Let us know about free updates
Simply sign up for myft AI digest and it will be delivered directly to your inbox.
China’s state-run data centers are washed away with artificial intelligence chips and are now the technology group’s expertise to bolster billions of dollars in investment as part of a national effort to increase adoption of high-speed development technologies. We are using.
According to people with knowledge of issues and public documents, local governments adopt merit interactive and startup infinite AI and silicon flows to help surge in AI data centers created across China was developed.
It shows that China is increasingly pooling resources from the private and provincial sectors as part of a push to accelerate its network of AI computing infrastructure.
“At this point, the bottleneck hasn’t gotten chips, but we’re finding ways to make it work in the cluster. This is a really complicated job,” said the Beijing-based chip investor.
The move came after Chinese startup Deepseek attracted global attention last month with the latest in a series of AI model releases that performed similarly to rivals such as Openai and Google. It was achieved with what appears to be the budget.
The Zhijiang-based company uses its engineering talent to narrow down as much computing power as possible from the Nvidia GPU, reducing the cost of model training and so-called inference, and reduce the process of calling large language models. did.
State-owned data centers are leveraging this type of private sector engineering talent amid the AI infrastructure boom across China.
Official data shows that by mid-2024, 250 AI data centres had been completed or under construction. According to people familiar with the deal, local governments are financing these billion-dollar projects to issue bonds to secure tips and tips for these billion-dollar projects, according to people has played a central role in However, they added that many of the projects suffered from a lack of technical know-how and wasted important chip resources.
Infinigence AI has emerged as one of the main beneficiaries of the state’s push to sign agreements with Ningxia and Tianjin’s national AI data centres.
Backed by Hongshan and Qiming Venture Partners, the Shanghai-based startup will lend Nvidia chips to developers building AI applications. It makes money between the difference paying the data center operator and what charges customers’ end access to the tip.
In December, the Shanghai government will be infinitely building a platform for customers to create AI applications on third-party models hosted in data centers run by local derivatives of state-owned China Telecom. We have announced a collaboration with
Both companies will build inference engines that speed up LLMS by adopting pre-trained models and adjusting appropriately with data and compressing the size of AI models.
One investor said such collaborations are important for courting AI clients. Technology companies run tests before signing contracts with data centers to ensure there is no confusion while running expensive training.
Zhijiang-based Merit Interactive also leverages its engineering expertise. The Zhijiang Media Outlet has published a meeting with Merit Interactive executives and local officials. There, the company was chosen for its role in building a large AI data center in Wenzhou city.
The group is associated with business partners of Liang Wenfeng, founder of his Quant Trading Fund High-Flyer Deepseek, according to someone with knowledge of the issue.
According to local media reports, Xu Jin, who co-founded High Flyer with Liang, previously worked as technical director at Merit Interactive. Merit Interactive has business benefits across the import and export of high-tech hardware.
The US has a similar set of “Neocloud” companies, including CoreWeave, Lambda and Vultr, which manage and rent access to Data Center’s computing clusters, providing companies with the ability to use tips in their payment ranges .
Recommended
Despite Washington tightening Nvidia’s export controls on high-end chips, China has been able to develop its AI industry. With black markets thriving for smuggled chips emerging, industry insiders have seen the hugely popular H100 prices in recent months as the number of illegal chips brought to China has swelled We are focusing on the decline.
State-owned data center operators have preferred access to Huawei’s Ascend AI chip, the leading challenger of China’s Nvidia for the growing reasoning market.
Like DeepSeek, both Infinigence AI and SiliconFlow are working to reduce inference costs to make using LLMS cheaper. We are also working with Huawei to make the Ascend chip compatible to infer with the major Chinese AI models.
This month, Huawei and Siliconflow announced a collaboration to make Deepseek’s AI model available through Telecom Giant’s cloud service. According to anyone familiar with the issue, Deepseek worked with both companies to deploy the model to Huawei chips on China’s New Year’s holiday.
Additional Reports by Michael Acton of San Francisco
