UBS believes the market is underestimating MiniMax's multimodal and engineering-driven monetization potential

Overview

This is a company research report by UBS on MiniMax Group (0100.HK). The report argues that although MiniMax's share price has risen significantly since listing, the market may still be underestimating its long-term potential in multimodal foundation models, compute scaling, inference efficiency, and the engineering application layer. UBS's focus is not a single AI coding scenario, but rather the monetization potential of improving model intelligence across broader scenarios such as creativity, video generation, office productivity, and enterprise developer platforms.

Core views

The report's core views include: first, MiniMax has differentiated advantages in compute scaling and inference efficiency, with TPM maintaining 10%-20% week-over-week growth and the M2.7 text model achieving a gross margin above 40%; second, MiniMax's models deliver intelligence close to major domestic AI lab models despite smaller parameter scale and more efficient activated parameters, while showing cost advantages in input/output pricing, latency, and output speed; third, the harness engineering layer can systematically improve AI Agent performance, and products such as MaxHermes help translate model capabilities into usable workflows; fourth, follow-on model upgrades such as Hailuo 3 and M3 may strengthen video understanding, agentic capabilities, and penetration into office scenarios; fifth, index inclusion and Stock Connect inclusion may serve as share price catalysts, though IPO cornerstone investors and some pre-IPO investors coming off lock-up could bring volatility.

Analysis framework

The report uses an analytical framework combining company fundamentals, AI model technology roadmap, inference economics, product deployment scenarios, and capital-market catalysts. The valuation method uses Price/ARR, along with 12-month forecast return, earnings forecasts, industry structure changes, the regulatory environment, and a short-term quantitative research questionnaire to assess investment attractiveness.

Methodology notes

• Valuation methodPrice/ARR

  Measures the valuation level of AI foundation model companies by price relative to annual recurring revenue.

  UBS explicitly uses Price/ARR as the valuation method for MiniMax, which is suitable for AI companies that are still loss-making but whose main pricing variables are revenue growth and commercialization expansion.

• Technology and product frameworkAgent = Model + Harness

  AI Agent capability is jointly determined by the foundation model and the external engineering system.

  The report defines harness as the system and engineering layer beyond the model, including task orchestration, tool integration, execution environment, and feedback loops, used to manage context, call tools, and execute multi-step workflows.

• Short-term quantitative assessmentQuantitative Research Review

  UBS's questionnaire-style quantitative assessment of analysts' views on short-term factors.

  This assessment reflects short-term judgments over the next six months on industry structure, regulatory environment, share price trends, and earnings surprise risk, with a time horizon different from the 12-month stock rating.

Asset mapping & comparison

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

• MiniMax Group (0100.HK)

  Research target

  Strengths

  It has exposure to AI foundation models, multimodal generation, open platforms, and AI-native applications; its model parameter efficiency, inference speed, and pricing competitiveness are strong; its compute supply and infrastructure optimization are considered advantageous.

  Weaknesses

  The company remains loss-making, with UBS forecasting negative EPS from 2026E to 2028E; valuation is highly dependent on future ARR growth and commercialization delivery.

  Comparison

  The report argues that MiniMax achieves model intelligence close to domestic peers with a smaller total parameter count and activated parameter scale, while showing advantages in input/output pricing, latency, and output speed.

  Risks

  Macroeconomic and geopolitical uncertainty, constrained AI adoption, competitive and operational challenges, model training data risks, user-generated content governance risks, disruptions from key service providers, and shareholder lock-up expiry pressure.

• Hailuo AI / Hailuo 3

  Core product and potential catalyst

  Strengths

  Its video generation and video understanding capabilities may become further differentiated, and harness optimization can lower the user barrier.

  Weaknesses

  Competition in video generation is intense, and expansion from professional users to mass users still requires continued improvement in product experience and cost.

  Comparison

  The report expects Hailuo 3 to widen its lead over peers in video understanding capability.

  Risks

  Model upgrade progress, user adoption, and paid conversion falling short of expectations.

• MaxHermes

  AI Agent product and harness case

  Strengths

  It applies harness engineering optimization in memory, scale, tool use, task state, and user feedback, enabling extraction of reusable skills from completed tasks and dynamic loading.

  Weaknesses

  Actual commercialization scale, retention, and paid conversion still need to be validated.

  Comparison

  The report cites LangChain research showing that upgrading only the harness architecture can raise the pass rate on Terminal Bench 2.0 from 52.8% to 66.5%, illustrating the value of the engineering layer.

  Risks

  AI Agent product experience, stability, and enterprise use-case deployment falling short of expectations.

Key data

• 12-month ratingBuyThe 12-month rating disclosed in the main body of the report.
• Target priceHK$1,000.00Corresponding to the current price of HK$777.50 on April 24, 2026.
• Forecast share price upside28.6%Forecast dividend yield is 0.0%, and forecast total stock return is 28.6%.
• Market capitalizationHK$244b / US$31.1bDisclosed in the report's trading data and key indicators.
• 52-week rangeHK$1,238.00-345.00The original report format is HKS1,238.00-345.00.
• TPM growth10%-20% WoWThe report believes MiniMax's compute expansion remains solid.
• M2.7 text model gross margin>40%UBS believes it is comparable to global leading vendors and higher than domestic peers.
• 2026E revenueUS$221,212kUBS forecast basis.
• 2026E EPSUS$(1.60)UBS diluted EPS forecast.
• 2026E net debt/EBITDA2.2xDisclosed in the report's trading metrics.
• Forecast excess return17.7%Forecast stock return of 28.6% versus an assumed market return of 10.9%.

Impact & implications

The report's investment implication is positive: the MiniMax investment story does not rely only on AI coding, but on whether model capability, multimodal synergy, inference cost, harness engineering, and expansion into real application scenarios can together drive ARR growth. If iterations such as M3, M4, Hailuo 3, and MaxHermes deliver as expected, the market may reassess its revenue scale, profit margins, and valuation multiple. However, if competition intensifies, commercialization progresses more slowly than expected, or lock-up expirations disturb sentiment, the share price could see substantial volatility.

Risks

• Macroeconomic and geopolitical uncertainty may limit AI adoption.
• Competitive and operational challenges may delay commercialization and compress profit margins.
• Data risks related to model training may affect model performance and compliance.
• User-generated content governance risks may create regulatory and brand pressure.
• Disruptions from key service providers may affect product delivery.
• Lock-up expirations for IPO cornerstone investors and some pre-IPO investors in July and October may trigger share price volatility.
• If advanced hardware export restrictions do not ease, model upgrades and compute supply may be constrained.

What to watch

• Potential major model upgrades in May-June, including progress related to M3 and M4.
• Hailuo 3's differentiated performance in video understanding and video generation.
• Adoption of MaxHermes and other harness-engineered products in office and creative scenarios.
• Potential inclusion in the HSCI and HSTECH indices in June.
• Potential Stock Connect inclusion in August.
• Share price performance after lock-up expirations for IPO cornerstone investors and some pre-IPO investors in July and October.
• Model economics indicators such as TPM growth, inference gross margin, input/output pricing, latency, and output speed.
• ARR growth, developer adoption of the open platform, and paid conversion of AI-native applications.