FOLLOWING GLOBAL TRENDS
With an in-house chip, DeepSeek would be joining other global AI developers in seeking greater control over the hardware behind their models and reduced dependence on Nvidia’s.
OpenAI last month unveiled Jalapeno, its first custom inference chip, developed with Broadcom, while Anthropic has been weighing building its own AI chips, Reuters reported in April.
For DeepSeek, the effort carries an added strategic dimension. US export controls bar Chinese companies from buying Nvidia’s most advanced chips, and Beijing has been pressing its technology champions to build domestic alternatives.
DeepSeek founder Liang Wenfeng said in a rare 2024 interview with a Chinese media outlet that chip export controls were a challenge for the company.
DeepSeek has used both Nvidia and Huawei chips. The company has said the foundation model underpinning R1, the reasoning model whose low-cost performance triggered a rout in US tech stocks in January 2025, was trained on Nvidia’s H800, a chip designed for the Chinese market that Washington banned in late 2023.
The company has since leaned increasingly on Huawei. In April, it released its V4 model adapted for Huawei’s Ascend chips, and Huawei said its processors were used in part of the training of V4-Flash, a lighter version of the model. Orders for Huawei’s Ascend 950 chips from Chinese tech conglomerates surged after the launch, Reuters has reported.
TAPPING INFERENCE DEMAND
A DeepSeek inference chip would target the fastest-growing segment of AI computing demand. As AI applications spread, more of the industry’s computing work is shifting from training models to running them, which relies on specialised chips that can be cheaper and less power-hungry than general-purpose GPUs.
However, there is no guarantee of success. Designing a competitive AI chip typically takes years and significant capital. Manufacturing poses another hurdle as the US bans Chinese designers from accessing the most advanced overseas foundries, while separate US curbs have cut China’s access to high-bandwidth memory, a component critical to AI inference chips.
DeepSeek’s chip push coincides with the company’s first embrace of outside capital. The company was slated to raise US$7 billion in a maiden funding round valuing it at between US$52 billion and US$59 billion, Reuters reported in June, reversing its years-long strategy of rejecting external investment.
Read the full article here
