DeepSeek Is Developing Its Own AI Chip for Its Neural Networks, Reuters Reports

DeepSeek, a Chinese artificial intelligence startup, has begun developing its own semiconductor chips to power AI systems. Reuters reported this on July 7, 2026, citing unnamed sources. DeepSeek has not issued any official confirmation.

Why does DeepSeek need its own chip?

The main reason is limited access to cutting-edge accelerators. US export restrictions, tightened step by step since 2023, have cut Chinese organizations off from Nvidia's H100, H200 and A100 chips, the very accelerators on which OpenAI, Google DeepMind and Anthropic build their infrastructure. Even the H20 series, which Nvidia designed specifically for the Chinese market with deliberately reduced specifications, was added to the restricted list in 2025.

DeepSeek has already shown it knows how to squeeze the most out of limited hardware: in early 2025 the startup released a series of models comparable to OpenAI's flagship offerings in quality but significantly cheaper to train. That approach, algorithmic efficiency instead of raw computing power, made DeepSeek a global sensation. A custom chip is the next logical step: optimizing hardware architecture for your own algorithms is considerably more efficient than adapting algorithms to someone else's hardware.

A proprietary hardware platform could potentially give the company:

• Independence from Nvidia supplies and from US export policy
• The ability to optimize the microarchitecture for the specifics of its own transformers
• Long-term reductions in the cost of training and inference
• A competitive advantage amid an ongoing technological blockade

Who has already gone down this road

Building a custom chip is a practice the largest Western players adopted long ago, but each took years and billions of dollars to get there. Google has been developing its TPU series since 2016, and today those chips underpin the training of Gemini. Amazon released Trainium for model training and Inferentia for inference. Meta is developing its MTIA accelerators, and Microsoft is investing in Maia chips for Azure.

In China, other companies are already on a similar path. Huawei created the Ascend series of accelerators, which are used in major domestic AI projects. Baidu develops Kunlun chips across several generations. Cambricon specializes in neural network processors and is publicly traded. All these examples show that the Chinese industry can move toward chip sovereignty, although the technological gap with Western leaders remains significant.

The fundamental problem for any Chinese chip program is manufacturing. TSMC, which produces the world's most advanced chips, is unavailable to Chinese customers because of US sanctions pressure. SMIC, China's largest manufacturer, is technologically limited to older process nodes, which significantly affects the energy efficiency and performance of any future chip.

What was left out

Reuters does not disclose the technical specifications of the future chip, does not name a development timeline and does not state the size of the investment in the project. It is unknown whether the work is at the stage of early architectural research or already at production prototyping. It is unclear whether the chip will be intended for model training, inference or both. DeepSeek has issued no official comment.

This is a typical picture for reporting on Chinese technology projects in semiconductors: companies very rarely announce developments before a product is ready, especially in a strategic area that has ended up at the center of the technological confrontation between the US and China.

What this means

If the project reaches the product stage, DeepSeek will join the narrow list of companies that control the full AI stack, from algorithms to silicon. For the global industry, this is a signal: the fragmentation of AI infrastructure into Western and Chinese ecosystems is becoming increasingly irreversible, and the race for technological sovereignty is moving beyond software down to the hardware level.

*Meta has been designated an extremist organization and is banned in Russia.