Zhipu Increases Losses by 60% Amid Fierce Competition in China's AI Market

Chinese AI company Zhipu ended 2025 with significantly deeper losses than analysts expected: net losses increased by 60%. For the market, this is not just a weak financial result of one company, but an indicator of how expensive the race for leadership in Chinese AI has become.

Losses Growing Faster

A 60% increase in losses indicates that Zhipu's expenses are growing faster than the company manages to turn interest in its models into sustainable revenue. And what matters is not only the figure itself, but also that it fell short of expectations. Typically this means the market underestimated the speed at which an AI startup needs to burn capital on model training, infrastructure, and launching new products.

For Zhipu, such dynamics are particularly telling because the Chinese AI market is currently operating in constant acceleration mode. One successful release is no longer enough: companies must regularly release updates, improve answer quality, reduce latency, and simultaneously make services cheaper for customers. In such an environment, even notable technological progress does not guarantee quick business economics improvement.

The Cost of the Race

From the outside, the AI boom often looks like a battle of models and benchmarks, but inside it is primarily a battle of budgets. To remain among the leaders, companies have to invest simultaneously in multiple directions, and each one is expensive.

• Training and fine-tuning models requires significant computational resources and expensive cloud infrastructure.
• Inference also costs money: the more users and corporate requests, the higher the fixed expenses.
• Strong engineers, researchers, and product teams in the AI sector are becoming increasingly expensive.
• Corporate pilots, integrations, and customer support require not only models but also a full-fledged service around them.

After that, margins are also pressured by commercial reality. When the market is overheated by competition, companies often resort to discounts, free quotas, favorable pilots, and softer terms for large customers. This helps gain users and use cases, but simultaneously delays the moment when the business can demonstrate normal operational efficiency. Therefore, rapid audience or deployment growth does not always mean that the company has already achieved a sustainable monetization model.

Competition in China

Zhipu's problem does not look isolated: it reflects the broader picture of the Chinese AI market, where pressure from competitors is only intensifying. Each new release effectively nullifies some of the advantages of the previous one. If one player improves model quality or lowers API pricing, the rest are forced to respond almost immediately, otherwise they risk losing developers, corporate customers, and investor attention.

An additional complexity is that Chinese AI companies have to compete simultaneously on multiple levels. On the one hand, there are other fast-growing startups ready to invest aggressively in growth. On the other—large technology platforms with their own infrastructure, sales channels, and data access.

In such a landscape, independent players need not only to create strong models but simultaneously prove they can withstand a long financial cycle and not fall out of the technological race. This is precisely why Zhipu's financial reports read as an indicator of a broader trend. The Chinese AI market has not yet entered a phase of calm monetization.

It remains a stage where capital is spent on speed, reach, and position retention. Those companies that look technologically strong today may find themselves under pressure tomorrow if they fail to convert these advantages into repeatable revenue and controlled costs.

What This Means

Zhipu's results show that the Chinese AI race is becoming increasingly intolerant of mistakes and increasingly costly for participants. For the market, this is a signal: winners will not only be those with stronger models, but also those who can finance growth longer, hold prices under competitive pressure, and more quickly convert technology into a working business.