Goldman Sachs Report Breaks Down China’s AI Foundation Model Competitive Landscape: Who Will Become the Long-Term Winner?

Author: Wall Street Insights, Bu Shuqing

Original title: Goldman Sachs in-depth report: Who will become the long-term winner in China’s AI large model industry?

China’s AI large models are at a historic turning point. Goldman Sachs believes that the intelligent performance of China’s open-source/open-weight large models has approached the global top proprietary models, while the adoption scale of domestic companies and global small and mid-sized enterprises is rapidly expanding. This, in turn, will create a data flywheel effect that further drives iterative upgrades of the models.

According to Follow the Wind Trading Desk, Goldman Sachs’ latest report summarizes this evolution path as: “from last year’s cost-efficiency moment of DeepSeek to this year’s model intelligence moment of Zhipu GLM.” In a 50-page report led by analyst Ronald Keung, the team conducts a systematic evaluation of four core questions: how Chinese AI models achieve high performance at low cost, why they choose an open-source route, how they monetize, where the core addressable markets are, and who will become the long-term winner.

On the competitive landscape assessment, Goldman Sachs has rolled out a “competitive positioning framework” based on pricing capability, cost advantages, and financial strength. Based on it, Goldman Sachs concludes that in the base text model space, Zhipu (first coverage) and DeepSeek (not publicly listed) are positioned as the strongest; in the multimodal space, ByteDance (not publicly listed) leads. Goldman Sachs also maintains Buy ratings on MiniMax and Kuaishou.

Small wins, big gains—efficiency wins

China’s large models can achieve performance close to that of similar US products at far lower costs. The core lies in a dual breakthrough in architecture innovation and parameter efficiency.

Goldman Sachs’ report states that in China, the parameter scale of open-source models is generally between 200 billion and 1.6 trillion, only 2% to 10% of the world’s top models—mainly because access to high-end compute resources is constrained. Meanwhile, innovations such as the Mixture of Experts (MoE) architecture and sparse attention mechanisms reduce the share of actually activated parameters to only 3% to 5% of total parameters, significantly lowering training and inference costs.

In terms of specific models, DeepSeek V4 Pro has 1.6 trillion parameters, Zhipu GLM5.2 has 0.7 trillion, and MiniMax M3 has 0.4 trillion.

Goldman Sachs attributes the recent leap in Chinese models’ programming ability to the coordinated effects of data filtering and post–reinforcement learning fine-tuning. On June 27, DeepSeek released its speculative decoding framework DSpark, which has been deployed in the online services of V4-Flash and V4 Pro. Without changing model weights or output quality, it increases per-user generation speed by 60% to 85% (V4-Flash) and 57% to 78% (V4 Pro).

Meituan’s LongCat 2.0 released on June 30 is viewed by Goldman Sachs as an important milestone for the self-sufficiency of China’s AI infrastructure—China’s first open-source MoE model with 1.6 trillion parameters trained and deployed entirely based on 50k domestically produced compute cards. Goldman Sachs believes this demonstrates the feasibility of a localized hardware stack during compute-intensive pretraining and has far-reaching significance for China’s AI models to break away from dependence on foreign high-end chips.

Market polarizes—stronger players get stronger

Goldman Sachs describes China’s AI model market as forming a “two-tier structure” and identifies two ARR-maximization quadrants.

In the high-end market, top models represented by Zhipu GLM5.2 and Alibaba’s Qwen3.7 Max are priced at about $1 per million tokens, which is 5 times that of low-end models. The inference gross margin is about 10% to 20% (estimated by Goldman Sachs). By comparison, top models in the US are priced at $4 to $8 per million tokens; China’s high-end models are only 10% to 25% of that, but thanks to a lower parameter activation ratio, they can still maintain positive gross margins.

In the low-end market, models targeting agent tasks are priced as low as $0.06 to $0.2 per million tokens, opening up a market for globally price-sensitive small and mid-sized enterprises and individual users. MiniMax has 60% to 70% of its revenue coming from overseas. Of note, DeepSeek has announced that starting from mid-July it will introduce peak and off-peak pricing for the V4 series, with the peak rate at 2 times the off-peak rate; blended pricing is about $0.35 per million tokens (V4 Pro) and $0.12 per million tokens (V4 Flash).

Goldman Sachs forecasts that China’s AI model API and subscription revenue will grow from 35 billion RMB estimated for 2026 to 879 billion RMB in 2030, corresponding to daily token consumption increasing from 350 trillion to 4,600 trillion, an approximately 25x increase.

Open-source strategy: broad penetration—monetization paths still need upgrading

Goldman Sachs’ report lays out in detail the strategic logic behind China’s widespread adoption of open-source/open-weight routes for AI models, and its limitations in monetization.

The core advantage of open-source strategy lies in deployment flexibility and community ecosystems. Alibaba’s Qwen series, DeepSeek, Zhipu GLM, and MiniMax M3 all adopt open-source or open-weight approaches; ByteDance’s Seed model is a major exception, using a fully closed proprietary route. The open model allows flexible deployment both inside and outside mainland China, and accelerates iteration through community feedback.

However, Goldman Sachs points out that the ARR figures disclosed by open-source model companies are likely to severely underestimate actual deployment scale and revenue potential. Taking Zhipu as an example, its target ARR by end of 2026 is $1 billion, but the real global deployment volume of GLM5.2 will be far higher than the token volume and revenue from Zhipu’s own API channels—Alibaba Cloud’s Bailian MaaS platform can directly host the open-source GLM5.2 model without paying any fees to Zhipu.

Goldman Sachs expects the industry to gradually shift from pure open-source (MIT license, completely free) to a “open-weight + community license” model—meaning commercial use requires revenue-sharing agreements signed with the model companies. MiniMax’s M series has taken the lead in adopting this model. Goldman Sachs believes this shift will significantly improve unit economics for AI model companies, because model companies can benefit from revenue-sharing agreements with platforms such as AWS Bedrock and Alibaba Cloud Bailian, without having to bear inference compute costs themselves.

From “token maximization” to prioritizing ROI

Goldman Sachs qualitatively characterizes international market expansion as the most important upside space for China’s AI models, especially in non-US markets.

Goldman Sachs’ US research team estimates that by 2030, agent AI will drive a 24x increase in global token consumption, reaching 120 trillion tokens per month. Enterprises’ agent AI contributes 55x growth, while consumers’ agent AI contributes 12x growth. In the global market (outside China), China’s AI models have already gained significant token share growth by leveraging performance improvements and pricing advantages.

Goldman Sachs’ report notes that the global enterprise AI usage paradigm is undergoing a fundamental shift from “token maximization” to “ROI first.” The former was dominant from late 2025 to early 2026, with enterprises equating high token consumption to organizational productivity. The latter focuses more on clear task boundaries, the number of daily active agents, backend workflow automation, and actual outputs. A Jellyfish AI engineering trends study shows that heavy AI users in enterprises consume 10x the tokens, but output increases only 2x.

On the channel front, Alphabet’s Gemini Enterprise Agent Platform and Amazon’s AWS Bedrock have already provided hosting services for China’s AI models such as DeepSeek, MiniMax, Moonshot, GLM, and Qwen. As reported by The Wall Street Journal, Microsoft CEO recently said Microsoft is considering hosting versions of DeepSeek on Copilot as an optional, low-cost model, emphasizing that if DeepSeek is hosted, the model will run within Microsoft’s cloud ecosystem to ensure customer data remains within Azure.

Who is the long-term winner?

Goldman Sachs builds a three-dimensional competitive positioning framework to assess each player’s probability of long-term outperformance using quantitative indicators. The core formula is: ARR scale × gross margin advantage + financial strength.

The pricing capability dimension examines listing speed (compared with prior versions and peer-level models), the LMArena arena score (based on evaluations from large-scale blind user tests), and the blended pricing level per million tokens.

The cost advantage dimension examines throughput (tokens per second), cache hit rate, parameter activation ratio, and inference gross margin. The financial strength dimension examines cash on hand, net cash as a proportion of total assets, and valuation multiples.

In base text model space, Goldman Sachs identifies Zhipu (first coverage, neutral rating, target valuation $16k) and DeepSeek (not publicly listed) as the strongest; both stand out in pricing capability and cost advantages. The total implied valuation of independent AI model companies overall exceeds $200 billion.

In multimodal/video generation, ByteDance leads with Seedance. According to LatePost and 36Kr, Seedance’s gross margin is as high as 70%, and its ARR run rate has exceeded $2 billion. Kuaishou Lingjing and MiniMax Hailuo/the upcoming H3 model are also viewed favorably by Goldman Sachs. It expects that in the second half of 2026, they will benefit from functional breakthroughs from the fusion of video generation and LLMs, as well as healthy pricing driven by supply tightness.

Goldman Sachs maintains a Buy rating on MiniMax, with a target price of HK$860. The rationale is that its M3 model sits in the ARR-maximization quadrant of high token volumes and attractive pricing, and its current valuation is only 13x its ARR by end of 2026. Compared with valuation multiples of similar companies in China and globally, there is an obvious discount, so the risk-reward profile is skewed upward.