China’s artificial intelligence large language model landscape is being reshaped at an unprecedented pace. As four leading companies \u2014 DeepSeek, Moonshot AI, Zhipu AI (2513.HK), and MiniMax (100.HK) \u2014 successively make waves with technological breakthroughs and capital market moves, the four founders behind them \u2014 Liang Wenfeng, Yang Zhilin, Tang Jie, and Yan Junjie \u2014 have been dubbed by the market as “China’s AI Big Four.” By current estimates, the combined valuation and market capitalization of the enterprises these four leaders control has surpassed 1.1 trillion yuan (approximately $162.6 billion), as a vast homegrown Chinese AI business empire takes shape.<\/p>\n
The four founders range in age from 34 to 49, and without exception possess top-tier academic credentials and deep technical expertise. Though they all stand at the forefront of China’s AI wave, the paths they have chosen diverge dramatically: one is dedicated to breaking computing power monopolies through architectural innovation, another is betting on long-context windows and open-source ecosystems, a third is immersed in foundational artificial general intelligence (AGI) research, and the fourth is pushing hard to expand the commercial frontiers of multimodal applications.<\/p>\n
Liang Wenfeng: From Quant King to AI Price Disruptor<\/p>\n
The 41-year-old Liang Wenfeng is the founder of DeepSeek. A former top scorer on the gaokao college entrance exam from Wuchuan in Zhanjiang, he began applying artificial intelligence to quantitative trading during his master’s studies at Zhejiang University, later founding the renowned quantitative investment firm High-Flyer Quant. It is precisely this background that has given DeepSeek an innate “cost reduction and efficiency improvement” DNA. Liang introduced the statistical thinking of quantitative trading into large model training, optimizing data sampling and distributed training frameworks to dramatically reduce reliance on brute-force computing power stacking.<\/p>\n
According to publicly available data, DeepSeek-V3’s training cost was just $5.576 million, yet it delivers performance comparable to GPT-4o, with inference costs slashed to roughly one-tenth of the industry average. This extreme cost-performance strategy has rapidly won over the market, with more than 200,000 enterprises globally now using DeepSeek models, and the company’s annualized revenue reportedly reaching $400 million to $500 million.<\/p>\n
In capital operations, Liang has demonstrated equally formidable control. DeepSeek completed a 51 billion yuan (approximately $7.5 billion) first-round funding in June this year, bringing in giants such as Tencent (0700.HK) and CATL (300750.SZ), lifting its valuation to between 350 billion and 400 billion yuan (approximately $51.7 billion to $59.1 billion). Despite introducing outside capital, Liang still firmly holds 84.29% equity and nearly 100% voting rights through a special shareholding structure. According to Bloomberg data, Liang Wenfeng’s personal net worth stands at approximately 256.7 billion yuan (approximately $37.9 billion), ranking him 60th on the global rich list. Recent market chatter suggests DeepSeek has initiated IPO preparations, planning to list on China’s STAR Market; if the listing succeeds, its business empire could expand further.<\/p>\n
Yang Zhilin: The 34-Year-Old Tech Icon and the Kimi K3 Shockwave<\/p>\n
At 34, Yang Zhilin is the youngest of the four and the figure who has recently caused the most violent tremors across global capital markets. A former top science scorer on the gaokao from Shantou, he graduated from Tsinghua University’s Department of Computer Science before pursuing a PhD at Carnegie Mellon University, where he contributed to heavyweight architectures including Transformer-XL and XLNet. After stints at Google Brain and Meta AI, Yang returned to China in 2023 to found Moonshot AI, establishing long-context technology as the core moat for the Kimi model series.<\/p>\n
Moonshot AI’s latest release, the Kimi K3 model, boasts 2.8 trillion parameters and a 1-million-token context window, scoring 1,679 points on the front-end coding arena Arena.ai to surpass Claude Fable 5 and claim the global number one spot. Although Yang has candidly acknowledged that K3’s overall performance still trails the strongest closed-source models, its open-source strategy and top-tier coding capability have nonetheless triggered a massive shockwave across the global AI industry.<\/p>\n
According to reports from multiple media outlets including the China Times, within 72 hours of Kimi K3’s release, the US-listed AI sector shed approximately $470 billion in combined market value, with the Philadelphia Semiconductor Index plunging 10% in a single week. Shares of Zhipu AI and MiniMax suffered two consecutive days of heavy selling: Zhipu’s cumulative two-day decline exceeded 42%, with its market cap briefly falling below HK$500 billion (approximately $63.8 billion); MiniMax’s decline over the same period approached 25%. It was not until July 21, as market sentiment recovered, that Zhipu surged 36.89% in a single day and MiniMax rebounded 15.17% \u2014 a vivid illustration of how hypersensitive secondary markets have become to shifts in model capabilities.<\/p>\n
K3’s explosive popularity even triggered an unforeseen computing power crisis. Moonshot AI issued an emergency statement announcing that user request volumes had far exceeded the cluster’s carrying capacity, forcing the suspension of new consumer subscriptions. Yang’s monetization efforts have also accelerated accordingly; while K3’s API pricing is already the most expensive in China, it still holds roughly a 40% cost advantage over GPT-5.6 Sol. Kimi’s enterprise clients have now surpassed 120,000, with paying users in high-value sectors such as finance and law accounting for over 60%. Moonshot AI’s latest valuation has reached approximately $31.5 billion (~213.4 billion yuan), and with his 51.83% stake, Yang Zhilin’s personal fortune exceeds 110 billion yuan (approximately $16.3 billion).<\/p>\n
Tang Jie and Yan Junjie: Alternative Paths in AGI and Multimodal AI<\/p>\n
Compared to the seismic waves Liang Wenfeng and Yang Zhilin have stirred in secondary markets, the 49-year-old Zhipu AI founder Tang Jie and 37-year-old MiniMax founder Yan Junjie appear more measured \u2014 but their strength is no less formidable.<\/p>\n
Beyond his role as an entrepreneur, Tang Jie is also a professor in Tsinghua University’s Department of Computer Science. Amid the clamor for commercial monetization, Tang issued an internal letter announcing a “Mo Gao Plan,” pledging to concentrate over 10 billion yuan on foundational AGI research over the next two years, spanning long-horizon tasks, autonomous agents, and extreme safety governance \u2014 explicitly not pursuing short-term application revenue. Zhipu’s GLM-5.2 model, open-sourced under an MIT license, has already attracted over 3 million developers globally. Although its stock price experienced violent swings following the Kimi K3 shock, Zhipu reached $1 billion in annual recurring revenue (ARR) in just five months, with its market capitalization still standing at 489.8 billion yuan (approximately $72.4 billion).<\/p>\n
Yan Junjie, who holds a PhD from the Institute of Automation at the Chinese Academy of Sciences, has steered MiniMax to focus on Mixture of Experts (MoE) architecture and multimodal technology. Its product Hailuo AI specializes in text-to-video generation, supporting real-time output at 4K resolution and 60 frames per second. As of July this year, multimodal product revenue had surpassed 40% of the total. Yan even pledged in an all-staff letter to forgo his salary until the company achieves its AGI goal, and committed 4% of his personal shares to incentivize the team. Based on MiniMax’s latest market cap of 67 billion yuan (approximately $9.9 billion), Yan’s 21.46% stake is worth approximately 14.4 billion yuan (approximately $2.1 billion).<\/p>\n
Impact Analysis and Future Outlook<\/p>\n
The establishment of the “China AI Big Four” framework signals that China’s large model industry has moved from isolated breakthroughs to a phase of collective emergence. Zhang Xiaorong, director of the Deep Tech Research Institute, notes that whether it is Moonshot AI or DeepSeek, once they enter public markets they will undoubtedly attract strong capital inflows \u2014 but the flip side of scarcity is enormous volatility. The roller-coaster ride in Zhipu’s stock price over just a few days has already proven that as long as the “release model \u2014 top the leaderboards \u2014 stock price swings” pattern persists, violent capital market fluctuations will not cease.<\/p>\n
Wang Chao, founder of Wenyuan Institute, believes the release of Kimi K3 may have shortened the generational gap between Chinese models and overseas frontier models to within three months, compared to a gap of 12 to 18 months just a year ago. However, the shelf life of such advantages is often extremely short. Shortly after K3’s release, Alibaba unveiled a preview of Qwen3.8 Max, and Elon Musk teased that a new Grok model is imminent. As Yang Zhilin often says, “A day in AI is a year in the human world” \u2014 until the next disruptive model is released, no one can rest securely on the throne.<\/p>\n
It is foreseeable that as DeepSeek and Moonshot AI successively signal IPO plans, the capital market poker game for China’s homegrown large models will gain yet another heavyweight variable. For the four leaders at the helm, finding the balance between technological iteration, commercial monetization, and capital market expectations will be the key to determining who has the last laugh in this trillion-yuan AI contest.<\/p>\n","protected":false},"excerpt":{"rendered":"China’s artificial intelligence large language model landscape is being reshaped at an unprecedented pace. As four leading companies…\n","protected":false},"author":2,"featured_media":114223,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[4],"tags":[6744,3013,6010,45062,58395,42601,56012,9755,930,9603,53363,52817,15838,9756],"class_list":["post-114222","post","type-post","status-publish","format-standard","has-post-thumbnail","category-agi","tag-agi","tag-artificial-general-intelligence","tag-deepseek","tag-glm-5-2","tag-hailuo-ai","tag-high-flyer-quant","tag-kimi-k3","tag-liang-wenfeng","tag-minimax","tag-moonshot-ai","tag-tang-jie","tag-yan-junjie","tag-yang-zhilin","tag-zhipu-ai"],"_links":{"self":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/114222","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/comments?post=114222"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/114222\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media\/114223"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media?parent=114222"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/categories?post=114222"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/tags?post=114222"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}