Zhipu AI developing Custom AI chips amid 27x surge in GLM-5.2 usage

Chinese artificial intelligence startup Zhipu AI is reportedly exploring the development of a custom application-specific integrated circuit (ASIC) to power its growing AI services, as demand for its latest GLM-5.2 large language model has surged dramatically. According to reports, usage of GLM-5.2 has increased 27-fold since its launch, prompting the company to evaluate dedicated AI hardware to improve performance, reduce costs, and lessen its dependence on third-party chip suppliers.

The move reflects a broader trend across the AI industry, where leading developers are investing in custom silicon to optimize AI workloads and secure long-term computing capacity amid rising demand for generative AI applications.

Zhipu AI Considers Developing a Custom AI Chip

According to The Information, Zhipu AI is studying the possibility of designing its own ASIC specifically for artificial intelligence inference and model deployment. Unlike general-purpose processors, ASICs are built to perform specific tasks more efficiently, offering improvements in speed, power consumption, and operating costs.

While the company is still in the early stages of evaluating the project, a proprietary AI chip could help Zhipu AI better support its expanding portfolio of foundation models and enterprise AI services.

Developing a custom ASIC would also provide greater control over hardware optimization, allowing the company to tailor chip architecture to the unique requirements of its GLM family of language models.

GLM-5.2 Adoption Accelerates

The reported chip exploration comes as Zhipu AI experiences rapid adoption of its latest language model, GLM-5.2.

According to the report, usage of GLM-5.2 has surged 27 times compared with earlier levels, reflecting growing interest from developers and enterprise customers seeking Chinese-developed large language models.

The rapid increase in inference requests places additional pressure on computing infrastructure, making hardware efficiency increasingly important as AI workloads continue to expand.

Growing demand also highlights the competitive position of GLM-5.2 in China’s rapidly evolving AI market, where domestic companies are racing to develop alternatives to Western foundation models.

Why Custom ASICs Matter for AI

AI companies increasingly view custom silicon as a strategic advantage.

Unlike traditional CPUs or even general-purpose GPUs, ASICs are designed for specific computational workloads. This specialization enables several benefits, including:

• Faster AI inference performance.
• Lower energy consumption.
• Reduced operating costs.
• Improved scalability for high-volume AI services.
• Better optimization for proprietary AI models.

As inference demand continues to grow, these advantages can significantly lower the cost of serving millions of AI requests every day.

Reducing Dependence on Third-Party Chips

One of the biggest motivations behind custom AI chip development is reducing reliance on external hardware suppliers.

Many AI developers currently depend on graphics processors from companies such as NVIDIA for both model training and inference. However, increasing global demand for AI accelerators, combined with export restrictions affecting advanced chip availability in China, has encouraged domestic AI firms to explore alternative hardware strategies.

By developing its own ASIC, Zhipu AI could gain greater control over hardware availability while optimizing infrastructure for its specific software stack.

China’s AI Hardware Push Continues

Zhipu AI’s reported plans align with China’s broader efforts to strengthen domestic AI capabilities across both software and semiconductor technologies.

Chinese AI companies have increasingly invested in locally developed foundation models while exploring partnerships with domestic chipmakers to reduce reliance on imported computing hardware.

Custom AI chips are becoming an important part of this strategy, enabling companies to improve efficiency while navigating supply chain challenges and evolving technology regulations.

Rising Competition in AI Infrastructure

The global AI industry is witnessing an increasing shift toward vertically integrated AI infrastructure, where companies develop both software models and specialized hardware.

Several leading AI developers are pursuing custom chip initiatives to optimize their platforms for large-scale deployment.

For AI providers, controlling both hardware and software offers multiple advantages, including improved performance, lower operational expenses, and the ability to rapidly introduce new AI features without depending entirely on external semiconductor vendors.

What It Means for the AI Industry

If Zhipu AI proceeds with its custom ASIC project, it would represent another milestone in the growing convergence of AI software and semiconductor design.

As large language models become more capable and attract millions of users, infrastructure costs are emerging as one of the industry’s biggest challenges. Purpose-built AI chips could play a critical role in making advanced AI services more efficient, scalable, and commercially sustainable.

The reported 27-fold increase in GLM-5.2 usage underscores the accelerating adoption of generative AI in China and highlights why companies are increasingly viewing custom silicon as a strategic investment rather than simply a hardware upgrade.

Looking ahead, the success of future AI platforms may depend not only on the intelligence of their models but also on the efficiency of the chips powering them.