In a significant development in the rapidly advancing world of artificial intelligence, Xiaomi is reportedly in the midst of establishing its own massive GPU cluster, which will undoubtedly bolster its efforts in AI model developmentAccording to recent insights from Jiemian News, as of December 26, Xiaomi aimed to create this extensive GPU system, embarking on a substantial investment to enhance its AI capabilitiesThe news highlights that Xiaomi's emerging AI development team started with a robust resource pool, comprising 6,500 GPUs already in its possession at the time of inception.

Furthermore, while the company has yet to publicly comment on this undertaking, rumors have circulated regarding key personnel moves within the organizationNotably, Luo Fuli, one of the main developers behind the DeepSeek open-source model, is set to join Xiaomi and possibly lead its AI laboratory

With an impressive background that includes a degree in computational linguistics from Peking University and experience at Alibaba's Damo Academy, Luo’s expertise will be vital as Xiaomi endeavors to carve out a greater presence in the competitive landscape of AI.

The application of AI Agents in smartphones has become increasingly mainstream, showcasing the profound impact of advancements in chip computing powerWith this burgeoning technology, mobile devices are expected to enhance human-computer interaction and intention recognition significantlyAI Agents, powered by foundational AI systems embedded within smartphones, are poised to understand user needs far more accurately, seamlessly executing tasks and evolving into highly customized personal assistants for usersThis paradigm shift represents a crucial strategic advantage for brands seeking to maintain a loyal customer base.

According to IDC forecasts, by 2025, China’s next-generation AI mobile phone market is anticipated to witness a staggering shipment of approximately 118 million units, reflecting a year-on-year increase of 59.8% and commanding a market share of 40.7%. This significant growth trajectory presents both opportunities and challenges for firms like Xiaomi, as emphasized in its third-quarter financial report for 2024, which revealed a revenue of RMB 47.5 billion from its smartphone division—a 13.9% increase compared to the previous year

The effectiveness of its strategy to upscale product offerings has begun to bear fruit; however, as the market matures into a saturation phase, the rising demand for high-end innovations highlights the pivotal role big model technologies will play in determining a brand's competitive edge.

AI technology has already seeped into almost every aspect of smartphones, from voice assistants to imaging enhancements and performance optimizationsNevertheless, the arrival of large language models signifies a transformative leap in the capabilities of mobile AI features, promising a more potent and responsive experience for usersHowever, when compared to peers such as OPPO and Vivo, Xiaomi’s promotion of its large model technology appears relatively subduedWhile it has made strides in areas such as imaging and semantic assistance, there is considerable room for growth in multi-modal interactions and cross-app connectivity.

The recruitment of Luo Fuli underscores Xiaomi's intention to harness specialized research talents aligned with its strategic direction within big models

This strategic move signals a broader intention to double down on its investments in AI, a field increasingly dominated by firms that can swiftly innovate and deploy advanced AI solutionsThe term “ten-thousand card cluster” is now described as a significant benchmark in the current round of AI model competition, kicking off a frenetic pursuit amongst technology giants for supremacyAs noted by Baidu's executive vice president, Shen Dou, at the Baidu Zhiyun Conference, the market can expect more ten-thousand card clusters to emerge shortly.

The exponential growth in demand for model training resources has prompted the creation of these expansive GPU clustersWith each passing year, the expectations for continuous reductions in model inference costs also grow, placing heightened demands on the stability and efficacy of GPU managementOn the same day, Baidu announced an upgrade to its AI heterogeneous computing platform, known as "Baihe 4.0", enabling it to deploy and manage ten-thousand card clusters—setting a new standard in the industry.

This current upheaval surrounding generative artificial intelligence can be largely attributed to a massive increase in computational power—a phenomenon somewhat akin to the adage “hard work pays off.” The industry has responded by ramping up massive power stacks, attaining significant performance leaps in large models

alefox

Consequently, the ten-thousand card cluster has been regarded as an essential component for firms aiming to penetrate the core of AI systemsHowever, even this vast computational capacity is not yet sufficient to meet burgeoning demands, leading many industry players to invest in even more substantial ten-thousand card setups to drive improved computational efficiency and performance within their models.

At a previous Yuncy Conference, Alibaba Cloud showcased its new infrastructure for the AI era, revealing that its networks had expanded to accommodate clusters comprising ten thousand cardsThis ambitious approach covers various technological facets, including chips, servers, networks, storage solutions, cooling technologies, power requirements, and data centers—all redesigned to facilitate advanced AI infrastructure tailored for future demands.

In another example of this escalating competition within the AI domain, Elon Musk announced on social media earlier this month that his AI startup, xAI, has launched a super AI training cluster called "Colossus." The construction of this complex took 122 days and involved 100,000 Nvidia H100 GPU accelerator cards

Musk further revealed plans to double this capacity in the coming months, with an additional 100,000 GPUs, half of which would utilize even more advanced H200 chips.

Following closely, Meta CEO Mark Zuckerberg had previously announced plans to acquire 350,000 Nvidia H100 GPUs, effectively multiplying Meta's computational power equivalent to what 600,000 H100 GPUs would offerWhile OpenAI has not disclosed specific numbers regarding its GPU usage, industry speculation suggests their resource pools approach ten thousand GPUsAccording to Wang Xiaochuan, CEO of Baichuan Technology, conversations during his visits to Silicon Valley indicated that OpenAI is working on a computational model designed to interconnect ten million GPUs—a vision compared to moon landing ambitions, illustrating the scale and audacity of aspirations in the AI sector.

As shared by researchers like Shen Xiangyang from Hong Kong University of Science and Technology, today, any general-purpose large model company lacking a ten-thousand card cluster might find their legitimacy questioned