The entrepreneurial history of China's chip companies

In the past two decades of development, China's capital market has invested in a wide variety of projects, but they all use the same core story - the vast Chinese market.

Hillhouse Capital founder Zhang Lei secured $10 million in initial funding from the Yale University Foundation thanks to his heavy investment in China. Shong Xiaoge, who returned to China to work in venture capital, has not exited a single project in seven years and still enjoys support from IDG. All of these successes are based on the vast and alluring Chinese market, capable of creating industries serving hundreds of millions of people and generating hundreds of billions or even trillions of dollars in annual revenue. This market has also captured the imagination of investors, driving investment in this market.

However, the semiconductor industry has never enjoyed such treatment. Chip semiconductor companies use today's investments to seize tomorrow's market. Chip companies need continuous and stable orders from brand manufacturers to maintain a healthy input-output rhythm. Otherwise, high capital and R&D investments will cause many companies to "die" today. However, the foundation of the Chinese "market" is not strong. For a long time, Chinese electronic products were collectively known as Huaqiangbei, known for their low prices and counterfeit products. Products can only be sold in large quantities, lacking the stable generational updates that drive stable upgrades and iterations of the supply chain system.

However, a country without its own technological reserves is ultimately in vain. In the government's industrial planning, ensuring that "national integrated circuit" products could meet the majority of domestic market demand has been a goal that has remained unfulfilled and unresolved for over two decades. Waves of scientific research and industrial talent have dedicated themselves to the task of producing a universal chip that is market-recognized for China. They began amidst national attention, and after the audience's enthusiasm faded, they were left to independently navigate the gap between scientific research achievements and market demand.

China's chip and semiconductor industry seems to have been on the defensive since its inception, repeatedly rehearsing and growing within the scenario of "what if a giant blocks us?" The secret to a company's long-term survival should be where the market is and what customers want. It's time to ask: Even if these companies, packaged by capital and national ambition, possess technology comparable to Intel and Nvidia, can they still secure the same customer support that sustains Intel and Nvidia?

Technology blockade

In 2015, before founding Cambricon, Chen Tianshi was still a researcher at the Institute of Computing Technology of the Chinese Academy of Sciences. In response to a question from The Paper, "Which country has a bigger gap in computing, China or the West?" he wrote that the main gap is the West's blockade of key technologies and equipment, as well as the market barriers set up by Western pioneers.

His words came true. Eight years later, Cambricon, a chip design company founded by Chen Tianshi, was listed on the "Entity List" of the US Department of Commerce along with 36 other Chinese chip semiconductor companies. Xingge Technology, an automotive chip design company under Cambricon, was also included in the "Entity List" even though its chips were still in the research stage and had not yet been mass-produced.

This means that companies and organizations listed on the Entity List who wish to purchase any products controlled by the U.S. Export Administration Regulations (EAR) or containing U.S. technology must apply for a license from the relevant U.S. agency and await government review. Failure to do so could result in sanctions. The Entity List creates entry barriers for normal commercial transactions, much like the QR code scanning required to enter and exit public spaces during the epidemic prevention and control period, and the need to report travel to and from high-risk areas to the local authorities.

The issue of erecting technological barriers has become a familiar one. Every researcher at the Institute of Computing Technology of the Chinese Academy of Sciences has heard the story of the "Glass House": In the 1990s, major Western countries signed the Wassenaar Arrangement, using patents and lawsuits to impose a comprehensive blockade on China's cutting-edge technology, including computers. At the time, when China purchased high-performance computers from the United States for civilian use, it was necessary to build a glass wall around the computer room because the US Department of Commerce would require technical experts to ensure that the machines were not used for military purposes. That glass wall, known as the "Glass House," is a crisp, transparent name that reflects the deep sense of shame a country feels under a technological blockade.

This experience also shaped the ethos of the Chinese Academy of Sciences—when conducting research, one must think about contributing something to the nation. CAS researchers and PIs frequently hear their directors, advisors, and project leaders use similar phrases, speaking of "Guozhong" (the abbreviation for the State Key Laboratory of Computer Architecture) and "Youqing" (the National Science Fund for Excellent Young Scholars). The nation is a concept with a strong presence in this narrative. The Institute of Computing Technology of the Chinese Academy of Sciences, located on Zhongguancun South Road, is like a national treasure chest, poised to reverse the situation at the snap of Thanos' fingers.

Cambrian founder Chen Tianshi was already an "Outstanding Young Scientist" at the Chinese Academy of Sciences before the age of 30, and according to the scientific researcher promotion system, he had a bright future. Chen Tianshi was the quintessential "other people's child": he was admitted to the USTC Junior Program at 16, graduated with a doctorate at 25, and at 31, developed the first AI chip—Cambrian, which is also the origin of Cambrian's name. He has an older brother, Chen Yunji, the second-youngest researcher in the history of the Chinese Academy of Sciences and also a member of the USTC Junior Program. They are one of the few "brother prodigies" in the history of the USTC Junior Program.

The research directions of these brothers are aligned with the needs of the national technological agenda. Chen Yunji, a student of Hu Weiwu, the team leader for the development of Loongson 1, China's first general-purpose CPU chip, is the chief architect of Loongson 3 and possesses engineering expertise and product development experience in general-purpose processor chips. Chen Tianshi excels in artificial intelligence algorithms. In 2013, the leadership of the Chinese Academy of Sciences transferred Chen Yunji from the Loongson team to research on intelligent chip processors, internally referred to as a "small national key research project" to distinguish it from the Loongson project.

China's chip semiconductor industry has been lame for too long.

China joined the WTO in 2001, and over the next two decades, it has accumulated a trade surplus of nearly $4 trillion through products like televisions, computers, and mobile phones. However, China's semiconductor industry has been consistently held hostage by foreign giants, resulting in an annual trade deficit of nearly $200 billion. In 2021, with chip supply shortages and companies scrambling to stockpile, the trade deficit widened to $280 billion.

During the long-term technological catch-up, scientific researchers in China's chip industry have always held their breath, looking for opportunities to overtake others.

In 2014, a project team led by Chen Yunji and Chen Tianshi published two papers on deep learning processors, DianNao and DaDianNao. These papers won Best Paper awards at the ASPLOS and MICRO architecture CCF Category A conferences, making a significant impact in the field of artificial intelligence. Years later, an American student recounted how their Indian professor, an Indian-American professor, struggled to pronounce DaDianNao in class while quoting from the papers.

The instruction set is the core of the computer hardware and software ecosystem. Intel and ARM control the PC and embedded ecosystems through their instruction sets. DianNaoYu is Cambricon's instruction set. Compared to traditional instruction sets, DianNaoYu can process a group of neurons with a single instruction. Thanks to its research achievements in AI chips, Cambricon has become the "top talent of the institute," garnering industry attention and discussion. Much like Loongson's success 20 years ago, Cambricon has become a new hope for the Institute of Computing Technology of the Chinese Academy of Sciences to surpass competitors in the field of AI chips.

Starting to climb

At the beginning of the century, the entire country's pursuit of "Chinese chips" reached a climax. In June 2000, the "Several Policies to Encourage the Development of the Software Industry and the Integrated Circuit Industry" mentioned that after 5 to 10 years, "national integrated circuit" products will be able to meet most of the domestic market demand and have a certain amount of exports.

After failing to persuade Liu Chuanzhi to build a chip, the leaders turned their attention elsewhere, using all available connections to recruit overseas students to return to China to start businesses. Leveraging the personal connections of Zhou Guangzhao, a founding member of the "Two Bombs, One Satellite" program, they recruited Deng Zhonghan and Zhang Hui to return to China and found Vimicro. Both were alumni of the University of California, Berkeley. The former had led the design and research of very large-scale integrated circuits at IBM, while the latter was responsible for the development of reconfigurable CPU architecture technology at Lucent Bell Labs. Li Delei, another early member of the "China Chip Project," had served as director of microprocessor design at Hitachi Semiconductor America before returning to China to found Vimicro.

Soon, they delivered results. In March 2001, Deng Zhonghan's Vimicro developed the Starlight-1 processor. In September 2002, Li Guojie's Institute of Computing Technology of the Chinese Academy of Sciences developed the Loongson processor. In December 2002, Li Delei's SMIC launched the Ark-2 processor. 2002 became the year of the chip, and headlines about China ending its "chipless" history were frequently picked up by the media and appeared in newspapers.

They also enjoyed conveniences unimaginable in that era. Registering a company in Zhongguancun took only six days, offering the opportunity to secure office space of at least 40 square meters and even a year's rent-free lease. If a company was deemed promising, it would receive nearly 80,000 yuan in the "International Student Entrepreneurship Support Fund," which did not require repayment, and could also receive government-guaranteed microloans and interest subsidies.

Furthermore, Deng Zhonghan's "Starlight 1" received 10 million yuan in support funds, and Fangzhou received 20 million yuan. However, these projects still lag far behind the more established Loongson, a general-purpose chip project from the Chinese Academy of Sciences, which received 100 million yuan in support funds and also enjoyed significant exposure. The launch event for Fangzhou 2 took place at the Great Hall of the People. The Zhongguancun Management Committee secured this opportunity for the chip after a complex application and approval process.

Despite the heated public reaction, those in the industry are well aware of the challenges and capital required to create a new chip. They avoid Intel's strengths and instead focus on niche markets. In addition to Loongson's attempts at general-purpose chips, Starlight 1 is a multimedia chip, powering cameras and mobile phones. Ark 2 is betting big on netbooks.

"Starlight No. 1" was a successful bet, capitalizing on the booming digital camera market. Nearly 1,000 chips were sold in 2002. However, "Ark No. 2" never found a use. The government offered a token purchase of a few thousand chips in 2002, but the chip failed to gain traction with domestic manufacturers. Li Delei spent all his funds and could not find a source of income.

Regarding their short-lived chip-making venture, Li Delei said, "We failed in the market," and he credited Liu Chuanzhi with being more resourceful. Li Delei was referring to the "Ni-Liu dispute," where Liu Chuanzhi's concern at the time was that the key to success wasn't whether chips could be produced, but whether they could be sold.

The situation with Loongson was not much better, with total sales of less than 100,000 units for Loongson 1 and a similar situation for Loongson 2. No matter how patriotic a PC manufacturer might be, it would be a shame to rush into using a chip that users didn't understand.

Back then, China's semiconductor industry was sparsely populated, with no incentives for exchange. Trading time for market share? Judging from the story of Shockley's invention of the transistor and the subsequent formation of "Silicon Valley" in Santa Clara, China was nearly 50 years behind. Compared to the semiconductor industry chains of South Korea, Taiwan, and Japan, China also started at least 20 years later. Trading capital for market share? The tens of millions of dollars in capital and R&D investment in China pale in comparison to the hundreds of millions or even billions of dollars invested by Silicon Valley companies.

The significance of being number one in China might be explained by Zhao Lun, then-general manager of Datang Microelectronics, at the Beijing IT Industry Development Seminar in 2004: "In terms of sales revenue, Datang is indeed number one in China, but internationally, we only rank 44th. We are still very immature." In fact, thanks to its specialized chips for mobile phone SIM cards and second-generation ID cards, Datang Microelectronics is already China's largest chip design company. However, its market size is limited.

Figure: Sales (10,000 yuan)

Industry insiders estimate that Chinese companies have no chance of success in developing chips from a market perspective for at least 10 years. However, a country without its own technological reserves will ultimately fail, so there's only one option: let the government take over, no matter the cost.

This commitment is exemplified by the Chinese Academy of Sciences' Loongson project. Since the birth of Loongson 1 in 2002, Loongson has spent nearly 20 years, supported by tens of millions in annual government subsidies, continuously developing breakthroughs and iterations. Its fourth-generation products, the Loongson 3A6000 and 3A7000, have barely caught up, with performance comparable to the current mainstream Intel 11th-generation mobile Core and AMD Zen 4 desktop CPUs. However, Loongson's revenue in 2021 was only 1.2 billion yuan.

Wanzhong Innovation Chip

Zhongguancun's enthusiasm for semiconductors soon shifted to the Internet.

In Silicon Valley, the internet boom began in the 1990s. Netscape went public in 1995, Yahoo in 1996, and Amazon in 1997. In Zhongguancun, waves of wealth creation followed one after another. Wang Zhidong's Sina and Zhang Chaoyang's Sohu brought multi-fold returns to the capital market in just a few years. From 2003 to 2004, companies like Tencent, Ctrip, Kongzhong, and Shanda went public in rapid succession. Yan Yan, then at Softbank China Venture Capital, achieved a 15-fold return on his Shanda investment in just two years.

Compared to the semiconductor industry's struggle to survive by scavenging scraps discarded by chip giants, the internet industry's sudden wealth creation is more appealing. After basking in the atmosphere of China's chip industry's breakthrough, public opinion shifted its focus to the internet's wealth creation movement: 24/7, comprehensive coverage of the industry, from financing information and IPOs to the financial freedom of founding teams and emerging trends. Employees at semiconductor companies dream of jumping to internet giants and doubling their salaries.

The next wave of enthusiasm for semiconductors would not arrive until 2014. The State Council issued the "National Integrated Circuit Industry Development Promotion Outline," stating that China's goal was to increase its self-sufficiency rate in integrated circuits to 40% by 2020 and to 70% by 2025. Local governments also launched industry guidance funds to support semiconductor companies.

The power of market-oriented operations is rising.

While the Loongson project didn't generate substantial profits, it did cultivate talent for China's semiconductor industry. Chen Yunji developed a deep understanding of the semiconductor industry within the Loongson team. Chen Tianshi's doctoral advisor, Yao Xin, also wrote his doctoral dissertation on complexity analysis of intelligent algorithms under the guidance of Li Guojie. This intergenerational transmission of academic research is crucial.

In terms of entrepreneurial projects, the Chinese Academy of Sciences has start-ups such as Cambricon and Yitu Technology, as well as flagship companies such as Inspur and Loongson.

The University of Science and Technology of China (USTC), located in Hefei, has also made a strong presence in this wave of semiconductor startups. Horizon Robotics Vice President Zhang Yongqian, Xinwang Microelectronics CEO Ding Xiaobing, Axin Semiconductor CEO Zhang Xuena, Hengshuo Semiconductor Chairman Lv Xiangdong, Xinyang Semiconductor founder Fang Shunong, Panqi Microelectronics President Li Baoqi, and Fudan Microelectronics General Manager Shi Lei are just a few examples. Even Cai Songsong, a fund manager known for his expertise in semiconductor industry investment, graduated from USTC's Junior Class.

The narrative template of education changing destiny that is prevalent in the semiconductor industry is reflected in USTC. The industry jokingly says that if USTC alumni go around, they will have all the resources for chip semiconductor design, manufacturing, and packaging.

Starting a business requires a strong sense of risk, and semiconductor manufacturing requires an even stronger sense of risk. A successful overtaking on a curve can result in a nearly 10-year lead.

Intel defined the CPU in 1971 and dominated the semiconductor industry for 40 years. Nvidia's multi-billion-dollar bet on CUDA in 1999 defined the GPU, and in 2020, its market capitalization surpassed Intel's for the first time. The market's enthusiasm for GPU chips has spurred investors to focus on GPU chips, scrambling to recruit executives from Nvidia and AMD, becoming a key talent acquisition process in this fierce competition.

Nvidia has no core R&D team in China, and AMD has only a small R&D team in Shanghai, making them key targets. Facts have proven that even the strongest corners can't withstand the constant poaching of domestic investors. Zhao Lidong, founder of Suiyuan Technology, and his entrepreneurial team came from AMD, as did Feng Xinpeng, founder of Zhaoguan Electronics.

Zhang Jianzhong, founder and CEO of Moore's Threads, a GPU chip design company, was formerly Nvidia's global vice president and general manager of China. In addition to Nvidia, its core members also include Intel, AMD, ARM and other companies.

When Zhang Wen decided to start BiRen Technology, he asked a Harvard graduate to compile a list of top chip industry talent. As for recruiting these individuals, Zhang Wen approached them one by one. Li Xinrong, former AMD Global Vice President and General Manager of AMD's China R&D Center, became Co-CEO, and Hong Zhou, head of HiSilicon's in-house GPU team and chief architect, became CTO.

Picture: Former AMD Global Vice President,

Li Xinrong, current co-CEO of BiRen Technology

Despite its relatively short development history, domestic GPU enthusiasm continues to surge. BiRen Technology set a record by raising 4.7 billion yuan in 18 months, with backers including Qiming Venture Partners, IDG Capital, and Hillhouse Capital. Moore Threads, founded just over two years ago, has raised nearly 5 billion yuan, with backers including Sequoia Capital, Tencent, GGV Capital, and ByteDance.

Riding on the opportunities created by the rise of domestic mobile phone and new energy vehicle brands, chip startups are like trending stars, attracting investment and attention. They are seen as the young dragon slayers battling giants. However, the key issue remains the same as it was twenty years ago: to develop general-purpose chips, one must not only manufacture them but also sell them.

Growth soil

The chip is a small, dark green, square piece, almost weightless when held in the palm of your hand. The circuitry beneath the package reveals a faint texture. If the color were changed to bluish-purple, it could be interpreted as the lines on the bodies of the people of Pandora's Planet that James Cameron designed in "Avatar."

But even this tiny area is as challenging as building a 50- to 60-story "microcosm" on an area measuring tens of millions or even tens of thousands of times the diameter of a human hair. This requires chip and semiconductor companies to continuously invest heavily in exchange for market orders. It's common for a project to sink billions of dollars and yet fail to make a splash. Due to this massive cash burn, it's common for startups to go bankrupt due to a single chip tapeout failure.

TSMC's development is inseparable from Apple's support. According to financial reports, Apple accounts for approximately 27% of TSMC's revenue. In reality, Apple's influence far exceeds this. Qualcomm, Apple's baseband chip supplier, is manufactured by TSMC, as are Sony, its camera system chip supplier, and Nvidia, its graphics card supplier. Apple's products significantly impact TSMC's revenue, far exceeding the 27% share. This is why Tim Cook can say no to TSMC's price increases.

An influential brand in the industry will continue to feed suppliers in the industrial chain with orders, carry out technological iterations at the same frequency, stay at the forefront of technology, and form a stable symbiotic relationship.

Take Huawei and HiSilicon, for example. Founded in 2004, HiSilicon released its first chip, the K3V1, five years later. However, due to its immature technology, it ultimately failed to reach the market. It wasn't until the release of the Kirin chip in 2014 that HiSilicon's fortunes began to change. From 2004 to 2014, HiSilicon cost Huawei over 160 billion RMB. The Kirin 980 chip alone cost HiSilicon 2 billion RMB in R&D. HiSilicon was able to sustain itself for over a decade thanks to Huawei's nearly 90% share of orders. But many chip companies lack HiSilicon's pedigree.

At the beginning of the century, Chinese consumer electronics held no place in the global market, and the Chinese industry chain lacked a prominent presence. Going back a decade or so, Chinese brands entering the international market needed to bolster their confidence by acquiring foreign brands. Lenovo, for example, expanded its international business by acquiring IBM's personal computer business, using it as brand endorsement. Not to mention Li Dongsheng, who, to facilitate TCL's international expansion, acquired Thomson's color TV business in France and Alcatel's mobile phone business. These two acquisitions nearly depleted TCL's finances and served as examples of failed international mergers and acquisitions for MBA business schools. Their actions can be simply described as throwing money away in a declining market.

Before the advent of smartphones, Chinese electronics could be summed up in one name: Huaqiangbei. Known for its low prices and counterfeit goods, they sold in high volumes, with styles and brands closely matching the market's hottest trends. Like deep-sea sharks, they encounter a school of fish, swarm in and snatch up all the goods, then sail off to try their luck elsewhere.

Even in the smartphone era, with press conferences filled with verbal attacks, Lei Jun, Richard Yu, Zhao Ming, and others have a wide variety of disparaging remarks about competitors at new product launches. Their attitude toward chips remains consistent: the latest Qualcomm Snapdragon. MediaTek always comes up short against Qualcomm's Snapdragon chips.

A similar situation is playing out in the new energy vehicle sector. Chinese NEV companies are struggling on the brink of losses, while Qualcomm is raking in the cash. In July of this year, Li Shufu's Zeekr Auto offered users a free chip upgrade to the Qualcomm 8155, a move that garnered widespread attention and widespread praise from users. This put pressure on other automakers. When a user inquired in the NIO group about a free upgrade, NIO founder William Li Bin replied, "With over 200,000 users, a free upgrade would cost over 2 billion yuan."

After the news of the United States banning the export of Nvidia A1000 chips to China came out, He Xiaopeng, chairman of Xpeng Motors, forwarded a message on WeChat Moments saying that "Nvidia A100 restrictions will severely damage China's autonomous driving," and wrote: "The bad news is that this will bring challenges to all autonomous driving cloud training. The good news is that we have already bought back the demand for the next few years in advance." He Xiaopeng's low EQ remarks indirectly illustrate that the Nvidia A1000 chip is a "social currency" worth showing off on the Internet.

Chinese semiconductor companies are often categorized as "others" in industry rankings, a euphemism for not making it into the top 10.

A glimmer of hope emerged this year, with industry chain rumors suggesting Yangtze Memory Technologies (YMTC) would integrate into Apple's supply chain, supplying flash memory chips for iPhones. In various versions of the rumor, Tim Cook, the master of balancing the two, was concerned about balancing Apple's supply chain security and was unwilling to see Korean chip companies dominate the flash memory market, so he supported Yangtze Memory Technologies. However, this ray of light was short-lived, extinguishing before it could illuminate anything. Yangtze Memory Technologies was added to the US Department of Commerce's Entity List, and while Cook sought to maintain balance, he was powerless to change anything.

Industry winter

The new wave of chip startups have not yet proven themselves in the market when they are swept into the cold winter of the semiconductor industry by the consumer electronics winter.

Affected by the new round of sanctions, the share prices of many A-share chip and semiconductor companies have further fallen. The Sci-Tech Innovation 50 Index, which focuses on semiconductor companies, has fallen 31% year-to-date, outperforming the Shanghai and Shenzhen Composite Indices. Cambrian, which listed on the Sci-Tech Innovation Board in July 2020 and once boasted a market capitalization exceeding 100 billion yuan, has seen its market capitalization evaporate by over 80 billion yuan in two years. Its financial reports show increased accounts receivable and rising inventories. The halo that once surrounded it has now become a source of external condemnation.

Cai Songsong, a fund manager who had previously been firmly bullish on the semiconductor industry, also changed direction and adjusted his positions from semiconductors to enterprise information services.

The industry is shifting from expansion to caution. Research firm IC Insights has revised its optimistic forecast for global semiconductor capital expenditures from its initial forecast of 24% growth at the beginning of the year, nearly $190.4 billion, to $181.7 billion, a growth rate of nearly 19%. This growth rate is nearly double the 35% increase in spending last year, when chip manufacturers rushed to ramp up production due to the chip shortage.

The weakness of the chip industry has made self-developed chips the most important "technical correctness" for domestic companies. Mainstream mobile phone manufacturers have to prove their value in the high-end market by developing their own chips.

Xiaomi's chip manufacturing dates back to 2014, and it officially released its first independent chip, the Surge S1, in 2017. In 2021, it also launched two more self-developed chips: the Surge C1 and the Surge P1. This year, it released its fourth chip, the Surge G1, a battery management chip. Oppo, not to be outdone, also released its second self-developed chip, the Mariana Y, at this year's INNO DAY. Vivo also has its own ISP chips, the V1 and V2. The V1 ISP chip is also primarily used for photography.

In 2017, Huawei launched the world's first artificial intelligence processor, the Kirin 970. After that, it was in high spirits and emphasized the "Da Vinci" plan to fully adopt its own chip architecture, kicking its previous partner Cambricon out of the cooperation list.

New energy vehicles are, to some extent, filling the gap in consumer electronics chip demand, but the overall market is more fragmented. Automakers are developing their own chips. NIO previously invested in Cambricon's Xingge Technology, and Xpeng Motors launched its own autonomous driving chip program last year. BYD, this year's top seller, has a subsidiary, BYD Electronics, which began developing IGBT power chips as early as 2005. BYD's automotive-grade IGBT production ranks second globally, second only to Infineon.

The rise of Chinese chip companies requires the emergence of "Huawei" and "BYD" companies across multiple industries, with sufficient market scale to support synchronized technological iteration across companies and supply chains. The rise of domestic mobile phone and auto brands has created favorable market conditions for localized chip production. This is unlike a decade ago, when a lack of support from leading brands led to a technological gap.

For chip startups, convincing established brands to use unproven chips is a significant challenge. Furthermore, in-house development is becoming a trend among industry leaders. Huawei created HiSilicon, BYD created BYD Electronics, and leading automakers are also considering developing their own chips.

Chip startups are waiting for the best opportunities of the times, but also face the most intense competition. But pessimists are right, and optimists are moving forward. In the industry opportunities created by the times, you must always fight hard to have a chance.

Latest articlesabout