摩根士丹利首次覆盖MiniMax：2025-2027年收入预测与全球AI模型领导者分析

报告核心观点概览

摩根士丹利首次覆盖中国人工智能公司MiniMax，给予其“增持”评级，并设定930港元的目标价。该行将MiniMax定位为“全球AI基础模型领导者”。这份报告的核心押注点并非短期财务表现，而是聚焦于两个关键维度：模型能力是否位居全球第一梯队，以及收入结构是否具备全球扩张的弹性。

Morgan Stanley initiated coverage on Chinese AI company MiniMax with an "Overweight" rating and a target price of HK$930. The firm positions MiniMax as a "global AI foundational model leader." The core bet of this report is not on short-term profitability but focuses on two critical dimensions: whether the model capability ranks among the global top tier, and whether the revenue structure possesses the flexibility for global expansion.

据追风交易台信息，负责此次覆盖的摩根士丹利分析师Gary Yu判断，MiniMax已成功跻身全球顶尖（SOTA）模型阵营，其多模态能力完善，商业化路径具备高度可扩展性。基于此，公司收入有望从2025年的7500万美元大幅增长至2027年的7亿美元，实现两年内9-10倍的跃升。报告指出，一旦技术能力形成代际优势，收入曲线将呈现“台阶式”而非线性的增长。

According to information from Zhui Feng Trading Desk, Gary Yu, the Morgan Stanley analyst responsible for this coverage, judges that MiniMax has successfully entered the global state-of-the-art (SOTA) model camp. Its multimodal capabilities are robust, and its commercialization path is highly scalable. Based on this, the company's revenue is expected to surge from $75 million in 2025 to $700 million in 2027, achieving a 9-10x increase within two years. The report notes that once technological capability establishes a generational advantage, the revenue curve will exhibit a "step-function" leap rather than linear growth.

报告对高估值的解释非常直接：MiniMax被视为一种“技术决定收入上限、全球化决定估值体系”的资产。如果其模型性能能持续保持在第一梯队，其增长天花板将由全球市场总规模（TAMTotal Addressable Market，总可寻址市场，指特定产品或服务在理论上能够达到的最大市场规模。）决定；而如果其收入主要来源于海外市场，其估值锚点自然会向国际可比公司看齐。

The report offers a straightforward explanation for the high valuation: MiniMax is viewed as an asset where "technology determines the revenue ceiling, and globalization determines the valuation framework." If its model performance can consistently remain in the top tier, its growth ceiling will be determined by the global Total Addressable Market (TAMTotal Addressable Market，总可寻址市场，指特定产品或服务在理论上能够达到的最大市场规模。). Furthermore, if its revenue is primarily sourced from overseas markets, its valuation anchor will naturally align with international comparable companies.

技术能力：估值的起点

摩根士丹利将MiniMax的核心竞争力归结为以下三点：

持续迭代能力
多模态布局
成本效率

Morgan Stanley attributes MiniMax's core competitiveness to the following three points:

Continuous Iteration Capability
Multimodal Layout
Cost Efficiency

在独立基准测试中，MiniMax的模型表现亮眼：其M2模型发布时位列全球大语言模型（LLM）排行榜第五；最新的旗舰模型M2.5位列第六，并在开源模型中排名第四。更具说服力的是实际应用数据：截至2026年2月中旬，M2.5在OpenRouter平台按token使用量排名第一，处理量达到1.97万亿token，在编程场景的市场份额高达58.8%。这些数据表明，MiniMax的模型已深入高频、真实的商业调用场景，而非仅仅停留在实验室评测阶段。

In independent benchmark tests, MiniMax's models have shown impressive performance: its M2 model ranked fifth on the global Large Language Model (LLM) leaderboard upon release; the latest flagship model, M2.5, ranks sixth overall and fourth among open-source models. More convincing is the practical application data: as of mid-February 2026, M2.5 ranked first on the OpenRouter platform by token usage, processing 1.97 trillion tokens, and held a 58.8% market share in programming scenarios. This data indicates that MiniMax's models have penetrated high-frequency, real-world commercial deployment scenarios, moving beyond mere lab benchmarks.

成本结构是另一个关键优势。公司采用混合专家（MoE）架构与线性注意力（Linear Attention）机制，在推理阶段的模型浮点运算利用率（Model Flop Utilization模型浮点运算利用率，衡量硬件在执行模型推理时实际利用计算资源的效率百分比，越高代表硬件利用率越好。）超过75%，显著高于行业40%-50%的平均水平。推理效率直接决定了API的定价竞争力和毛利率弹性，是规模扩张时利润率能否同步改善的核心。大摩预计，公司毛利率将从2024年的12%提升至2027年的32%。

Cost structure is another key advantage. The company employs a Mixture of Experts (MoE) architecture and Linear Attention mechanisms, achieving a Model Flop Utilization模型浮点运算利用率，衡量硬件在执行模型推理时实际利用计算资源的效率百分比，越高代表硬件利用率越好。 rate exceeding 75% during the inference stage, significantly higher than the industry average of 40%-50%. Inference efficiency directly determines the competitiveness of API pricing and the flexibility of gross margins. It is the core factor determining whether profitability can improve alongside scale expansion. Morgan Stanley forecasts the company's gross margin to increase from 12% in 2024 to 32% in 2027.

然而，报告也明确指出，技术领先并不直接等同于盈利。大摩预计，尽管毛利率改善，公司的经营亏损仍将持续扩大，2027年非国际财务报告准则（非IFRS）下的经营亏损预计约为4.84亿美元。这并非基于盈利拐点的投资逻辑，而是一条“先扩大技术与规模，再看利润”的发展路径。技术能力虽不保证短期盈利，但决定了长期收入的天花板。

However, the report clearly states that technological leadership does not directly translate to profitability. Morgan Stanley expects that despite improving gross margins, the company's operating losses will continue to widen, with non-IFRS operating losses projected to be approximately $484 million in 2027. This is not an investment thesis based on a profitability inflection point but rather a development path of "first expanding technology and scale, then focusing on profits." While technological capability does not guarantee short-term profitability, it determines the long-term ceiling for revenue.

收入结构：决定增长斜率

MiniMax的商业模式并非依赖单一产品驱动，而是三条业务线并行发展：

2C（面向消费者）：智能体与陪伴类产品，如Talkie/Xingye。
2P（面向合作伙伴/专业用户）：海螺AI、MiniMax Audio等。
2B（面向企业）：开放平台API服务。

MiniMax's business model is not driven by a single product but operates on three parallel business lines:

2C (To Consumer): Agent and companion products, such as Talkie/Xingye.
2P (To Partner/Professional): Products like Hailuo AI, MiniMax Audio.
2B (To Business): Open Platform API开放平台应用程序编程接口，允许第三方开发者通过标准化接口调用MiniMax的模型能力，集成到自己的应用或服务中。 services.

截至2025年前九个月，公司月活跃用户（MAU）已从2023年的310万增长至2760万，付费用户达177万。其收入结构正趋于多元化，其中开放平台（Open Platform）的收入占比持续提升。摩根士丹利预计，开放平台收入占比将从2024年的29%提升至2027年的40%，未来三年复合年增长率（CAGR）将超过200%。

As of the first nine months of 2025, the company's Monthly Active Users (MAU) grew from 3.1 million in 2023 to 27.6 million, with paying users reaching 1.77 million. Its revenue structure is becoming more diversified, with the revenue share from the Open Platform continuously increasing. Morgan Stanley forecasts that the Open Platform's revenue share will rise from 29% in 2024 to 40% in 2027, with a compound annual growth rate (CAGR) exceeding 200% over the next three years.

报告强调了一个行业关键特征：基础模型公司的增长往往由关键代际模型的突破所触发，呈现“跳跃式”放量，而非平滑的线性爬坡。OpenAI的ChatGPT 3.5、Anthropic的Claude 3.5 Sonnet都在模型重大升级后带来了收入的阶跃式增长。MiniMax是否会复制这种节奏，很大程度上取决于其计划于2026年中推出的下一代模型的表现。

The report emphasizes a key industry characteristic: the growth of foundational model companies is often triggered by breakthroughs in key generational models, resulting in "jump-style" volume expansion rather than smooth linear growth. Both OpenAI's ChatGPT 3.5 and Anthropic's Claude 3.5 Sonnet led to step-function revenue increases following major model upgrades. Whether MiniMax replicates this pattern depends largely on the performance of its next-generation model, slated for release in mid-2026.

全球化：估值的前提

摩根士丹利特别强调了MiniMax的“天生全球化”（Born Global）路径。其海外市场收入占比已从2023年的19%快速提升至2025年前九个月的73%。区域分布为：亚太地区61%，美洲24%，欧洲、中东及非洲（EMEA）15%。

Morgan Stanley particularly emphasizes MiniMax's "Born Global" trajectory. Its overseas revenue share has rapidly increased from 19% in 2023 to 73% in the first nine months of 2025. The regional breakdown is: Asia-Pacific 61%, Americas 24%, and Europe, Middle East & Africa (EMEA) 15%.

在全球基础模型市场规模预计将从2024年的107亿美元增长至2029年的2065亿美元（年复合增长率80.7%）的宏大背景下，MiniMax当前全球市场份额仅约0.3%。这意味着，只要其市场份额获得小幅提升，就会带来巨大的收入弹性。

Against the grand backdrop of the global foundational model market size projected to grow from $10.7 billion in 2024 to $206.5 billion in 2029 (CAGR 80.7%), MiniMax's current global market share is only about 0.3%. This implies that even a slight increase in its market share would result in significant revenue elasticity.

更重要的是估值体系的切换。如果公司收入主要来自海外市场，且客户以API调用和订阅为主，那么其估值逻辑将更接近于国际AI同行（如OpenAI、Anthropic等），而非传统的中国软件公司。这也是摩根士丹利给予其54倍2027年市销率（P/SPrice-to-Sales Ratio，市销率，公司市值与其营业收入的比例，用于评估成长型公司的估值水平。）估值的核心依据。

More importantly, this enables a shift in the valuation framework. If the company's revenue is primarily from overseas markets and its clients are mainly API users and subscribers, then its valuation logic aligns more closely with international AI peers (e.g., OpenAI, Anthropic) rather than traditional Chinese software companies. This is the core rationale behind Morgan Stanley's valuation of 54x 2027 Price-to-Sales (P/SPrice-to-Sales Ratio，市销率，公司市值与其营业收入的比例，用于评估成长型公司的估值水平。) ratio.

估值分歧与风险聚焦

报告通过三种清晰的情景假设来划分估值分歧：

基准情景：2027年收入达7亿美元，对应54倍P/SPrice-to-Sales Ratio，市销率，公司市值与其营业收入的比例，用于评估成长型公司的估值水平。，目标价930港元。
乐观情景：2027年收入达10亿美元，目标价1240港元。
悲观情景：2027年收入为4亿美元，目标价300港元。

The report delineates the valuation divergence through three clear scenario assumptions:

Base Case: 2027 revenue reaches $700 million, corresponding to a 54x P/SPrice-to-Sales Ratio，市销率，公司市值与其营业收入的比例，用于评估成长型公司的估值水平。, target price HK$930.
Bull Case: 2027 revenue reaches $1 billion, target price HK$1240.
Bear Case: 2027 revenue is $400 million, target price HK$300.

决定不同情景下估值差异的核心变量只有一个：2026年中推出的下一代模型是否达到或超越全球SOTA水平。

The core variable determining the valuation difference across scenarios is singular: whether the next-generation model launched in mid-2026 reaches or surpasses global SOTA levels.

报告同时指出了明确的风险：

GPU供应紧张与地缘政治限制。
与OpenAI及其他超大规模玩家的资源差距。
模型商品化可能带来的价格压力。
持续的现金消耗。

The report also identifies clear risks:

GPU supply constraints and geopolitical restrictions.
Resource gap compared to OpenAI and other hyperscale players.
Potential pricing pressure from model commoditization.
Sustained cash burn.

结论：一场关于“技术兑现能力”的定价

摩根士丹利并未回避现实：目前尚无纯粹的AI基础模型公司实现稳定盈利。MiniMax在2025年预计月均现金消耗约2790万美元，盈利可见度有限。

Morgan Stanley does not avoid the reality: no pure-play AI foundational model company has achieved stable profitability yet. MiniMax is projected to have an average monthly cash burn of approximately $27.9 million in 2025, with limited profitability visibility.

但报告的核心判断在于——基础模型行业的竞争本质不是营销战，而是代际技术突破的竞赛。技术能力决定了收入的天花板，而全球市场定位决定了估值的锚点。如果模型升级能带来非线性的收入扩张，那么当前的估值只是对未来市场规模的提前折现；反之，如果模型未能站稳全球第一梯队，估值收缩也同样会迅速发生。

However, the report's core judgment is that competition in the foundational model industry is inherently not about marketing but about the race for generational technological breakthroughs. Technological capability determines the revenue ceiling, while global market positioning determines the valuation anchor. If model upgrades lead to non-linear revenue expansion, then the current valuation is merely a discounting of future scale. Conversely, if the models fail to maintain a position in the global top tier, valuation contraction can occur just as rapidly.

这本质上是一场关于“技术兑现节奏”的押注。摩根士丹利的选择是，站在“全球顶尖基座模型稀缺资产”这一边。

This is essentially a bet on the "pace of technological realization." Morgan Stanley's choice is to side with "scarce assets of globally top-tier foundational models."

特别声明：以上内容(如有图片或视频亦包括在内)为自媒体平台“网易号”用户上传并发布，本平台仅提供信息存储服务。

Notice: The content above (including the pictures and videos if any) is uploaded and posted by a user of NetEase Hao, which is a social media platform and only provides information storage services.