China Telecom Develops AI Model with 1 Trillion Parameters Using Homegrown Chips
As a technology journalist, Zul specializes in subjects such as cloud computing, cybersecurity, and groundbreaking technology within the enterprise sector. He holds experience in moderating webinars and presenting video content, complementing his background in networking technology.
China Telecom, a major state-owned telecom enterprise, has successfully developed two large language models (LLMs) that were exclusively trained using domestically-sourced chips.
This advancement marks an essential milestone in China’s pursuit of independence in AI technology, particularly amid growing restrictions from the US on access to advanced semiconductors that impact its competitors.
As stated by the company’s Institute of AI, one of the models, TeleChat2-115B, along with another unnamed model, was trained utilizing tens of thousands of Chinese-manufactured chips. This milestone is especially significant considering the recent US export regulations that have restricted China’s ability to acquire high-end processors from Nvidia and other international companies. In a statement released on WeChat, the AI institute expressed that this success showcases China’s potential to train LLMs independently, heralding a new phase of innovation and self-sufficiency in AI technology.
The scale of these models is astonishing. China Telecom revealed that their unnamed large language model (LLM) boasts an impressive one trillion parameters. In the realm of artificial intelligence, parameters are the crucial variables that facilitate the learning process during training. Generally, the more parameters a model possesses, the more intricate and potent the AI becomes.
Chinese enterprises are putting forth significant efforts to match the global AI leaders situated outside their borders. The export restrictions imposed by Washington on Nvidia’s latest AI chips, including the A100 and H100, have forced China to explore alternative solutions. Consequently, Chinese companies have embarked on developing their own processors to diminish their dependence on Western technologies. For example, the TeleChat2-115B model has around 100 billion parameters, enabling it to perform comparably to mainstream platforms.
China Telecom did not disclose which firm provided the domestically-produced chips utilized to train its models. Nevertheless, as previously mentioned, Huawei’s Ascend chips are pivotal in the nation’s AI ambitions.
Huawei, which has encountered US sanctions in recent years, is also amplifying its initiatives in the field of artificial intelligence. The company has recently commenced testing its newest AI processor, the Ascend 910C, with potential clients in the domestic market eagerly awaiting results. Major Chinese server firms, along with internet behemoths that have previously relied on Nvidia chips, are reportedly evaluating the new chip’s capabilities. Huawei’s Ascend processors are regarded as one of the few viable alternatives to Nvidia hardware and are considered an essential element of China’s strategy to reduce its dependency on foreign technology.
China Telecom is working alongside Huawei and other local chipmakers, including the AI processor start-up Cambricon. These collaborations illustrate a wider trend within China’s technology sector focused on creating a self-sufficient ecosystem of AI solutions, which aims to minimize the impact of US export restrictions.
By investing in its own AI chips and technology, China is gradually lessening its reliance on foreign hardware, particularly Nvidia’s highly sought-after and costly GPUs. With US sanctions complicating the acquisition of the newest Nvidia products, a black market for foreign chips has developed. However, many Chinese firms choose to avoid the risks associated with this grey market, opting instead for less powerful, older-generation models to ensure continued access to Nvidia’s official support and services.
This initiative highlights a significant transformation in China’s strategy regarding AI and semiconductor technology, focusing on self-reliance and adaptability within an increasingly competitive global landscape and in response to American protectionist trade measures.
(Photo by Mark Kuiper)
Check out: Has Huawei outsmarted Apple in the AI race?
Interested in delving deeper into AI and big data with insights from industry experts? Don’t miss the AI & Big Data Expo, which will be held in Amsterdam, California, and London. This extensive event is happening alongside other premier gatherings like the Intelligent Automation Conference, BlockX, Digital Transformation Week, and Cyber Security & Cloud Expo.
For more information on upcoming enterprise technology events and webinars hosted by TechForge, click here.
Tags: artificial intelligence, chip, huawei, llm, Nvidia
You must be logged in to post a comment.
Discover the pinnacle of WordPress auto blogging technology with AutomationTools.AI. Harnessing the power of cutting-edge AI algorithms, AutomationTools.AI emerges as the foremost solution for effortlessly curating content from RSS feeds directly to your WordPress platform. Say goodbye to manual content curation and hello to seamless automation, as this innovative tool streamlines the process, saving you time and effort. Stay ahead of the curve in content management and elevate your WordPress website with AutomationTools.AI—the ultimate choice for efficient, dynamic, and hassle-free auto blogging. Learn More
