Key Takeaways DeepSeek is creating a proprietary AI semiconductor tailored for inference operations rather than training This strategic shift aims to decrease dependence on Nvidia and Huawei
According to three informed sources, Chinese artificial intelligence company DeepSeek has initiated development of a proprietary semiconductor. This processor is specifically engineered for inference tasks—the computational process that enables AI systems to generate outputs—as opposed to the resource-intensive training of new models.
DeepSeek is reportedly developing its own AI chip, per Reuters.
The chip is being designed for inference, not training, as DeepSeek looks to reduce reliance on Nvidia and Huawei chips.
DeepSeek is reportedly in talks with foundries, memory suppliers and chip-design firms, while… pic.twitter.com/xi3kFertIC
— Wall St Engine (@wallstengine) July 7, 2026
DeepSeek has not issued any official statement regarding these reports.
Following this news, Nvidia’s stock price decreased by roughly 2% in premarket trading sessions.
Inference processors differ significantly from the high-performance GPUs required for training AI systems. These chips generally cost less and consume substantially less power. With artificial intelligence applications expanding across industries, the market for inference computing power is experiencing rapid expansion.
DeepSeek’s semiconductor is specifically designed to address this expanding market segment.
The initiative remains in preliminary development stages. DeepSeek has initiated conversations with chip design companies, semiconductor fabrication facilities, and memory manufacturers. Additionally, the firm has been discreetly recruiting chip engineering talent through private channels rather than public job postings, according to two sources.
Historically, DeepSeek has utilized processors from both Nvidia and Huawei for developing and operating its AI systems. American export regulations prohibit Chinese enterprises from purchasing Nvidia’s cutting-edge processors. These restrictions have compelled DeepSeek to increasingly depend on Huawei’s Ascend chip lineup in recent periods.
This past April, DeepSeek unveiled its V4 model optimized for Huawei’s Ascend platform. Demand for Huawei’s Ascend 950 processors experienced a significant spike following this product launch.
However, Huawei’s dominance in China’s $50 billion AI semiconductor market faces mounting challenges. Both Alibaba and Baidu are engineering proprietary chips and capturing market territory.
DeepSeek’s semiconductor initiative places it alongside similar industry movements. OpenAI introduced Jalapeno, its inaugural custom inference processor developed with Broadcom, last month. Anthropic has similarly explored potential chip development projects.
For DeepSeek, additional obstacles exist. American regulations prevent Chinese chip designers from utilizing the most sophisticated international semiconductor foundries. Additional restrictions constrain access to high-bandwidth memory, an essential component for inference processors.
Liang Wenfeng, DeepSeek’s founder, acknowledged in a 2024 interview that export restrictions presented genuine obstacles for the organization.
This semiconductor initiative coincides with DeepSeek’s first acceptance of external capital. Reuters reported in June that the company planned to secure $7 billion at a valuation ranging between $52 billion and $59 billion. This represents a significant departure from years of rejecting outside investment.
DeepSeek captured worldwide attention in early 2025 following the viral success of its R1 reasoning model. This model underwent training using Nvidia’s H800 chip, a China-specific variant that Washington subsequently prohibited.
Despite emerging as one of China’s most scrutinized AI companies, DeepSeek has maintained a relatively discreet public presence. The ultimate success of its semiconductor venture will hinge on overcoming both engineering complexities and persistent American trade restrictions.
The post DeepSeek Develops Custom AI Chip as Nvidia (NVDA) Stock Falls 2% appeared first on Blockonomi.
