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

The artificial intelligence race may be entering a new phase where the biggest competitive advantage is no longer having the most powerful model, but offering comparable performance at a much lower price. According to Jefferies' latest GREED & FEAR report, China's newest large language model, GLM-5.2, could prove to be an even bigger competitive challenge for Western AI leaders than DeepSeek, particularly in the enterprise market where businesses are focused on costs, security and return on investment.

GLM-5.2 wave

Developed by Hong Kong-listed Z.ai, formerly known as Zhipu AI, GLM-5.2 is being described by Jefferies strategist Christopher Wood as another "DeepSeek moment." While DeepSeek disrupted the AI landscape by demonstrating that sophisticated models could be built at lower costs, GLM-5.2 takes the competition a step further by targeting enterprise customers. According to the report, industry feedback suggests the model delivers performance that is close to Anthropic's leading AI systems while costing only about one-quarter as much per token.

MUST READ: India needs its own DeepSeek to avoid AI dependence, warns Bernstein

Anthropic

That pricing advantage could have significant implications for the AI industry. Anthropic has emerged as one of the fastest-growing AI companies globally, with its annualised revenue run rate jumping from $9 billion at the end of 2025 to $47 billion in May 2026. OpenAI, too, continues to expand aggressively while preparing for a potential public listing. Both companies have built their enterprise businesses around premium AI offerings. If businesses begin switching to lower-cost alternatives with similar capabilities, the economics of enterprise AI could change dramatically.

China's latest AI models

Jefferies believes this is where China's latest AI models could have the greatest impact. Instead of competing solely on benchmark scores or advanced capabilities, Chinese developers are competing on affordability. As AI models become increasingly capable across the board, businesses may find it difficult to justify paying substantially higher prices for only incremental performance improvements.

Evidence that this shift is already underway can be seen in usage trends. Data cited by Jefferies from AI platform OpenRouter shows that Chinese AI models processed 21.37 trillion tokens during the week ended June 21, up sharply from 4.37 trillion tokens in late April. During the same period, the leading US AI models processed 5.76 trillion tokens, highlighting the rapid adoption of lower-cost Chinese alternatives.

MUST READ: India attracts AI billions, but HSBC's top investment picks are China, Japan and Korea; know why

The report argues that this reinforces a broader trend—the gradual commoditisation of large language models. As the technology matures, pricing, deployment flexibility and data privacy could become more important competitive differentiators than raw model performance.

Lower token costs may also encourage enterprises to deploy AI differently. Rather than relying exclusively on public cloud providers, companies may increasingly choose to run smaller AI models on their own servers to better protect sensitive corporate data. This shift towards on-premises AI could reduce dependence on cloud infrastructure while addressing growing concerns around cybersecurity and data governance.

Cheaper AI models

Ironically, cheaper AI models may end up benefiting semiconductor companies rather than hurting them. Citing the Jevons Paradox, Jefferies argues that lower AI costs will encourage businesses to deploy AI across more applications, driving higher overall demand for computing power, AI servers and memory chips. The brokerage therefore remains bullish on AI hardware despite intensifying competition among model developers.

MUST READ: OpenAI's next flagship AI model faces launch delay; White House demands safety checks

At the same time, Jefferies cautions that the biggest long-term risk to the AI investment cycle is not the rise of Chinese models, but whether investors eventually question if companies such as OpenAI and Anthropic can generate sufficient returns on the massive sums being invested in AI infrastructure. Until that happens, the brokerage expects AI capital expenditure to remain robust. In that context, GLM-5.2 may ultimately prove more disruptive than DeepSeek—not because it is dramatically smarter, but because it challenges the industry's ability to maintain premium pricing.

MUST READ: ‘I handed Claude my salary’: Viral X post shares 7 prompts for financial freedom