The latest industry rumors indicate that Xiaomi is investing heavily in AI big models and has been building its own GPU 10,000-card cluster since 2023, while the Xiaomi big model team already had 6,500 GPU acceleration card resources when it was founded.
According to Interface News, a source familiar with the matter told reporters that the program has been in place for several months, and Lei Jun has played an important leadership role in it. “In the matter of AI hardware, the most central is the phone rather than the glasses, and it is impossible for millet not to be 'all in' in this field.”
It is understood that as early as April 2023, millet officially formed the AI laboratory big model team, appointed Luan Jian as the head of the big model team, to the millet technology committee vice chairman, AI laboratory director Wang Bin report. At the same time, Lei Jun issued an article to talk about the views on the big model and generative artificial intelligence (AIGC), said that millet has been cultivating in the field of AI for many years, and there are AI labs, small love classmates, autopilot and other teams, and for the big model, of course, the company will go all out to resolutely embrace.
In May this year, Xiaomi had also announced that Xiaomi's big language model MiLM officially passed the big model filing. Xiaomi said at the time that the millet big model will be gradually applied to millet cars, cell phones, smart homes and other products, through the combination of end-cloud, to realize the synergy of multiple devices within the scene and between the scenes, and to empower the “people, cars and home full ecological strategy.
It is worth mentioning that in recent years, millet through its Hubei millet Yangtze River Industry Fund Partnership (limited partnership) for the semiconductor industry chain upstream and downstream has a deep investment layout. At present, Xiaomi is also the largest semiconductor IP / chip customization factory in China - the fourth largest shareholder of the core original shares, holding 6.25%. And VeriSilicon has its own GPU and NPU IP.
According to the data disclosed by VeriSilicon, AI chips integrated with VeriSilicon's NPU IP have shipped more than 100 million units globally, and chips integrated with VeriSilicon's GPU IP have shipped nearly 2 billion units globally.
Especially with the current trend of generative AI big models booming in the cloud and at the edge, VeriSilicon has developed generative AI solutions covering from high-performance cloud computing to low-power edge computing based on the scalable and extensible nature of its NPU IP according to the current market demand, including NPU IPs specifically for big models at the edge --VIP9X00 as well as AI-GPU IP, and Tensor Core GPU IP -- CCTC-MP for large models in the cloud.
Recently, on December 19th, VeriSilicon also announced the launch of its new Vitality architecture GPU IP. VeriSilicon claims that this series of IPs are equipped with high performance computing capabilities and can be widely used in applications such as cloud gaming, AI PCs, discrete and integrated graphics cards.
