China's latest AI model claims to be even cheaper to use than DeepSeek

Chinese AI startup Z.ai announced its new GLM-4.5 model, which significantly undercuts competitors like DeepSeek on pricing, charging 11 cents per million input tokens and 28 cents per million output tokens. This open-source, 'agentic' model is half the size of DeepSeek's and optimized for Nvidia's China-specific H20 chips, demonstrating China's progress in developing cost-efficient and competitive AI despite U.S. export controls. Z.ai, backed by major investors and reportedly planning an IPO, exemplifies a broader trend of Chinese firms releasing advanced AI models that challenge global leaders, even as the U.S. adds them to its entity list.

Analysis

The Chinese artificial intelligence landscape is demonstrating rapid, capital-efficient innovation, exemplified by startup Z.ai's new GLM-4.5 model. This development is significant not just for its technological claims of 'agentic' AI, but for its aggressive pricing strategy, which substantially undercuts competitors. Z.ai is charging 11 cents per million input tokens and, more notably, just 28 cents per million output tokens, compared to DeepSeek's $2.19 and Alibaba-backed Moonshot's $2.50 for the same output. This signals an intense price war and a focus on commoditizing AI services within China. Furthermore, the model's ability to operate on only eight of Nvidia's China-specific H20 chips highlights a successful effort to circumvent the impact of U.S. export controls by optimizing for less powerful hardware. This trend, echoed by recent open-source releases from Tencent and Alibaba, confirms that Chinese firms are developing competitive AI capabilities despite geopolitical pressures. Z.ai's substantial backing of over $1.5 billion from investors including Alibaba and Tencent, combined with its reported IPO plans and inclusion on the U.S. entity list, positions it as a key player to watch in the global AI rivalry.

Key Decisions for Investors

• Investors should recognize that the Chinese AI sector is developing models that are not only technologically competitive but also aggressively priced, potentially compressing margins for global players and creating distinct investment opportunities in key backers like Alibaba (BABA) and Tencent.
• The hardware efficiency of new models like GLM-4.5, which operate effectively on restricted chips, suggests U.S. sanctions may not be a long-term barrier to Chinese AI progress, a factor that must be priced into the competitive outlook for U.S. AI firms.
• For investors in Nvidia (NVDA), the development of highly efficient models that require fewer chips could temper long-term sales forecasts for its China-specific hardware, warranting a closer look at demand elasticity in the region.
• Given Z.ai's significant venture funding and reported plans for a Greater China IPO, institutional investors should monitor for public filing announcements to assess a new, high-profile entry into the public AI market.