Chinese AI startup DeepSeek is developing its own chip for inference to reduce reliance on Nvidia and Huawei, a strategic shift driven by U.S. export controls. The early-stage project coincides with DeepSeek's first external funding round, which seeks to raise $7 billion at a valuation up to $59 billion. The move follows a global trend of AI firms like OpenAI and Anthropic bringing hardware development in-house.
DeepSeek chip development initiative
- ▪The chip is designed for inference, the stage where a trained model generates responses, rather than for training new models
- ▪The company is holding discussions with external partners, including chip-design, foundry, and memory companies
- ▪DeepSeek has been privately hiring chip-design engineers in recent months without public job postings
- ▪Chinese AI startup DeepSeek is developing its own artificial intelligence chip
- ▪The chip development project is in its early stages and began about a year ago
Chinese AI chip market dynamics
- ▪A U.S. ban on advanced chip exports to China helped Huawei gain around half of the $50 billion domestic AI chip market
- ▪Huawei's market dominance is weakening as rivals like Alibaba and Baidu also develop their own AI chips
- ▪Developing an in-house chip could reduce DeepSeek's reliance on processors from Nvidia and Huawei
- ▪DeepSeek has previously used both Nvidia and Huawei chips, adapting its V4 model for Huawei's Ascend processors
Global AI hardware trends
- ▪Developing in-house chips is a global trend, with AI developers seeking greater control over hardware
- ▪AI company Anthropic is also reportedly considering building its own AI chips
- ▪OpenAI recently unveiled its first custom inference chip, named Jalapeno, which was developed with Broadcom
U.S. export control restrictions
- ▪U.S. restrictions also cut China's access to high-bandwidth memory and prevent Chinese designers from using the most advanced overseas foundries
- ▪DeepSeek's R1 model was trained on Nvidia's H800, a chip designed for China that the U.S. banned in late 2023
- ▪U.S. export controls bar Chinese companies from buying Nvidia's most advanced chips
AI inference chip demand
- ▪Specialized inference chips can be cheaper and less power-hungry than general-purpose GPUs used for training
- ▪Inference chips target the fastest-growing segment of AI computing demand as more work shifts from training models to running them
DeepSeek funding strategy shift
- ▪The company is slated to raise $7 billion in a funding round, valuing it at between $52 billion and $59 billion
- ▪Seeking external investment reverses DeepSeek's long-standing strategy of rejecting outside capital
- ▪DeepSeek's chip initiative coincides with the company raising outside capital for the first time
Story comments
Loading comments…