Losing Huawei, How chip giant Cambrian hit the Science Plate

The outbreak did not close the market, and another unicorn began the process of listing. On the evening of December 28, 2019, China’s C.I.A. and Citic Securities signed an A-share listing counseling agreement that will open the process of listing the company, the official website of the Beijing Securities Regulatory Commission announced on the evening of February 28, 2019.

Wen/Gu Quanjun

Losing Huawei, How chip giant Cambrian hit the Science Plate

It is widely believed that if The Cambrian successfully landed on the Science and Technology board, it would become the “first stock of AI chips” without suspense.

The Cambrian period, which was founded in 2016, has been buzzing about listings in the past year.

In March 2019, Liu Qingfeng, chairman of the board of directors, said at the recent “Koda News AI New Product Media Tasting Conference” that the company’s investment in Cambrian and other enterprises are actively exploring the possibility of landing on the Science and Technology board.

However, in an interview with the Economic Observer, Mr. Cambrian later said, “There are no plans to become a public company, let alone an IPO, and a hundred and eighty thousand miles away from the Science and Technology Board.” “

But the change in business information that followed the Cambrian period was clearly in preparation for the IPO.

On November 29, 2019, the name of Cambrian Company was changed from “Beijing Zhongke Cambrian Technology Co., Ltd.” to “China Cambrian Science and Technology Co., Ltd.” and the type of enterprise changed from “Limited Liability Company (Natural Person Investment or Holding)” to “Other Limited Company (Unlisted)”, in order to comply with the Securities Act Relevant regulations for listing.

Tianeye data show that since the establishment of the Cambrian period, a total of five rounds of financing, the most recent is in September 2019, the investor for the Oriental Production and Finance Bank International Capital and Profit Capital, the amount of financing has not been disclosed.

Losing Huawei, How chip giant Cambrian hit the Science Plate

The highest amount of public financing is round B. In June 2018, Cambrian announced the acquisition of hundreds of millions of dollars in round B, the overall valuation after the investment reached 2.5 billion U.S. dollars, investors including the Chinese Academy of Sciences funds, Science and Technology, China Investment Ventures, Ali and so on.

In the past few rounds of financing, the national character background of the investors, Cambrian is also regarded as the AI chip “national team.”


Cambrian is a typical “technological establishment”. Chief scientist, CEO Chen Yunxuan, Chen Tianshi brothers are graduated from the China University of Science and Technology “junior class”, Ph.D. graduated from the Chinese Academy of Sciences Computing Institute, Ph.D., Chen Yunxuan’s research direction is chip, Chen Tianshi mainly do artificial intelligence, artificial intelligence plus chips, there is a later Cambrian.

In 2015, long before the cambrian establishment, the Chen brothers, who were also in the Chinese Academy of Sciences, led the development of the world’s first deep learning-specific processor prototype chip, which is also the name of the Cambrian company. At the same time, the company’s tens of millions of angel round financing also from the Chinese Academy of Sciences.

In addition to the Chen brothers, The Cambrian’s main founding team also has experience in the development and design of China’s first universal MIPS architecture CPU, “Loongson”.

Cambrian focuses on artificial intelligence chips, a concept relative to traditional chips. At present, AI chips mainly refer to GPU, FPGA, ASIC and other AI acceleration chips (Cambrian is mainly ASIC-specific integrated circuits), mainly used to solve the huge demand of artificial intelligence computing power.

Currently, the main application scenarios for AI chips are cloud computing data centers and edge computing, which includes camera IPC, autonomous driving, mobile phone Soc, and so on.

In the past few years, Cambrian has also released a range of related products. The Cambrian reputation was the 1A processor released in 2016, which it claimed was the world’s first commercial deep learning processor, and Huawei later installed the solution on its Soc.

The most recent release was November 14, 2019, when Cambrian released the Edge AI series of SI 220 (MLU220) chips and M.2 acceleration card products.

This is the First AI chip for edge-oriented intelligent computing, according to officials, the Siyuan 220 chip uses the latest generation of smart processor MLUv02, to achieve a maximum of 32TOPS (INT4) computing power, while power consumption is only 10 watts (W).

Because of its “speciality”, the special integrated circuit is a chip type with relatively large investment and relatively high commercial risk, which needs to be customized for specialized application scenarios, while Cambrian is also specialized in aI chip this kind, and does not involve CPU, GPU and other general-purpose chips.

The current product model used by Cambrian is also “authorization and finished product” two-legged walk, the former similar to ARM, the AI chip IP licensed to downstream manufacturers, such as the most well-known partner Huawei;

“Speciality” Returns and Risks

When it comes to the Cambrian era, it is Huawei that has to be mentioned.

In 2017, Huawei launched a new generation of mobile processor Kirin 970, featuring AI performance, which is equipped with NPUIP, is from the Cambrian, the following year, the Kirin 980, still chose the Cambrian cooperation, Mate10, Mate20, P20 and other flagship aircraft, are equipped with the Cambrian NPU, And the shipments of these machines are in the tens of millions.

Thanks to Huawei’s huge shipments, The Cambrian stake has also been made.

Things have changed in 2018.

In October 2018, Huawei unveiled two AI chips, the Ying Teng 910 and the Tin 310, at the All-Connect Conference, using Huawei’s self-developed AI architecture, Da Vinci, rather than the Cambrian program.

In this regard, Chen Tianshi said in an interview, Huawei’s Da Vinci plan, I sawmedia reports to know. But it’s a surprise: “If a powerful, platform-based giant like Huawei doesn’t plan to study AI chips, it just means that AI chips aren’t important enough.”

He argues that Huawei’s Launch of the Tin 310 and the Cambrian MLU100 are not competitive because the two scenarios are different – the former is mainly on the edge, while the latter is the cloud, with different peak performance.

By June 2019, Huawei’s nova5 was powered by Huawei’s latest mid-range mobile processor, the Kirin 810, the first mobile AI chip to use Huawei’s self-developed Da Vinci architecture, rather than its former partner, Cambrian.

The end-of-year flagship platform, the Kirin 990, still uses the Da Vinci architecture, and its run on a AIBenchMark reached 476% of the Kirin 980, which is also seen as a signal of Huawei’s successful departure from suppliers.

“From the shipment of the Kirin 970 mobile phone, if the license fee is $5 per tablet, more than 40 million mobile phone shipments will bring about $200 million ($1.4 billion) in revenue from cambrian shipments, according to the report. “

Huawei is seen as the biggest source of revenue for the Cambrian period.

Huawei is undoubtedly a big customer and gives the market a glimmer of doubt about the prospects for the Cambrian era. In order to dispel this perception, the startup must come up with strong technical strength.

In addition to Huawei, Cambrian also has many wealthy and powerful competitors, such as Alibaba, one of the investors in the Cambrian, which set up Pingtou Semiconductor Co., Ltd. in 2018 to integrate Zhongtian Microsystems Co., Ltd. (hereinafter referred to as Zhongtian micro) and Damo Institute’s self-research chip business.

In July 2019, Alibaba Group Vice President Qi Xiaoning announced that Pingtou brother’s first end smart chip Xuan Iron 910 released, using RISC-V architecture, using 12nm process, main frequency 2.5GHz, 7.1Co/MHZ, performance two said, but its target of the end of the cloud market, is highly integrated with the Cambrian.

In addition, Baidu, Google and other Internet giants have been high-profile agency AI chips, although the application scenario may have a focus, but The Traditional Vida, Intel, AMD and other traditional chip factories are unlikely to let go of this market.

Notably, Invista ranked first in the AIChipsetIndexTOP24 list released by compassIntelligence in 2018, with Huawei HiSilicon 12th and Cambrian 23rd.

The Market for AI chips has a very promising outlook. According to the ABI report, the cloud AI reasoning and training services application market will grow from $4.2 billion in 2019 to $10 billion in 2024, a battleground.

Losing Huawei, How chip giant Cambrian hit the Science Plate

As you can see, the top vendors are internet giants or hardware giants, and for “specialists” like the Cambrian, Avery has revealed some of the possible risks in his report.

The semiconductor industry has two manufacturing models, one is their own complete disenfranchisement of the design to the packaging of the whole process, one is the production design, to the contract ingress factory production, and the Cambrian naturally belongs to the latter, the foundry control process and production capacity, the upstream manufacturers have a strong bargaining power;

At the same time, chip design enterprises also have their own upstream EDA tools, design companies in turn on the upstream bargaining power is not high, sandwich edgy survival.

In addition, the chip industry is pursuing the scale effect, streaming costs are high, the need for huge shipments to maintain profits and costs of the screen, this order of magnitude is at least 10 million.

Huawei is the largest customer of the Cambrian period, and the impact on shipments is obvious after it is self-sustaining.

After the completion of the IPO, Cambrian will have more capital to invest in research and development, which is the Cambrian technology companies are very important, how to get rid of the impact of the departure of large customers, how to open up more markets, how to deal with technology iterations, etc. , are the problems in front of us.