Leading model developers are increasingly seeking to control more of their compute stacks, particularly for inference, as AI models move from training labs into the mass market.

DeepSeek is the latest company to pursue this strategy by developing a new in-house inference chip, Reuters reports.

If successful, this effort would mark a major shift for the Chinese AI chatbot company and add new pressure on Nvidia's long-term dominance across the AI compute market. It would also add pressure on Huawei, which has benefited from US export controls that blocked Chinese firms from buying Nvidia's most advanced AI chips.

US companies such as Meta Platforms, Microsoft, OpenAI, and Anthropic are also increasingly turning to in-house chips to reduce AI infrastructure costs.

According to the report, DeepSeek's effort to design a new inference chip is in the "early stages," suggesting that developing a competitive AI chip could take a few years.

A new Bloomberg Intelligence survey found that an increasing number of Chinese companies are shifting away from Nvidia's advanced AI chips.

Chinese executives expect to allocate 46% of their AI accelerator budgets to domestic AI infrastructure over the next 12 months, up from 30% today, according to the survey.

The findings point to growing momentum behind Beijing's push to replace Nvidia in domestic chip stacks.

"China's drive to substitute locally made AI semiconductors for foreign ones is making progress, which is likely to benefit domestic markers such as Huawei and Hygon," said the report, which surveyed dozens of executives at Chinese software, finance, manufacturing and retail companies.

For Nvidia, the risk is not an immediate loss of market share, but a longer-term shift in which Chinese companies and AI chatbot startups increasingly seek in-house or domestically produced chips, suggesting mounting headwinds for Nvidia.

Meanwhile, the token-maxxing fiasco appears to have accelerated a global shift toward low-cost models, a trend that may favor Chinese AI chatbots.

Bloomberg cited OpenRouter data showing that Chinese models are capturing a growing share of developer usage, suggesting that China's AI firms are becoming more competitive in global model adoption.