The sudden emergence of DeepSeek as a serious provider of artificial intelligence, which led to a group of local cloud service providers, chip manufacturers and others adopted the AI model of this startup.
Although Beijing is happy to promote the company, although the model cost of DeepSeek is widely believed to be Underestimated (More than the following) Local analysts are willing to succeed in the company's “watershed” for the company.
Moore threads and Hygon information technology created AI chips and hoped to compete with NVIDIA. They said on Monday that their computing clusters and accelerators will be able to support the R1 and V3 models of DeepSeek.
Please refer to: China has promoted DeepSeek AI startup before the US stock investment '
“We pay tribute to DeepSeek,” Moore Threads's post title on WeChat is the title, and added that the company's model has made progress with the use of domestic graphics processing units (GPUs), which can “fire” China's AI industry.
Huawei integrated DeepSeek model
On Saturday, Huawei Technology Co., Ltd., which also has its own AI chip series, said that it is working with AI infrastructure to launch silicon flow to enable DeepSeek models to be used on its Ascend Cloud Service.
It added that their performance is comparable to models running on high-end chips worldwide.
Bernstein analysts said in a report on Sunday that Huawei's news that Huawei has integrated DeepSeek's model with rising chips marks “Basin Moment”.
They added: “DeepSeek proves that competitive large language models (LLM) can be deployed on the” good enough “chips that have been deployed in China, relaxing the dependence on the cutting -edge American hardware.”
Alibaba, Baidu and Tencent's cloud weapons also said that they allowed Deepseek models to access through their services.
Last month, Deepseek launched a free AI assistant, which said the assistant to use a small part of the existing service cost.
Within a few days, Apps surpass our competitor Chatgpt in Apple's App Store downloadTrigger global technology stocks.
The company earlier mentioned the global AI community’s concern in a December research paper, which said training for DeepSeek-V3 requires less than $6 million in computing power from NVIDIA’s H800 chip, and Compared with billions of dollars such as technology giants, metapore and Microsoft.
However, according to recent reports, the estimated actual cost is higher.
“$1.6 billion to build a computing cluster”
“DeepSeek's total server capital expenditure may have reached $ 1.6 billion, and about 50,000 NVIDIA Hopper GPUs have been established, including the operating cost of the calculators of $ 944 million.” South China MorningCited the semi -analysis of American research companies.
But analysts also praised the young startup's innovative software.
Morgan Stanley analysts Shawn Kim and Duan Liu wrote in the report on Sunday: “Although we can debate the actual GPU cost of Deepseek's actual GPU, it is undeniable that the factors are its output.” “This is a change of the game to change the game Rules, support AI applications and agents based on it. “
Analysts added: “This model is smaller, and with other innovations, the cost of training and operation is greatly reduced.”
It is not surprising that Beijing welcomes its success and transforms the startup of the Eastern City Hangzhou and the company's founder Liang Wenfeng into a popular cultural celebrity.
Microsoft and Amazon's cloud services have also begun to provide DEEPSEEK models, but several countries such as Italy and the Netherlands have blocked or are investigating the privacy of AI applications that are investigating DeepSeek.
- Jim Pollard's additional input and editor's Reuters
Please refer to:
Containing China with our AI model can be hard: Analysts
Yuan established “War Rooms” to learn DeepSeek AI model – Fortune
DeepSeek Breakthrough or Theft? US investigates “AI data breach”
Technical selling expands to Japan, because Deepseek will focus on the cost of AI
China's DeepSeek rock US tech giant's “AI breakthrough”
UAE's SoftBank joins Openai's $50 billion data center agreement
Bytedance's plan to have an AI infrastructure plan of $20 billion this year.
BIDEN limits the access of AI chips for American companies and their allies
US rules restrict investment in Chinese chips, quantum and AI
China says the United States' prohibition of investment is not good for the AI industry.
A way of “deceiving” a “deception” in China is a failure: a special envoy
