Chinese AI Laboratories Innovate Amid Hardware Limitations

Research institution Delphi Digital has reported that hardware constraints are prompting Chinese AI laboratories to adopt more cost-effective models and efficient technology stacks. Export control restrictions have limited access to advanced chips, making it challenging for these labs to achieve cutting-edge training capabilities. While domestic accelerators can manage inference tasks, the need for innovation in system efficiency, model architecture, training data, and reinforcement learning has become critical.

Innovations are underway as companies like DeepSeek have developed methods to minimize GPU idle data movement, with ByteDance and Moonshot expected to release similar technologies soon. The shift in architecture is leading to the adoption of cheaper digital formats that work well with domestic chips. Sparse architectures are being utilized to decrease runtime per token, and the attention mechanism has undergone three redesigns to manage memory costs associated with long contexts.

In terms of data optimization, denser prediction targets and improved optimizers are enhancing the learning signal per token. DeepSeek's GRPO has emerged as a standard in reinforcement learning, with major players like ByteDance, Alibaba, and MiniMax planning to introduce successor models within the next year. As a result, Chinese AI models are becoming some of the most cost-effective cutting-edge systems available.

Projections indicate that the share of token consumption by Chinese-developed models on OpenRouter will increase significantly, from less than 1.2% at the end of 2024 to a majority by 2026. The Chinese AI market is anticipated to undergo a price war from 2024 to early 2026, with revenue multiples for several laboratories estimated between 47-117 times, compared to 15-21 times for companies like Anthropic and OpenAI. Despite the advancements, the supply of domestic accelerators remains below the estimated demand.

© 2026 KLEA News. All Rights Reserved. This article is provided for informational purposes only. It is not offered or intended to be used as legal, tax, investment, financial, or other advice.

Source: KLEA News