MiniMax ARR Significantly Exceeds Expectations; Near-Term Gross Margin Under Pressure but Long-Term Revenue Forecasts Raised

Overview

The report focuses on MiniMax's stronger-than-expected ARR growth, pressure on 1H26 gross margin, model launch plans, and earnings forecast revisions. Morgan Stanley believes commercialization progress and long-term revenue potential have strengthened, while near-term margins, R&D investment, and computing costs remain headwinds. It therefore maintains its Overweight rating and HK$900 target price.

Core views

ARR and commercialization progress were significantly stronger than expected. Management said ARR had exceeded US$800 million as of August 2026, surpassing Morgan Stanley's previous expectation of US$600 million; the report therefore raised its year-end ARR forecast from US$1 billion to US$1.3 billion. Of this US$800 million in ARR, toB accounts for 80% and toC for 20%. Elsewhere, the report states that August ARR reached RMB800 million and that management is confident it will exceed RMB1 billion by year-end; both sets of figures are presented as stated in the original report. Demand-side growth was similarly rapid: Token consumption increased 20-fold from January to July 2026, while the customer count rose tenfold from 200,000 at the end of 2025 to 2 million currently. The report attributes the growth to adoption of the M3 text model and H3 video model, as well as improved commercialization execution. Enhanced platform stability and higher TPS have enabled customers to undertake production-grade deployments; use cases have expanded from coding and agents to broader office productivity applications; industry price increases have improved the competitive environment; and H3 began contributing incremental growth in August. M3.1, M3 Pro, and H3.1 comprise the main product pipeline for the second half of 2026, and existing computing capacity is expected to support these iterations. MiniMax is also accumulating computing capacity to train a 10-trillion-parameter model for 2027. 1H26 gross margin was below expectations, but management believes the pressure mainly reflected one-off or temporary factors during the M3 launch phase rather than structural deterioration. AI-native products themselves maintained positive gross margins and were not a drag. Specific negative factors included investments in platform stability and customer compensation during the transition, low inference efficiency in the initial launch stage, a higher proportion of low-margin Token packages priced aggressively to accelerate adoption, and an increased revenue contribution from lower-margin text models relative to higher-margin multimodal models. Management expects gross margin to improve sequentially in 2H26, driven by continued reductions in inference costs, greater commercialization discipline, and premium models such as M3 Pro supporting higher pricing and better unit economics. After incorporating 1H26 results, Morgan Stanley raised its 2026, 2027, and 2028 revenue forecasts by 25.3%, 25.0%, and 24.6%, respectively, to reflect higher ARR expectations. At the same time, the report lowered its gross margin forecasts because 1H26 gross margin missed expectations and raised its R&D expense assumptions due to increased training and computing costs for large-parameter models. Consequently, its non-IFRS operating loss forecasts for 2026–2028 widened by 29.6%, 14.4%, and 15.4%, respectively, while loss-per-share forecasts widened by 34.6%, 16.2%, and 18.0%, respectively. The financial summary forecasts revenue of US$500 million, US$1.497 billion, and US$3.727 billion for 2026–2028, respectively, and EBITDA losses of US$788 million, US$848 million, and US$619 million, respectively, indicating that despite rapid revenue expansion, the company is expected to remain loss-making throughout the forecast period. The report treats the performance of the next-generation large-parameter model scheduled for release in October 2026 as a key scenario variable for 2027 growth. The bull-case scenario assumes the new model becomes global SOTA and outperforms peers, thereby driving demand, market share, and potential price increases; it forecasts 2027 revenue of US$2.5 billion and a target price of HK$2,230 based on 40x 2027 P/S. The base-case scenario assumes the model reaches global SOTA levels and generates nonlinear revenue growth; it forecasts 2027 revenue of US$1.5 billion and a target price of HK$900 based on 25x 2027 P/S. The bear-case scenario assumes the model fails to outperform peers, with 2027 revenue growing only linearly to US$1 billion and a target price of HK$220 based on 10x 2027 P/S. For valuation, Morgan Stanley uses DCF with assumptions of a 15% WACC and a 3% perpetual growth rate. The HK$900 target price remains unchanged because higher long-term revenue forecasts offset the widening of near-term net loss forecasts; the target price implies 25x 2027 P/S, below the company's historical average one-year forward P/S of 137x. Based on Morgan Stanley's forecasts, the stock was then trading at 10x 2026 P/ARR and 9x 2027 P/S, which the report considers undervalued. Its Overweight thesis is also based on the nonlinear expansion opportunities in global AI foundation models and MiniMax's industry position in advanced technology, omni-modal capabilities, diversified AI applications, and a scalable business model.

Analysis framework

The report first compares management's disclosed August ARR with Morgan Stanley's previous expectations, then explains the sources of growth through customer count, Token consumption, business mix, and the model pipeline. It subsequently dissects the pressure on 1H26 gross margin, distinguishing temporary launch-related factors from structural issues. On this basis, the research team adjusts its revenue, gross margin, R&D expense, and loss forecasts, constructs bull-, base-, and bear-case scenarios based on the relative performance of the next-generation model scheduled for October 2026, and finally presents the target price and valuation implications through cross-referencing DCF and P/S multiples.

Methodology notes

• Valuation MethodDCF Discounted Cash Flow

  Discounted Cash Flow Valuation

  The report discounts future cash flows to their present value, using a 15% WACC and a 3% perpetual growth rate to derive a target price of HK$900.

• Valuation MethodPS valuation

  Price-to-Sales Valuation

  The report uses 2027 P/S to assess the target price and different scenarios: 40x, 25x, and 10x for the bull, base, and bear cases, respectively, and compares the base-case multiple with the then-current 9x 2027 P/S and the historical average one-year forward P/S of 137x.

• Valuation Method

  Bull-, Base-, and Bear-Case Scenario Analysis

  Based on whether the next-generation model scheduled for October 2026 can reach or exceed global SOTA levels, the report establishes different demand, revenue, and valuation paths to illustrate the impact of key technological outcomes on 2027 revenue and the target price.

• Corporate Fundamentals and Financial Framework

  Morgan Stanley ModelWare Framework

  Unless otherwise stated, the financial metrics and Morgan Stanley forecasts in the report are based on the ModelWare framework, which standardizes forecasting methodologies for revenue, expenses, profit, and the balance sheet.

Asset mapping & comparison

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

• MiniMax (00100.HK)

  The report believes MiniMax benefits from expanding demand for AI foundation models, commercialization of M3 and H3, omni-modal capabilities, and next-generation model iterations, and maintains its Overweight rating.

  Strengths

  Advanced technology, omni-modal capabilities, diversified AI applications, a scalable business model, and rapidly growing customer count and Token consumption.

  Weaknesses

  1H26 gross margin was weighed down by launch investments, customer compensation, low initial inference efficiency, and a low-margin product mix, while R&D and computing investments further widened near-term losses.

  Comparison

  The report believes whether its next-generation model can reach or exceed global SOTA and peer performance is key to determining whether 2027 revenue shifts from linear to nonlinear growth.

  Risks

  Geopolitical risks, intensifying competition and price wars, and model performance lagging peers.

Key data

• August 2026 ARRAbove US$800mnDisclosed by management, above Morgan Stanley's US$600mn expectation
• Year-End ARR ForecastUS$1.3bnRaised from US$1bn
• ARR Business MixtoB 80%; toC 20%Based on the US$800mn ARR disclosed by management
• Token Consumption Growth20xFrom January to July 2026
• Customer Count2mnUp tenfold from 200k at the end of 2025
• 2026–2028 Revenue Forecast Revisions+25.3%, +25.0%, +24.6%Reflecting higher ARR expectations
• 2026–2028 Non-IFRS Operating Loss Forecast Revisions+29.6%, +14.4%, +15.4%Losses widened due to lower gross margin forecasts and higher R&D expense assumptions
• 2026–2028 Loss-Per-Share Forecast Revisions+34.6%, +16.2%, +18.0%Loss per share widened
• 2026–2028 Revenue ForecastsUS$500mn, US$1,497mn, US$3,727mnMorgan Stanley forecasts
• Core DCF AssumptionsWACC 15%; perpetual growth rate 3%Used to value the HK$900 target price
• Base-Case Scenario2027 revenue US$1.5bn; 25x 2027 P/S; HK$900Assumes the next-generation model reaches global SOTA levels
• Bull-Case Scenario2027 revenue US$2.5bn; 40x 2027 P/S; HK$2,230Assumes the next-generation model becomes global SOTA and outperforms peers
• Bear-Case Scenario2027 revenue US$1bn; 10x 2027 P/S; HK$220Assumes the next-generation model fails to outperform peers
• Current Valuation10x 2026 P/ARR; 9x 2027 P/SBased on Morgan Stanley forecasts

Impact & implications

The report believes that rapid growth in customer count and Token usage indicates that M3 and H3 have accelerated commercialization sufficiently to support higher medium- and long-term revenue forecasts. The near-term cost is pressure on gross margin and losses due to launch investments, inference efficiency, low-margin packages, and higher R&D and computing costs. If inference costs decline, premium-model pricing increases, and next-generation model iterations deliver as expected, revenue growth and unit economics should improve. The higher long-term revenue forecasts and wider near-term losses offset each other, leaving the target price unchanged at HK$900.

Risks

• Geopolitical risks may affect the company's business development and valuation.
• Intensifying market competition and price wars may weaken pricing power and unit economics.
• If model performance lags peers, 2027 revenue may grow only linearly and fail to achieve the expected step-up.

What to watch

• Monitor the launches and commercialization contributions of M3.1, M3 Pro, and H3.1 in the second half of 2026.
• Monitor whether the next-generation large-parameter model scheduled for October 2026 can reach or exceed global SOTA levels.
• Monitor whether year-end ARR can reach Morgan Stanley's upgraded forecast of US$1.3 billion and management's stated target of more than RMB1 billion.
• Monitor whether lower inference costs, commercialization discipline, and premium-model pricing can drive sequential improvement in 2H26 gross margin.
• Monitor progress in training the 10-trillion-parameter model for 2027 and accumulating computing capacity.