel + Harness

  Agent = Model + Harness

  The report defines harness as the system and engineering layer outside the model, including task orchestration, tool integration, execution environment, memory/context, and feedback loops, used to enable the model to perform multi-step tasks in real-world applications.

• Short-term quantitative viewQuantitative Research Review

  Analysts' quantitative assessment of short-term factors

  The report includes UBS's quantitative research assessment, showing that most short-term factors such as industry structure, recent stock performance, and EPS revision risk are skewed positive, although this assessment has a different time horizon from the 12-month rating.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• MiniMax Group (0100.HK)

  Core covered target in the report

  Strengths

  Has positioning in AI foundation models, multimodal generation, consumer applications, and open platforms; TPM growth is solid, M2.7 inference gross margin exceeds 40%, and harness engineering capabilities help expand Agent applications.

  Weaknesses

  The company remains in a loss-making and commercialization expansion stage, with EPS still negative in 2026E-2028E and margins under pressure.

  Comparison

  Compared with some domestic AI model peers, MiniMax has smaller model parameters but broadly comparable intelligence performance, while offering advantages in input/output pricing and some aspects of inference efficiency.

  Risks

  Macroeconomic and geopolitical uncertainty, slower AI adoption, intensifying competition, operational execution, data training, governance of user-generated content, disruption from key service providers, and share price volatility caused by lock-up expirations.

Key data

• 12-month ratingBuyThe report lists MiniMax Group's 12-month rating as Buy.
• Target priceHK$1,000.0012-month target price, corresponding to the price of HK$777.50 on April 24, 2026.
• Current priceHK$777.50The price timestamp is April 24, 2026, 18:24:04 HKT or based on the local market close.
• Projected share price upside28.6%Both the projected share price upside and expected stock return are 28.6%.
• Market capitalizationHK$244b/US$31.1bListed in the report's trading data and key indicators section.
• 52-week rangeHK$1,238.00-345.00The original text lists it as 52-wk range HKS1,238.00-345.00.
• TPM growth10%-20% WoWUBS believes MiniMax's computing power expansion is solid, with tokens per minute maintaining 10%-20% week-over-week growth.
• M2.7 gross margin>40%The report states that the latest text model M2.7 has a gross margin above 40%, comparable to global leaders and higher than domestic peers.
• MiniMax M2.7 pricingInput US$0.30/million tokens; output US$1.20/million tokensThe chart shows that MiniMax has a cost advantage in input/output pricing relative to some domestic peers.
• 2026E EPSUS$(1.60)UBS's diluted EPS forecast is shown as (1.60) for 12/26E in the table.

Impact & implications

If MiniMax can continue to improve its multimodal model capabilities, reduce inference costs, and expand AI Agent and workspace application deployment through harness, the market may reassess its commercialization potential and long-term growth prospects. Potential index inclusion and Southbound Stock Connect inclusion may also provide capital-market catalysts, but lock-up expirations and industry competition could bring share price volatility.

Risks

• Macroeconomic and geopolitical uncertainty may constrain AI adoption.
• Competition and operational challenges may delay commercialization and pressure margins.
• Data-related model training risks may affect model quality or compliance.
• User-generated content governance risks may increase regulatory and platform operating pressure.
• Disruptions from key service providers may affect business continuity.
• Lock-up periods for IPO cornerstone investors and some Pre-IPO investors expire in July and October, which may trigger share price volatility.

What to watch

• Potential important model upgrades such as M3/M4 that may emerge in May-June.
• Potential inclusion in the HSCI and HSTECH in June.
• Potential Southbound Stock Connect inclusion in August.
• Progress of Hailuo 3 in video understanding capabilities and adoption by ordinary users.
• The deployment effectiveness of MaxHermes and harness engineering in workspaces, tool use, and multi-step tasks.
• Changes in token demand, TPM growth, inference gross margin, and input/output pricing.