Save AI chip

10 minutes, 17 seconds Read

Under the wave of AI, some AI chip companies broke out of the siege, and some AI chip companies fell to the bottom. But now, after the tide, the life of AI chip enterprises represented by Cambrian is not very prosperous. After wave after wave of capital inflow, AI chips seem like a pool of muddy water.
Many AI chip companies are facing the problem of survival.
Chinese AI chip enterprises have passed their growth period
In the past two years, China’s AI chips have been very popular. AI chips set off wave after wave of investment. Last year, there were 109 investment events in the AI chip industry, with an investment amount of 39.636 billion yuan. Among them, at least 8 cases of single financing amounted to more than 1 billion yuan, and the maximum amount of single financing amounted to 5.35 billion yuan. In 2021, four AI chip manufacturers, including Muxi IC, Xingyun Zhilian, and Moore Threads, all received billions of yuan of financing.
Under the pressure of capital, AI chip companies have grown rapidly. Since the AI chip boom broke out in 2016, traditional chip manufacturers, algorithm companies, Internet giants and other companies have poured in, and hundreds of companies have devoted themselves to core manufacturing. In 2017, there were only 1110 AI chip enterprises in China, and by 2021, the number of enterprises had soared to 13492.
At the same time, AI chip companies also launched various types of AI chip products, including the Cambrian third generation cloud AI chip Siyuan 370; Journey 5 of Horizon vehicle intelligent computing platform; Kunlun core 2; Ali Pingtou Brother contains light 800; Suiyuan “Shensi” 2.5 cloud AI inference chip; Hanbo Semiconductor AI inference chip SV100, etc.
But as the tide recedes, will AI enterprises survive well?
When Cambrian, China’s first AI chip R&D manufacturer, landed on the Science and Technology Innovation Board in 2020, its share price soared to 295 yuan per share, and its total market value once exceeded 100 billion yuan. However, recently, Cambrian was frequently reduced by shareholders, and the core executive Liang Jun left, and the stock has dropped to 67 yuan/share.
In addition, the Cambrian has lost money for five consecutive years, or nearly 3 billion yuan. The revenue cannot cover the R&D investment. For the Cambrian pre-loss performance in 2021, Guojin Securities Review Company “the loss is difficult to improve, and the cost is higher than expected”. Guojin Securities believes that in terms of research and development costs, the company is expected to invest 1.044 billion yuan to 1.276 billion yuan in 2021, with a year-on-year increase of 35.9% to 66.1%, significantly higher than its previous expectation of 950 million yuan.
Things are changing and stars are disappearing. The era of AI chip 1.0 has become a thing of the past. The prologue of AI chip 2.0, which tends to be calm, is slowly beginning.
Can AI chip companies survive next year?
fierce competition
The market for AI chips is huge. According to relevant data, the market size of China’s AI core industry will reach 400 billion yuan in 2025, including 174 billion yuan for basic layer chips and related technologies. Such a high potential market means that AI chip companies are not the only ones in the market.
In recent years, many enterprises have made cores across the border and have targeted AI chips. For example, Baidu began to adopt FPGA self-developed AI chips in 2010 and released Kunlun Core 2 in 2021; Alibaba set up “Pingtouge Semiconductor Co., Ltd.” to enter the AI chip market in 2018, and launched the first AI chip “Light 800” in 2019; Tencent will build AI chips on its own in 2020; Bytes were also exposed from the self-developed cloud AI chip. In addition, there are also many AI unicorns in machine vision and voice circuits that have also begun to focus on the chip field, and intend to put their accumulation in AI algorithms and hardware together.
The big cake of AI chips has long been targeted by many enterprises.
R&D level
Although a large number of companies rush into the AI chip circuit, the research of AI chips is not so simple. The correct architecture depends on the understanding of AI. Some chip experts have strong chip design ability, but they do not have a deep understanding of AI computing or application characteristics; Some AI algorithm scientists lack basic knowledge. From the laboratory to the real scene, the chip needs to face a lot of variables. It should not only achieve low power consumption, high performance and price, but also well deal with various problems in the scene.
The high demand for research and development has made the AI chip industry suffer from losses. In his financial report, Cambrian once pointed out that the losses were mainly caused by further promoting the “cloud edge” industrial layout, expanding product lines, increasing R&D investment and talent introduction, and said that smart chip R&D required a lot of expenditure, and there would be continuous losses and potential risks in the future, especially the inability to ensure profits in the next few years.
Obvious ecological demand
But the development of AI chips requires not only wishful thinking, but also the cooperation of chip ecology. In addition to meeting customer requirements, the chip itself also needs the ecological cooperation of software, tool chain and solutions.
When facing different scenarios, the utilization and compatibility of AI chips need to be improved, and it is difficult for heterogeneous devices based on different AI chips to cooperate. However, it is difficult for enterprises to develop from soft to hard, while it is relatively easy to develop from hard to soft. Because the learning cycle of software is relatively fast, and the cycle of chip is based on years. Once the person who makes hardware has a suitable software team, he can iterate quickly, but it takes a long learning cycle from software to hardware.
For AI chip companies with many start-ups, a big challenge is that start-ups have to find out the direction of the market, and even the way of product promotion and customer needs. They need to make chips while enabling customers to use a certain product, which requires ecological support including software, tool chain, solutions, etc. This is a very challenging aspect for chip companies.
Difficulty in mass production
There are two key data for mass production, the first is 18 months, and the second is one million. 18 months means that the R&D cycle of an AI chip is generally about 18 months. After an AI chip product comes out, it may take N iterations to gain a large market share. One million chips means that the design and development cost of a chip is high, and the sales volume needs to reach the level of one million chips to reach the break-even point.
Both of these data put forward great requirements for AI chip companies. The chip industry itself is an industry with high investment. The R&D cost is very high. It may cost tens of millions of yuan for each chip, and the failure rate of the chip is very high. AI special chips are generally customized for specific application needs. Once taped, the functions cannot be changed, so there is a need for quantitative assurance, which is a great challenge for some chip companies that are still in the early stage of development.
After the chip is successfully streamed, it is also necessary to consider the issue of shipment volume. The sales requirement of millions of chips also raises questions about whether some AI chip companies can achieve this goal.
At present, most driverless companies are more willing to buy NVIDIA’s general GPU chip. Although the price is expensive and the power consumption is high, the performance is more stable. Without market validation, domestic AI chips still have a long way to go before being widely commercialized.
Save AI chip
“AI chip start-ups need to continue to move forward, transform the leading technological advantages into commercial advantages, and then support technology research and development in turn. When everyone else dies, you will be successful if you live.” said an AI chip investor.
Under the dual pressure of revenue scale and loss, how to survive has become an important proposition for AI chip companies.
Looking back at China’s AI enterprises, the four dragons of AI are looking for new ways out.
Shangtang Technology has positioned itself as an “AI factory”. In order to support the continuous operation of the whole “factory”, Shangtang Technology has invested about 5 billion yuan to build a supercomputing center and open source core algorithm. During the IPO period, 60% of the raised funds were invested in research and development, including expanding AIDC computing power, strengthening artificial intelligence chip design, self-developed existing chip solutions, improving model-related capabilities and further developing products.
Cloud is running from technology to artificial intelligence that provides efficient human-computer cooperative operating system and industrial solutions. From the original idea of AI chip making a cut, it has become a comprehensive AI solution for smart finance, smart governance, smart travel and smart business.
In the selection of AI track, Ito Technology has shifted from medical image analysis to AI chip+computing power manufacturer. The raised capital of 7.505 billion yuan is mainly used for five projects, including the new generation AI IP and high-performance SoC chip project, the edge computing system project based on visual reasoning, and to supplement working capital.
Kuangsi Technology is mainly using the Internet of Things as the carrier of artificial intelligence technology landing, developing innovative AIoT software and hardware integration solutions, which is a software and hardware integration product system combining AI, software and hardware.
The commercial landing of AI companies has not yet found a viable business model, and the adjustment and selection of AI track is the difficulty that every start-up AI company needs to overcome.
On the other hand, AI enterprises want to build their own “moat”, either by their own products or by having real core algorithms and technical capabilities in the field of segmentation.
AI chips should learn to develop comprehensively and combine in different fields. The combination here does not mean that companies in other fields make cores across borders. AI chips are difficult to become an independent product, and the landing of AI application scenarios is the real test.
In the face of the strong impact of giants, the start-up AI chip companies want to compete with it, need to dig into the vertical field rather than general products, and open the blockade of giants by relying on the differentiation of their own products and technologies. Only by accumulating users and data in the vertical field and combining the advantages of technology and algorithms can we have the opportunity to become a subverter in the vertical field.
In the field of segmentation, integration is an innovative way. Take the currently scarce auto AI chip market as an example, the current auto AI market is either AI chip company making auto AI chips, and has launched Huashan series, Huawei Shengteng series, Horizon Journey series, and Cambrian MLU series chips. Or the MCU chip company is making the vehicle regulation MCU. Similar companies include Fudan Microelectronics, Jiefa Technology, Saiteng Microelectronics, Shanghai Aircore, Microelectronics, Jihai Semiconductor, Mega Innovation, Microelectronics, etc. However, the cooperation between domestic auto AI chip companies and domestic auto gauge MCU companies is not deep, resulting in the separation of AI and MCU development on auto chips.
Therefore, AI landing in the sub-field scenario is the way to break the situation for AI chip companies.
It’s summer to stay through winter
In the chip market, which type of enterprise has a larger foam?
Yang Lei, partner of Northern Lights Venture Capital, has observed that for excellent companies, many semiconductor companies will increase their net profit by 50% to 100% every year. These are good companies. From the perspective of today’s ecological environment, whether in terms of IP, design services, supply chain management, etc., the design of the chip industry has become easier and easier. There is a slogan in the industry called “There is no hard chip in the world”.
Under the storm, the number of AI chip companies has soared, but not all AI chip companies can survive to the end.
There are indeed many people who can make chips, but the answer is not sure whether the chips made can become a chip company and go on the way of chips for a long time.

Similar Posts