Forget ChatGPT? China's cheaper AI models are quietly taking over enterprise AI, says Jefferies

The competitive landscape of artificial intelligence is evolving, with a shift from simply having the most advanced models to offering similar performance at significantly lower prices. A recent report by Jefferies highlights that China's latest large language model, GLM-5.2, could pose a substantial challenge to Western AI giants, particularly in the enterprise sector where cost, security, and return on investment are paramount. Developed by Z.ai, formerly known as Zhipu AI and listed in Hong Kong, GLM-5.2 is being compared to the notable DeepSeek moment, as described by Jefferies strategist Christopher Wood. DeepSeek previously revolutionized the AI arena by demonstrating that sophisticated models could be developed affordably. Now, GLM-5.2 aims to capture enterprise customers by delivering performance comparable to Anthropic's top AI systems at just a quarter of the cost per token. This pricing advantage could significantly reshape the AI landscape. Anthropic has quickly risen to become one of the fastest-growing AI companies, with projections of its annual revenue jumping from $9 billion at the close of 2025 to an astounding $47 billion by May 2026. OpenAI is also expanding its reach and preparing for a potential public offering. Both firms have built their enterprise strategies around high-end AI products. However, if businesses begin to transition to more affordable options with similar functionalities, it could dramatically alter the economics of enterprise AI. Jefferies indicates that this is where Chinese AI models may have the most substantial influence. Instead of competing on benchmark scores, developers from China focus on affordability. As AI capabilities continue to improve, it may become harder for businesses to rationalize higher costs for only marginal performance gains. Evidence of this trend is illustrated in recent usage statistics. According to data from AI platform OpenRouter, Chinese models processed 21.37 trillion tokens in the week ending June 21, a stark increase from 4.37 trillion tokens in late April. In contrast, leading US models processed only 5.76 trillion tokens during the same timeframe, showcasing the swift uptake of these cost-effective Chinese alternatives. The report suggests that this trend signifies a broader movement towards the commoditization of large language models. As the technology matures, elements such as pricing, flexibility in deployment, and data privacy are likely to become more crucial competitive factors than raw performance. Lower token costs could also lead enterprises to rethink their AI deployment strategies. Instead of relying solely on public cloud providers, businesses might prefer to run smaller models on their own infrastructure, which could enhance data security and address increasing concerns surrounding cybersecurity and data governance. Interestingly, these cheaper AI solutions might benefit semiconductor manufacturers rather than diminish their market. Jefferies cites the Jevons Paradox, suggesting that reduced AI costs will motivate businesses to implement AI across a broader range of applications, thus driving greater demand for computing power and related technologies. While Jefferies maintains a positive outlook on AI hardware amidst growing competition, it also warns that the main long-term threat to AI investments may not stem from the rise of Chinese models, but rather from potential investor skepticism regarding whether companies like OpenAI and Anthropic can deliver adequate returns on the significant investments made in AI infrastructure. Until such concerns arise, Jefferies anticipates that AI capital expenditure will continue to thrive. In this context, GLM-5.2 may ultimately be more disruptive than DeepSeek—not due to superior intelligence, but because of its challenge to the industry's pricing structure.