Beyond the Hype: The Untold Story of China’s First AI Bubble Burst

In the bustling lexicon of China’s tech industry, few terms have captured the national imagination quite like “AI 四小龙” (AI Sì Xiǎo Lóng). For an American audience, the literal translation, “The Four Little Dragons of AI,” might not immediately resonate. But think of it as the 21st-century, tech-infused equivalent of the “Four Asian Tigers”—a term that evoked an era of explosive economic growth and ambition in the late 20th century. These four companies—SenseTime (商汤科技), Megvii (旷视科技), Yitu (依图科技), and CloudWalk (云从科技)—were more than just startups; they were symbols of China’s technological ascendance, standard-bearers in the global race for artificial intelligence supremacy.

Between 2016 and 2020, the atmosphere was electric. Fueled by a national strategic push and a seemingly endless torrent of venture capital, these firms, all specializing in computer vision (CV), became darlings of the investment world.¹ They were celebrated in state media, courted by global investors, and armed with billions of dollars to build what many believed would be an insurmountable lead in a foundational technology of the future.³ Their technology was everywhere: unlocking smartphones, enabling “scan-your-face-to-pay” systems, and powering the vast networks of smart city surveillance cameras across the country.

Yet, as quickly as they rose, their luster began to fade. The past few years have seen a dramatic reversal of fortune. The once-unthinkable headlines became commonplace: plummeting stock prices, torturous and ultimately failed IPOs, and balance sheets bleeding billions of dollars in losses. The dragons, once breathing fire, seemed to be running out of air.

This raises a crucial question: How did these celebrated national champions, once seen as invincible, fall from grace so spectacularly? What went wrong? The answer is not a simple story of a market downturn or a single misstep. It is a complex tale of a fundamentally flawed business model, a culture that prized academic brilliance over commercial reality, and a painful collision with the unforgiving logic of public markets and geopolitical headwinds. And as a new generation of AI companies focused on large language models—the so-called “AI Six Little Dragons”—now rises to take their place, their story serves as a critical, cautionary tale.⁵ Are the lessons from the original dragons’ fall being learned, or is China’s tech ecosystem poised to repeat its past mistakes?

Part I: The Gold Rush – A Tsunami of Capital and Hype

To understand the fall of the Four Little Dragons, one must first appreciate the dizzying heights from which they fell. Their ascent was propelled by a venture capital gold rush of historic proportions, a frenzy ignited by the confluence of technological breakthroughs, government ambition, and investor euphoria. In the years following Google DeepMind’s AlphaGo victory in 2016, AI was no longer an academic curiosity; it became a cornerstone of national strategy in China, and investors were desperate to back the country’s champions.¹

No company embodied this boom better than SenseTime. Founded in 2014 by a team from the Chinese University of Hong Kong, its journey was a masterclass in leveraging technical prowess to attract capital. After its facial recognition algorithm surpassed Facebook’s accuracy, investors took notice.⁶ The money flowed in waves, each larger than the last. In 2017, it closed a $410 million B-round, a global record for an AI company at the time.⁷ A year later, it secured a $600 million C-round led by e-commerce giant Alibaba, followed by a staggering $1 billion D-round from SoftBank’s Vision Fund.⁷ By the time it prepared for its IPO, SenseTime had completed 12 funding rounds, raising a colossal $5.2 billion and attracting over 50 of the world’s most prominent investment firms.⁴ Its valuation soared to an eye-watering $13 billion before it even hit the public market.⁶

While SenseTime was the undisputed heavyweight, the other dragons were hardly starving. Megvii, known for its Face++ platform, raised a $460 million C-round.⁸ Yitu Technology, another academic powerhouse, secured a $200 million C+ round in 2018, with backing from prestigious names like Sequoia Capital and Jack Ma’s Yunfeng Capital.⁸

Then there was CloudWalk, a company with a unique and powerful advantage: the backing of the state. Born from a research team at the state-run Chinese Academy of Sciences, CloudWalk was positioned as the “国家队” (guójiāduì), or the “national team”.⁹ This wasn’t just a catchy moniker; it was a core part of its identity and business strategy. Its investor list was a who’s who of state-backed entities: the China Internet Investment Fund, Shanghai Guosheng, and even the Industrial and Commercial Bank of China (ICBC).⁹ By 2020, it had amassed over 5.3 billion RMB (about $750 million) in funding, almost entirely from domestic, state-linked sources.⁹ This “national team” pedigree gave CloudWalk unparalleled access and trust, particularly for sensitive government and financial sector contracts where data security is paramount.⁹

This tidal wave of capital created a powerful, self-reinforcing feedback loop. The process was deceptively simple yet incredibly effective. First, a company like Yitu would win a prestigious global competition, such as the facial recognition challenges hosted by the US National Institute of Standards and Technology (NIST).¹⁰ This technical validation would be trumpeted as proof of world-leading capabilities. VCs, eager to invest in a strategically vital sector and back a national champion, would then pour in capital at a record-breaking valuation. This record-breaking funding round would, in itself, become a major news story, generating immense media hype. This hype reinforced the company’s image as a “winner,” which in turn helped it attract the best talent and justify an even higher valuation in the next funding round. It was a cycle of hype fueling funding, and funding fueling more hype, that inflated valuations far beyond the reality of the underlying businesses. Investors weren’t just buying a stake in a company; they were buying a piece of a national narrative of technological destiny.

Company	Founded	Core Technology	Peak Private Valuation (USD)	Pre-IPO Funding (USD)	Key Investors	IPO Status
SenseTime (商汤科技)	2014	Computer Vision, Facial Recognition	~$13 Billion	~$5.2 Billion	Alibaba, SoftBank, IDG Capital, Qualcomm	Listed in Hong Kong (Dec 2021)
Megvii (旷视科技)	2011	Computer Vision, Face++ Platform, IoT	~$4 Billion	>$1 Billion	Alibaba (Ant Group), Bank of China Group Investment	Withdrew STAR Market IPO (Nov 2024)
CloudWalk (云从科技)	2015	Computer Vision, Human-Machine Collaboration	~$3 Billion	~$750 Million	“National Team” Funds (CIIT, ICBC, etc.)	Listed on STAR Market (May 2022)
Yitu (依图科技)	2013	Computer Vision, AI Chips, Healthcare AI	~$2.5 Billion	>$600 Million	Sequoia Capital, Hillhouse, Yunfeng Capital	Withdrew STAR Market IPO (Jul 2021)

Data compiled from sources.⁴

Part II: The Cracks in the Foundation – A Fatally Flawed Business Model

For all the billions raised and the breathless headlines generated, the Four Little Dragons were built on a foundation of sand. The cooldown was not a sudden event but the inevitable collapse of a business model that was fundamentally broken from the start. As one Chinese media outlet aptly put it, their story was one of “technological singularity, commercial collapse”.¹¹ It was a failure of business design, not of technical brilliance. The core of the problem lay in a toxic combination of who their customers were, how they spent their money, and a culture that misunderstood the difference between an algorithm and a product.

The “To G” (To-Government) Trap

The primary customers for the Four Little Dragons were not consumers or a diverse array of private businesses, but rather government bodies and large state-owned enterprises. This is known in China as the “To G” (to-government) market. Their technology was the backbone of massive public sector projects: “smart cities” that promised to optimize traffic and services, public security systems for surveillance, and facial recognition gates at airports and train stations.⁹ On the surface, these seemed like lucrative, stable contracts. In reality, this reliance on the government was a trap with three fatal flaws.

First, the projects were defined by high customization and a lack of scalability. Each city’s surveillance system or each bank’s security protocol had unique requirements. This meant that every new contract required a significant new investment in bespoke development and on-site implementation.¹¹ Unlike a software-as-a-service (SaaS) product that can be sold to thousands of customers with minimal changes, the dragons’ solutions could not be easily replicated. This kept their marginal costs stubbornly high and prevented them from ever achieving the economies of scale that define successful tech companies. They were, in essence, running a high-tech but low-margin consulting and integration business.

Second, government clients are notoriously slow to pay, leading to cripplingly long payment cycles and a mountain of bad debt. As the companies took on more and more large-scale projects, their accounts receivable ballooned.¹⁴ This created a severe cash flow crunch. They were “revenue rich” on paper but “cash poor” in reality, constantly struggling to pay for their massive operational and R&D expenses while waiting for checks that might not arrive for months or even years.¹¹

Finally, the “To G” model created a vacuum of user feedback. In many cases, the government agency that commissioned a project was not the day-to-day user of the technology. This disconnect made it incredibly difficult for the companies to get the kind of direct, real-world feedback necessary to iterate and improve their products.¹¹ They were building what the government asked for, not necessarily what the market needed, trapping them in a cycle of delivering one-off projects rather than building a living, evolving product ecosystem.

The Unquenchable “Money Burn”

The flawed “To G” model was compounded by an almost pathological rate of spending. The dragons burned through their venture capital at a breathtaking pace, leading to staggering and persistent losses. In 2023, SenseTime reported a net loss of 6.44 billion RMB (about $900 million).¹⁵ Megvii’s prospectus revealed a cumulative loss of roughly 3.36 billion RMB in just three years.¹⁶ Yitu was also hemorrhaging cash before it abandoned its IPO plans.¹⁴

The most alarming metric was the ratio of research and development spending to revenue. For both CloudWalk and SenseTime, R&D expenses were not just a large portion of revenue; they exceeded 100% of revenue, reaching 119% and 109.6%, respectively.¹¹ This was not an aggressive investment in growth; it was a structurally unsustainable business where every dollar earned cost more than a dollar in R&D alone, to say nothing of sales and administrative costs.

This was exacerbated by razor-thin gross margins. Because their “solutions” often involved selling a large amount of third-party hardware (cameras, servers) alongside their software, their profitability was crushed. Megvii’s gross margin plummeted to just 33.11% in 2020, a figure more akin to a hardware reseller than a cutting-edge software firm.¹⁶ The dream of high-margin AI software collided with the low-margin reality of government contracting.

A Culture of “Technology First, Product Second”

The financial unsustainability was a symptom of a deeper, cultural problem. The Four Little Dragons were founded by brilliant academics and researchers, and they built their companies in their own image.¹¹ This created a culture that fetishized technical excellence and academic achievement over the grittier, less glamorous work of product engineering and commercialization.

Their focus was on “algorithm accuracy and academic output”.¹¹ They celebrated winning international competitions and filing thousands of patents.⁶ While impressive, this came at a cost. A huge portion of their R&D budget was “consumed on high salaries for personnel, rather than on producing replicable product capabilities”.¹¹ They were hiring PhDs, not product managers.

The result was a paradox that defined their existence: they had “technology leading but products absent” (技术领先但产品缺位). They had many orders from government clients but few real customers in a scalable, repeatable sense. They had ample financing on their balance sheets but were constantly starved of actual cash flow.¹¹

Ultimately, the dragons fell victim to a fundamental mismatch between the investment model they adopted and the business reality they inhabited. They were built using the Silicon Valley venture capital playbook of the 2010s: “blitzscaling,” burning massive amounts of cash to capture a market, achieving network effects, and worrying about monetization later.¹⁸ This model works for consumer internet or SaaS businesses with high margins and scalable products. But the dragons applied this playbook to a low-margin, project-based, government services world where none of those conditions applied. There were no network effects to be had in building a custom surveillance system for one city. The massive cash burn didn’t lead to market dominance or a clear path to profitability; it just led to bigger losses. The VCs were using the wrong map for the territory, and the companies followed them right off a financial cliff. As an executive from Yitu later admitted, using the internet “burn money” playbook for a B2B or B2G business is simply “not workable” (走不通的).¹⁹

Part III: The Great Reckoning – Public Markets and Geopolitical Headwinds

After years of living in the insulated, optimistic bubble of the private markets, the dragons were forced to face a harsh reality. The great reckoning came from two directions: the skeptical scrutiny of public market investors and regulators, and the escalating geopolitical tensions between the United States and China. These forces didn’t create the cracks in their foundation, but they exposed them for all to see and accelerated the inevitable collapse.

The IPO Gauntlet

For venture-backed companies, an Initial Public Offering (IPO) is the holy grail—an exit for early investors and a vital infusion of new capital. For the Four Little Dragons, the path to an IPO became a brutal gauntlet that most would not survive.

SenseTime’s journey to the Hong Kong Stock Exchange was the most dramatic. In December 2021, just days before its planned listing, the U.S. Treasury Department placed the company on an investment blacklist, forcing it to postpone the IPO.²⁰ It scrambled to restructure the offering, replacing American investors with a syndicate of Chinese state-backed funds, and finally managed to list at the end of the month.²² The debut was initially a spectacular success, with its stock price surging and its market capitalization briefly soaring past HK$325 billion.⁷ But the victory was fleeting. As the lock-up period for early investors expired and the company’s staggering losses continued to mount, the market’s sentiment turned viciously. Within seven months, SenseTime’s stock had crashed, losing over 200 billion HKD in value and falling into the dreaded territory of a “仙股” (xiāngǔ), a “penny stock”.²⁴

The sagas of Megvii and Yitu were even more painful. Both companies embarked on torturous, multi-year quests to go public, first trying their luck in Hong Kong before pivoting to Shanghai’s tech-focused STAR Market.¹⁴ Both ultimately failed, withdrawing their applications after repeated and lengthy delays.¹⁴ The official reasons often cited were technicalities, such as their financial statements having “expired,” requiring them to refile updated numbers.¹⁴ But these were symptoms of a deeper malaise. Public market regulators were clearly uncomfortable with their astronomical losses, questionable paths to profitability, opaque business dealings with government entities, and rising concerns over data security.²⁶ The private market hype simply could not withstand the rigorous scrutiny of a public listing.

Only CloudWalk, the “national team” member, managed a relatively smoother listing on the STAR market in May 2022.³⁰ Yet even its success has been qualified, with its stock performance being volatile and the company still struggling to turn a profit, though its 2023 financial report did show some signs of improvement in cash flow and a shift towards higher-margin business.³⁰

The Weight of the “Entity List”

Compounding their business model and IPO woes was a significant geopolitical headwind. All four of the dragons were eventually placed on the U.S. Department of Commerce’s “Entity List”.¹³ This designation, which severely restricts a company’s ability to purchase American technology and components, was primarily justified by allegations that their facial recognition technology was being used in the widespread surveillance of ethnic minorities, particularly Uyghurs, in China’s Xinjiang region.²¹

Publicly, the companies issued strong denials, calling the accusations baseless and arguing that technology should not be politicized.²¹ They often claimed the direct business impact would be minimal.²⁰ However, the sanctions were a significant blow. They created uncertainty in their supply chains, particularly for the high-end GPUs from American firms like Nvidia that are essential for training complex AI models. More importantly, the blacklisting and the human rights allegations tarnished their international reputations and made their IPO processes vastly more complicated, scaring away international investors and triggering intense questioning from regulators.²¹

It is crucial to understand the role the sanctions played. They were not the root cause of the dragons’ decline; the fatal flaws in their business models existed long before the Entity List. However, the sanctions acted as a powerful and decisive catalyst. They exposed and dramatically exacerbated the pre-existing fragility. For companies built on a model that required a constant infusion of cash to survive, the sanctions made raising that capital from international sources significantly harder and more expensive. For companies reliant on cutting-edge technology, the sanctions threatened their access to the best hardware. This dual pressure on capital and technology squeezed their already-thin margins and precarious cash flow. The sanctions didn’t start the fire, but they effectively cut off the water supply, turning a chronic illness into an acute crisis.

Part IV: A New Chapter? Pivots, Survivors, and the Great AI Shift

The story of the Four Little Dragons is not over. While their collective star has fallen, the companies themselves are fighting to survive, adapt, and reinvent themselves in a radically different landscape. Their struggles, and the broader technological shifts in the AI industry, are shaping a new chapter for Chinese AI—one marked by strategic pivots, a search for sustainable business models, and a worrying sense of déjà vu.

The Survivors’ Strategic Pivots

The ground has shifted beneath the feet of the entire AI industry. The narrative is no longer dominated by computer vision but by the transformative power of generative AI and large language models (LLMs), sparked by the global phenomenon of ChatGPT.¹⁵ This paradigm shift has forced the original dragons to adapt or face obsolescence. Their responses have split them into two distinct camps.

SenseTime and CloudWalk have chosen to go all-in on the new wave. Leveraging the one truly valuable asset they built during the boom years—massive data centers filled with tens of thousands of GPUs—they are aggressively pivoting to become large model companies.⁶ SenseTime is now marketing its AI Data Center (AIDC) infrastructure, offering both its own “SenseNova” large model and the raw computing power as a service to other companies. This new generative AI business is showing explosive growth, accounting for a significant portion of its recent revenue.⁶ Similarly, CloudWalk has launched its “Congrong” (从容) large model and is focusing on creating industry-specific applications for finance, transportation, and government, rebranding itself as one of China’s new “Five Tigers of Large Models” (大模型五虎).³⁰

In contrast, Yitu and Megvii are executing a strategic retreat. Having been burned by the “growth-at-all-costs” mindset and their failed IPOs, they are now seeking defensible niches with clearer paths to profitability. Yitu, after a period of silence, has re-emerged with a new focus on the security and smart city sectors it knows well, but with a radically different business model.¹⁹ Instead of acting as a systems integrator on massive, low-margin government projects, it is now selling integrated hardware-and-software products—like AI-powered servers—on a “cash on delivery” basis. This model is designed to ensure high gross margins and avoid the cash flow trap that plagued its past.¹⁹ Megvii has also pivoted away from the nebulous concept of smart cities toward the more tangible and commercially viable world of “supply chain IoT,” developing AI-powered robotics and logistics solutions for warehouses and factories.³⁷

The New Hype Cycle and the “Capital Winter”

As the original dragons regroup, a new class of AI startups has emerged to capture the spotlight. Dubbed the “AI Six Little Dragons,” these companies—including names like Zhipu AI, Moonshot AI, and MiniMax—are all focused on building China’s answer to OpenAI.⁵ Yet, their rise is taking place in a very different environment, one often described as a “资本寒冬” (zīběn hándōng), or “capital winter,” where venture funding is far scarcer and more discerning than it was during the last boom.¹⁸

There is a palpable sense of déjà vu. These new LLM companies are also burning through cash at an astonishing rate with highly uncertain paths to profitability, raising the critical question of whether they are simply repeating the mistakes of their predecessors.⁴¹ However, there is one crucial difference in this new hype cycle. The primary investors are no longer just traditional VCs. Instead, China’s established tech giants, Alibaba and Tencent, have become the new kingmakers.⁵ They are not just providing cash; they are providing the single most valuable and scarce resource in the AI world today: computing power. In deals like Alibaba’s recent major investment in Moonshot AI, a significant portion of the investment comes in the form of credits to use Alibaba’s cloud computing services.⁵ This creates a symbiotic relationship where the giants secure a strategic stake in the future of AI while ensuring the next generation of startups is built on their cloud platforms.

Conclusion: A “Failure Textbook” for the Future?

The dramatic cooldown of China’s Four Little Dragons is a powerful cautionary tale, a story not of technological failure but of commercial hubris. They fell from grace because they were built on a foundation of unsustainable hype and a fundamentally flawed business model. Their reliance on the unscalable, low-margin, cash-flow-draining “To G” market was their original sin. This was compounded by a culture born in academia that prioritized R&D breakthroughs over product-market fit, and a venture capital playbook that was dangerously mismatched with their business reality. The geopolitical sanctions from the United States did not create these problems, but they acted as a harsh catalyst, exposing every weakness and forcing a reckoning that was long overdue.

One Chinese technology publication aptly described the saga of the Four Little Dragons as a “failure textbook” (失败的经营教科书) for the entire industry.¹¹ It is a case study filled with invaluable lessons on the importance of scalable products, disciplined spending, and positive cash flow.

The critical question now is whether the new generation of Chinese AI companies—and their powerful backers—have read this textbook. The signs are mixed. The strategic pivot by companies like Yitu toward a more sustainable, product-focused, high-margin business model suggests that some have learned the hard lessons of commercial discipline. However, the broader LLM space shows familiar, worrying signs of a hype-driven, cash-burning frenzy, where staggering valuations are once again being built on technological promise rather than proven business models.⁴¹

The story of China’s AI ambitions is far from over. The relentless cycle of innovation, ambition, and capital continues to turn. But the market is now older, arguably wiser, and certainly more cautious. The fall of the first dragons has forced a necessary and painful reckoning, stripping away the naive optimism of the last decade. The future of Chinese AI will belong not to the companies that can raise the most money or publish the most papers, but to those that have finally mastered the difficult lessons from this textbook and learned how to build a real, sustainable business.

Works cited

1. 中国AI崛起：大量资金涌入，应用层面追上美国 – 21财经, accessed June 30, 2025, https://m.21jingji.com/article/20171013/herald/4d21fc7203a5e3b8dc439b85d86b64bb.html
2. 商汤科技、云从科技持续亏损“AI四小龙”光环褪色 – 中国工业新闻网, accessed June 30, 2025, https://www.cinn.cn/p/403176.html
3. 中国AI投资飙升三年后规模超7800亿 – 大公网, accessed June 30, 2025, http://www.takungpao.com/231106/2025/0408/1075866.html
4. 刚刚！商汤科技成功上市，成AI 领域全球最大IPO – 物联网智库, accessed June 30, 2025, https://www.iot101.com/mobile/news/1476.html
5. 瞄準中國人工智慧新秀阿里巴巴騰訊投資「AI六小龍」, accessed June 30, 2025, https://aict.nkust.edu.tw/digitrans/?p=6712
6. 商汤十年：69亿股股权归属尘埃落定，公司控制权生变？合计募资超 …, accessed June 30, 2025, https://www.stcn.com/article/detail/1376520.html
7. 商汤科技- 维基百科，自由的百科全书, accessed June 30, 2025, https://zh.wikipedia.org/zh-cn/%E5%95%86%E6%B1%A4%E7%A7%91%E6%8A%80
8. 依图科技获2亿美元C＋轮融资AI独角兽竞争激烈 – 中国财富网, accessed June 30, 2025, https://m.cfbond.com/zclb/detail/20180613/1000200000018311528867522684941971_1.html
9. 云从科技已完成18亿元最新融资-移动支付网, accessed June 30, 2025, https://m.mpaypass.com.cn/news/202005/14100938.html
10. 关于我们 – 依图科技, accessed June 30, 2025, https://www.yitutech.com/cn/about-us
11. 技术领先≠商业成功:AI大模型公司会重演AI四小龙的失败剧本吗?, accessed June 30, 2025, https://www.iot101.com/news/9894.html
12. 技术领先≠商业成功:AI大模型公司会重演AI四小龙的失败剧本吗? – 物联网智库, accessed June 30, 2025, https://www.iot101.com/mobile/news/9894.html
13. 美国制裁下，中国人工智能强国路还能走得通？, accessed June 30, 2025, https://www.voachinese.com/a/impact-of-us-sanctions-on-chinese-ai-companies-20200828/5562829.html
14. 依图科技科创板IPO“弃考” “AI明星团”上市阻滞人工智能热度降温- 21 …, accessed June 30, 2025, http://www.21jingji.com/article/20210708/36b3161225df9aa5b44d61fdda1144d3.html
15. 从CV“四小龙”到大模型“四龙五虎” AI混战进行时 – 东方财富, accessed June 30, 2025, https://wap.eastmoney.com/a/202407203135791836.html
16. 旷视科技再度冲击IPO，商业模式如何破冰？ | IPO见闻 – 华尔街见闻, accessed June 30, 2025, https://wallstreetcn.com/articles/3639749
17. 依图科技- 维基百科，自由的百科全书, accessed June 30, 2025, https://zh.wikipedia.org/zh-cn/%E4%BE%9D%E5%9B%BE%E7%A7%91%E6%8A%80
18. “资本寒冬”背后：市场资金发生了何种变化未来如何破局？ – 东方财富, accessed June 30, 2025, https://wap.eastmoney.com/a/202312252943294671.html
19. 对话“AI四小龙”依图科技：回归聚焦智能城市与安防赛道，AI大模型将 …, accessed June 30, 2025, https://www.yitutech.com/cn/news/%E5%AF%B9%E8%AF%9D%E2%80%9Cai%E5%9B%9B%E5%B0%8F%E9%BE%99%E2%80%9D%E4%BE%9D%E5%9B%BE%E7%A7%91%E6%8A%80%EF%BC%9A%E5%9B%9E%E5%BD%92%E8%81%9A%E7%84%A6%E6%99%BA%E8%83%BD%E5%9F%8E%E5%B8%82%E4%B8%8E%E5%AE%89%E9%98%B2%E8%B5%9B%E9%81%93%EF%BC%8Cai%E5%A4%A7%E6%A8%A1%E5%9E%8B%E5%B0%86%E5%8A%A0%E9%80%9F%E5%AE%89%E9%98%B2%E6%95%B0%E6%99%BA%E5%8C%96
20. 商汤之后，旷视、依图、云从等8家高科技被美国列入投资黑名单, accessed June 30, 2025, http://finance.cnr.cn/2014jingji/djbd/20211217/t20211217_525690278.shtml
21. 被美国纳入“军工复合体”黑名单后，商汤科技延迟香港上市, accessed June 30, 2025, https://www.voachinese.com/a/China-SenseTime-postpones-767-MLN-Hong-Kong-IPO-after-US-ban-20211213/6352224.html
22. AI独角兽商汤赴港上市中共天网帮凶角色惹议 – 美国之音, accessed June 30, 2025, https://www.voachinese.com/a/china-s-ai-giant-sensetime-rescued-by-state-backed-funds-has-relaunched-its-ipo-in-hong-kong-20211228/6372534.html
23. 商汤科技上市三天股价翻倍：经历大风大浪，AI价值如何重估？ – 华尔街见闻, accessed June 30, 2025, https://wallstreetcn.com/articles/3648819
24. 半年蒸发2000亿，商汤科技“亏不起”, accessed June 30, 2025, https://m.yicai.com/news/101479334.html
25. 3年亏损243亿？“AI四小龙”一哥商汤科技如何撑起120亿美金估值？ | IPO见闻, accessed June 30, 2025, https://wallstreetcn.com/articles/3639208
26. 资本宠儿旷视科技上市任重道远屡碰用户数据隐私红线 – 中国网财经, accessed June 30, 2025, http://m.gxfin.com/article/finance/zq/xgdt/2020-11-02/5421935.html
27. 提交注册3年未果，AI独角兽旷视科技撤回科创板IPO-观察者网, accessed June 30, 2025, https://www.guancha.cn/economy/2024_12_02_757375.shtml
28. 旷视科技撤回科创板IPO – 证券时报, accessed June 30, 2025, https://www.stcn.com/article/detail/1433292.html
29. 巨亏72亿、终止IPO，依图科技之后医疗AI故事该怎么继续？ – 21世纪经济报道, accessed June 30, 2025, https://m.21jingji.com/article/20210715/herald/4a3c4751fe0f87edacb4c5499e1875f5_zaker.html
30. 云从科技2023年泛AI业务增长超20倍毛利率大幅提升 – 证券时报, accessed June 30, 2025, https://www.stcn.com/article/detail/1179815.html
31. 公司公告_云从科技：2024年年度报告新浪财经, accessed June 30, 2025, https://vip.stock.finance.sina.com.cn/corp/view/vCB_AllBulletinDetail.php?stockid=688327&id=11074241
32. 声明：关于商汤科技被列入美国“实体清单”-新闻中心 – SenseTime, accessed June 30, 2025, https://www.sensetime.com/cn/news-detail/794?categoryId=72
33. 云从科技：从容大模型 – 828企业服务平台- 人民网, accessed June 30, 2025, http://828.people.com.cn/GB/458218/458221/index.html?article_id=24
34. 云从科技签约国家网安基地大模型训推一体化项目 – 证券时报, accessed June 30, 2025, https://www.stcn.com/article/detail/2285433.html
35. 云从科技, accessed June 30, 2025, https://www.cloudwalk.com/
36. 云从科技亮相鲲鹏昇腾开发者大会2025 发布从容大模型智用一体机 – 证券时报, accessed June 30, 2025, https://www.stcn.com/article/detail/1847116.html
37. 旷视魔方智能分析盒, accessed June 30, 2025, https://www.megvii.com/products/hardware/Intelligent_analysis_Unit
38. 旷视科技撤回科创板IPO – 证券时报, accessed June 30, 2025, https://www.stcn.com/article/detail/1433292.html?u_atoken=1cb7dda60e4d6247cb50b4153f4179ad&u_asig=ffbfd
39. 中国科技初创企业遭遇“资本寒冬” – FT中文网, accessed June 30, 2025, https://www1.ftchinese.com/story/001085799
40. 寒冬已去？创新药企业重磅BD不断、资本市场高歌猛进！专家：进入新发展阶段 – 东方财富, accessed June 30, 2025, https://finance.eastmoney.com/a/202506263441070834.html
41. 亏损1.6亿，云从科技被AI大模型狠狠“整顿”了- OFweek通信网, accessed June 30, 2025, https://tele.ofweek.com/2024-05/ART-8320512-8500-30634833.html