DeepSeek V4 Will Run on Huawei AI Chips, Report Says - Intermediate - EN

Read in: Difficulty

DeepSeek is reportedly preparing its next artificial intelligence model, called V4, and a new report says the system will run entirely on Huawei AI chips. That detail is important because more Chinese technology companies are turning to domestic hardware as access to some Nvidia products becomes more difficult in China.

According to the report, DeepSeek V4 may use a nearly 1-trillion-parameter architecture and possibly hundreds of thousands of Ascend 950PR chips. If that is true, the model could reach 1.8 times faster inference speeds, support a 1-million-token context window, and improve efficiency through a technology called Engram.

The report also says DeepSeek has worked with Huawei for months and with the Chinese chipmaker Cambricon to rewrite and test important parts of the model’s core code. Neither DeepSeek nor Huawei has officially confirmed these claims, but the idea is believable because DeepSeek has already used Ascend chips in earlier AI models.

V4 could arrive in the coming weeks and may focus on high-performance coding and reasoning. The company is also said to be developing two more V4 variants that use China-made chips and may launch later this year. If the report is accurate, DeepSeek’s next model could become another sign that China’s AI industry is building more of its future on local hardware.

Table of Contents

Vocabulary Words List

artificial intelligence model — a computer system built to perform AI tasks
run entirely — operate fully without using another system
Huawei AI chips — Huawei chips designed for AI work
technology companies — businesses that create or use technology
domestic hardware — technology equipment made inside the country
Nvidia products — computer products made by Nvidia
1-trillion-parameter architecture — a huge AI design with about one trillion settings
Ascend 950PR chips — Huawei’s advanced chips for AI
inference speeds — the speed at which a model gives results
1-million-token context window — the ability to handle a very large amount of text at once
efficiency — the ability to work well with less waste
technology called Engram — a system mentioned as helping the model work better
worked with Huawei — cooperated with Huawei
Chinese chipmaker — a company in China that makes computer chips
rewrite — to write something again in a new way
test — to check if something works correctly
core code — the main programming that runs a system
officially confirmed — publicly said something is true
claims — statements that may or may not be true
believable — easy to accept as possible
Ascend chips — Huawei’s AI chip family
earlier AI models — older AI systems made before
coming weeks — the next few weeks
high-performance coding — very strong ability to write or understand code
reasoning — careful thinking to solve problems
developing — creating over time
V4 variants — different versions of V4
China-made chips — chips produced in China
launch later this year — be released sometime later this year
local hardware — technology equipment made nearby or in the same country

Fill In The Blanks Listening Practice

DeepSeek is reportedly preparing its next , called V4, and a new report says the system will on . That detail is important because more Chinese are turning to as access to some becomes more difficult in China.

According to the report, DeepSeek V4 may use a nearly and possibly hundreds of thousands of . If that is true, the model could reach 1.8 times faster , support a , and improve through a .

The report also says DeepSeek has for months and with the Cambricon to and important parts of the model’s . Neither DeepSeek nor Huawei has these , but the idea is because DeepSeek has already used in .

V4 could arrive in the and may focus on and . The company is also said to be two more that use and may . If the report is accurate, DeepSeek’s next model could become another sign that China’s AI industry is building more of its future on .