DeepSeek lives up to market disruptor reputation with in-house chips talks

Chinese AI startup DeepSeek is designing an in-house chip to run its models, according to sources close to the matter. The disclosure, if true, would reduce its dependence on both Nvidia and Huawei. 

For a company that already forced a rethink across the industry once, the plan especially puts pressure on the American chipmaker, losing ground in China, and on the domestic suppliers racing to replace it.

The race to be the leader of inference

The Hangzhou firm is aiming the chip at inference, the stage where a trained model answers user queries. Inference is where AI systems earn their keep at scale, and it is the part of the market Huawei has just started to win with its Ascend line. 

A DeepSeek chip built for the same job would put the startup in competition with the very homegrown supplier that Chinese tech companies have been scrambling to buy from.

Since DeepSeek released its R1 model in January 2025, which matched top American AI models while claiming to use far less computing power. That event challenged the assumptions about the quality and how much hardware is needed. 

When the company shipped V4 in April 2026, it optimized the model for Huawei’s Ascend chips, which is a deviation from the Nvidia GPUs it had leaned on before. 

DeepSeek’s decision set off a buying rush as ByteDance, Tencent, and Alibaba all approached Huawei about orders for its Ascend 950 processors, according to three people familiar with the matter. Shares in SMIC, which manufactures Huawei’s chips, climbed 10% in Hong Kong after the launch.

“Huawei’s Ascend chips are the country’s best homegrown alternative to Nvidia, and supporting DeepSeek V4 shows that top Chinese AI models can now run on Chinese hardware,” He Hui, director of semiconductor research at Omdia, told reporters in April.

Why is DeepSeek entering the chip-making market now?

The backdrop is a Chinese market pulling away from Nvidia at speed. Executives at Chinese firms plan to route 46% of their AI-accelerator budgets to domestic products over the next 12 months, up from 30% today, a recent Bloomberg Intelligence survey of 60 companies found. 

Nvidia’s H20, once the most capable chip it could legally sell in China, has been blocked from import after Beijing urged its tech firms to avoid it, and the newer H200 remains stuck in regulatory limbo.

For now, Huawei seems to be the biggest beneficiary of Nvidia’s loss of market share in China. However, the same US export controls that pushed Chinese buyers toward Huawei are also capping how many Ascend 950 chips it can build, as there are restrictions on advanced chipmaking equipment, which in turn limits production at the most cutting-edge nodes. 

A domestic rival designing its own inference silicon would add another claimant to a market already short on capacity.

DeepSeek’s chip comes as the company is reportedly chasing its first outside funding, a round targeting $7 billion. Controlling its own hardware would fit a strategy that has already compressed costs across the board. In May 2026, DeepSeek cut V4-Pro prices by 75%, dropping the top rate to under $0.85 per million tokens from $3.30.

However, designing a competitive chip is hard, and US rules that bar Chinese firms from advanced manufacturing tools will also come into play and potentially affect their ability to produce one at volume. 

Given the Chinese government’s investment in the AI space, companies like Huawei and Deepseek may be able to see major breakthroughs soon.

If you're reading this, you’re already ahead. Stay there with our newsletter.