tech

Chinese tech firms from Huawei to Inspur push ‘all-in-one’ DeepSeek AI servers

Dozens of Chinese hi-tech manufacturers – from Lenovo Group and Huawei Technologies to Inspur Group – are pushing new “all-in-one” servers that include DeepSeek‘s advanced artificial intelligence (AI) models to private and public enterprises across the country, ramping up democratisation of the technology.

According to a report on Thursday by state-owned investment bank China International Capital Corp (CICC), about 5 per cent of the country’s AI server demand will be generated from industries that produce sensitive information including government, finance, public services and healthcare. These sectors are expected to adopt the DeepSeek-powered AI machines, lifting total 2025 sales in the domestic AI server market to 54 billion yuan (US$7.4 billion).

More than 30 Chinese server manufacturers currently produce these boxes with embedded DeepSeek AI models, according to the CICC report. These include China’s largest server maker Inspur, second-ranked H3C Technologies, Huawei and Lenovo, the world’s top personal computer vendor.

Do you have questions about the biggest topics and trends from around the world? Get the answers with SCMP Knowledge, our new platform of curated content with explainers, FAQs, analyses and infographics brought to you by our award-winning team.

These servers, equipped with AI chips, are designed to allow enterprises that lack certain technical resources to quickly deploy AI applications in their operations.

Apart from their easy set-up process, these new AI servers have become popular for being deployed at enterprises’ on-premise data centres, which satisfies a security requirement for sensitive data in certain industries, according to the CICC report.

The rapid adoption of DeepSeek technology in the country’s server industry reflects how the Hangzhou-based start-up has made AI more accessible to enterprises across various industries.

DeepSeek’s breakthrough was its development of advanced open-source AI models, V3 and R1, built at a fraction of the cost and computing power that major tech companies typically require for large language model (LLM) projects.

LLMs are the technology underpinning generative AI services like ChatGPT, Baidu‘s Ernie Bot and DeepSeek’s namesake chatbot.

Open source gives public access to a program’s source code, allowing third-party software developers to modify or share its design, fix broken links or scale up their capabilities. Open-source technologies have been a huge contributor to China’s tech industry over the past few decades.

Inspur last month said its new all-in-one servers with DeepSeek’s R1 allowed customers to deploy the reasoning model at various scales from its full 671-billion parameter version to a distilled 1.5-billion version.

Parameters refer to variables present in an AI system during training. The sophistication and effectiveness of an AI model depend largely on the scale of parameters involved in the training process.

Inspur’s plug-and-play machines can use both foreign AI chips, such as Nvidia‘s H20, as well as domestic alternatives. Chinese-made AI chips were expected to become the primary computing processors for these machines, as they align more with customer demand, according to the CICC report.

The H20 – an AI chip tailored by Nvidia for the Chinese market amid US tech export restrictions – is currently also at risk of being barred for export to the mainland, as the Trump administration considers new export control measures.

Huawei’s Ascend graphics processing units (GPUs), including the 910B, are widely used in DeepSeek AI servers now available in the market, according to the CICC report.

Huawei last month launched its own DeepSeek AI server, the Atlas 800I A2. The Shenzhen-based company said this server lets customers flexibly deploy either DeepSeek’s V3 or R1 model with low operational and maintenance costs.

Also last month, China Telecom launched its own DeepSeek-integrated AI server that the carrier said achieved “full localisation” because it used Huawei’s Kunpeng and Ascend chips.

Many Chinese integrated circuit developers have also adapted their products to support DeepSeek, offering the AI industry more home-grown options.

Moore Threads Technology, the Chinese AI chip design unicorn founded by former Nvidia China general manager Zhang Jianzhong, last month said that it would “pay tribute to DeepSeek” by “using locally made GPUs to set China’s AI ecosystem on fire”.

A growing number of local governments in China are also rushing to adopt DeepSeek’s AI models to enhance administration and governance, reflecting the rising adoption of the technology in the public sector.

This article originally appeared in the South China Morning Post (SCMP), the most authoritative voice reporting on China and Asia for more than a century. For more SCMP stories, please explore the SCMP app or visit the SCMP’s Facebook and Twitter pages. Copyright © 2025 South China Morning Post Publishers Ltd. All rights reserved.