MiniMax Goldman Sachs conference call: Confident in reaching $1 billion ARR this year, model advantage lies in "organizational agility," and highly integrated with domestic chips.

On July 4, according to news from Chasewind Trading Desk, Goldman Sachs released its latest research report on July 3, stating that the MiniMax conference call sent a strong signal of commercialization and technological evolution: The management is highly confident in achieving the goal of $1 billion in annual recurring revenue (ARR) by the end of 2026.

The report points out that the most core catalyst is that China's AI large model industry is ushering in a turning point of the "price war"—as competitor DeepSeek announced price increases during peak hours, industry pricing is returning to rationality.

MiniMax, relying on over 90% computing power utilization, high integration with domestic chips, and unique "organizational agility", maintains highly competitive pricing (M3 model blended price $0.22 per million tokens) while achieving gross margins far above peers. Additionally, the soon-to-be-launched H3 video generation model will further expand the imagination space of the multimodal market.

Goldman Sachs maintains a Buy rating with a 12-month target price of HKD 860, implying a 141% upside from the current price of HKD 356.80.

ARR Growth Trajectory: From $100 Million to $1 Billion, Management Gives Clear Path

The report states that MiniMax management systematically outlined the ARR (Annual Recurring Revenue) growth milestones during the conference call:

End of December 2025: ARR reaches $100 million;February 2026: ARR rises to $150 million;April 2026: ARR doubles again compared to February;Before M3 model goes live on June 1: ARR further accelerates upward.

Management clearly stated full confidence in achieving $1 billion ARR by the end of 2026.

On pricing strategy, M3 keeps the same pricing as the previous generation M2.7, but management emphasized its sustainability from a gross margin perspective—reason being that upgrades to training and inference architecture have brought over double cost savings, basically offsetting the cost increase caused by doubled parameters.

The company also previewed that a larger-scale M3 series model will be launched in the second half of 2026, aiming to further enhance intelligence while maintaining strong cost performance.

This ARR growth curve is the core basis supporting Goldman Sachs’ revenue forecast—Goldman Sachs expects MiniMax’s revenue to jump from $79 million in 2025 to $300 million in 2026, further reaching $880.1 million in 2027, and breaking $2,469.6 million in 2028.

DeepSeek Price Increase: Icebreaking Signal for Rational Industry Pricing, Direct Benefit to MiniMax

This is the most market-catalytic external event in the Goldman Sachs report.

DeepSeek announced this week that its official V4 version will be launched in mid-July, simultaneously introducing a peak/off-peak API differentiated pricing mechanism: Peak hours (Beijing time 9 am to 12 pm, 2 pm to 6 pm) will be charged at twice the off-peak rate, with blended prices about $0.35 (Pro version) / $0.12 (Flash version) per million tokens.

Goldman interprets this as: Since the end of April 2026, the aggressive pricing by Chinese AI model companies (some players with zero or even negative gross margins) is entering an early stage of transition to more rational pricing, essentially reflecting real inference cost pressure in pricing.

By comparison, MiniMax M3’s blended price is $0.22 per million tokens, with a significant advantage in performance/price ratio and obviously higher gross margin than peers—thanks to its higher proportion of self-built optimized computing power and efficient inference architecture using fewer activated parameters.

MiniMax also highlighted its self-operated computing power achieving over 90% utilization, balancing peaks and valleys by serving knowledge workers and developers during peak hours and using idle computing power for experiments and data sorting during off-peak times, supporting cost advantage for long-duration agent workflows.

H3 Video Model: Release Within Weeks, Deeply Integrated with M3 Architecture

Meanwhile, MiniMax is about to launch the next-generation video generation model H3, expected to officially release "in the next few weeks."

H3’s core upgrade features are in two dimensions:

• Comprehensive improvement in video generation quality and functional diversity, driven by major architectural upgrades (including optimization of annotation/classification/feedback loops);
• Deep integration with the M3 model architecture: Large language model capabilities are embedded in H3’s DiT (Diffusion Transformer) architecture, such as enhanced understanding of human movements and basic physical relationships.

Additionally, MiniMax is bringing in vertical field experts, gradually entering the feature film/series production market to expand the commercial boundaries of video generation.

Competitive Landscape: From "Hundred Model Battle" to Convergence, "Organizational Agility" Becomes Core Barrier

Goldman Sachs believes MiniMax’s judgment on China’s AI model competitive landscape is strategically significant: The market is quickly concentrating towards the top players from hundreds of players one or two years ago.

In the conference call, in response to competition from AI labs under domestic internet giants, MiniMax defined its advantages as:

• Efficient enterprise organizational structure;
• Higher infrastructure utilization rate;
• Faster model iteration capability;
• Quick response to emerging agent opportunities—for example, rapidly commercializing MaxClaw after the rise of OpenClaw, and rapid deployment of MiniMax Code product.

Management believes that as AI model competition shifts from "one-off benchmark rankings" to "continuous product iteration and real-world deployment," sustainable ROI will become the core evaluation standard, and the value of organizational agility will be increasingly highlighted under this new competitive paradigm.

Global Infrastructure and Domestic Chip Integration: Accelerating Localization

At the computing power infrastructure level, MiniMax adopts a dual-track parallel strategy:

Directly renting computing power from global cloud service providers (CSP);Deep cooperation with emerging cloud service providers (neo-cloud).

Currently, MiniMax’s localized inference infrastructure has covered more than 200 countries and regions worldwide, with highly dispersed customer structure and no risk of over-concentration in any single country.

In the China market, MiniMax has highly integrated domestic AI chips (ASIC) for inference tasks. As domestic chip capabilities continue to improve, this localization process is accelerating. This layout not only helps reduce dependence on overseas computing power, but also builds supply chain resilience in the context of China-U.S. tech competition.

In talent strategy, MiniMax supports high-intensity technology competition with a very lean team:

• 400 to 500 employees company-wide, of which over 80% are engaged in R&D;
• 300 to 400 employees participate in about 7% equity via ESOP (Employee Stock Ownership Plan) to strengthen talent retention through equity incentives;
• Continuously recruits fresh graduates from top Chinese and overseas universities;
• Introduces vertical field experts via "10X Talent Plan", converting industry know-how into model training and real task optimization capabilities.

~~~~~~~~~~~~~~~~~~~~~~~~

The above exciting content comes from Chasewind Trading Desk.

For more detailed interpretation, including real-time commentary and frontline research, please join Chasewind Trading Desk▪Annual Membership】

Risk Warning and DisclaimerThe market carries risks; investment should be cautious. This article does not constitute personal investment advice and does not consider the unique investment goals, financial status, or needs of individual users. Users should consider whether any opinions, viewpoints, or conclusions in this article fit their specific circumstances. Invest accordingly, at your own risk.