The Chip and the Cost

The chip in question is the one Nvidia adapted for China in order to comply with U.S. export controls. The American company estimated earlier this month that the U.S. would allow it to resume sales to China after a three-month pause, though it is unclear when shipments will begin.

Zhang said the company does not need to purchase more of the chips for now, as it currently has sufficient computing power. However, he declined to disclose how much Z.ai spent to train the AI model. Further details will be published later, he said.

The Cost of the Model

In January, DeepSeek shocked (mainly) global investors with its apparent ability to bypass U.S. chip restrictions and create an AI model that not only competed with OpenAI’s ChatGPT but also undercut it in training and operational costs.

DeepSeek claimed that training costs for its V3 model were under $6 million, although some analysts said that figure was based only on the company’s hardware expenses, which had exceeded $500 million over time.

Z.ai stated that for its new GLM-4.5 model, it will charge 11 cents per 1 million input tokens, compared to 14 cents for DeepSeek R1, and 28 cents per 1 million output tokens, compared to $2.19 for DeepSeek. Tokens are a way of measuring data used in AI model processing.

In July, Moonshot, backed by Alibaba, launched Kimi K2, which claimed to outperform OpenAI’s ChatGPT and Anthropic’s Claude in certain coding capabilities. Kimi K2 charges 15 cents per 1 million input tokens and $2.50 per 1 million output tokens, according to the company’s website.

Sanctions and IPO Plans

At the end of June, OpenAI named Zhipu in a warning about the progress of Chinese AI. The U.S. has also added the startup to its entity list, which prohibits U.S. companies from doing business with it.

Z.ai was founded in 2019 and is reportedly planning an initial public offering (IPO) in the Greater China region.

The Chinese startup has raised over $1.5 billion from investors including Alibaba, Tencent, and Qiming Venture Partners, according to PitchBook. Prosperity7 Ventures, backed by Aramco, and municipal funds from Hangzhou and Chengdu are also among Z.ai’s supporters, according to the database.

In recent weeks, several other Chinese companies have also announced new open-source AI models. During the World Artificial Intelligence Conference in Shanghai, Tencent launched the HunyuanWorld-1.0 model for generating 3D scenes for game development. Last week, Alibaba announced its Qwen3-Coder model for computer code generation.

source: https://www.ot.gr/2025/07/29