AI chip's rivers and lakes in the 'China Core' call and the trend of artificial intelligence sweeping across all industries, the fire do not want.
The traditional chip giants Nvidia, Intel, Qualcomm, Internet giants Google, Alibaba, Baidu, and entrepreneurial leaders such as the Cambrian, Shenzhen Science and Technology, Horizon, etc., have all raised a lot of high-profile investment from the capital level. The amount of financing starts at 10 million US dollars.
In addition, there are also the dark horses such as Bit Continent and Kennan Soochi that have emerged directly from the application market. From a single bitcoin mining, they ran out of mining machine chips and began to extend to other application areas.
In a market where big gamers compete, is there a chance for start-ups? Where are the opportunities? How to seize the opportunity? Yibang Power talks with the rather low-key Jinyun Technology. A member of the team, the background of the team is not expected to be gorgeous. The team's soul character Lu Yongqing is the British Royal Academy of Engineering, the Institute of Electrical and Electronics Engineers (IEEE), the British Academy of Sciences academician, Imperial College professor, in 1992 began to study custom computing chips, Participated in the development of the world’s first high-level compilation software for custom chips, Handel-C.
Many of its CEO NIU Yu-yu and the team’s core R&D staff are students of Academician Lu Yongqing. Niu Yu-Yu said that Jinyun Technology has almost evacuated the talents of the major AI laboratories of the Imperial College of Technology.
At present, Jinyun Technology has a laboratory research team in the UK, and has algorithms, hardware, products, operations and other teams in Shenzhen and Shanghai.
Different from other high-flying artificial intelligence chips, Jiyun Technology's initial technological capabilities are aimed at providing a customized end-to-end AI application solution for IoT terminals and vertical domains.
'Our chip architecture is divided into two layers, the upper layer is customized data, the bottom is a unified chip architecture. Customization can ensure that the chip performance is strong enough, generalization can ensure that the cost is low enough. The technology accumulated in our laboratory is through the compiler Deploying the requirements of different algorithms to the chip, automating the personalization with a compiler, both solving the customization and guaranteeing high performance while solving the common problem. ' This technology simulates the human design process through a compiler Automatic generation of customized chips, which greatly reduces the threshold and cost of chip design.
Yang Ge, a founding partner of Xingyi Capital, who invested in Jinyun Technologies in Pre-A, also mentioned that chips are a highly-invested industry, which can cost millions to tens of millions, and require continuous iterative upgrades. The entire artificial intelligence application market is still very elementary, and applications in many fields are at an experimental stage. The market is extremely fragmented, and traditional chips are too expensive for the market. It is too expensive to develop chip costs for market segments. Cloud technology can achieve a great deal of customization and versatility, and can provide a better market price.
The opportunities of LoT market start-ups are more to grow with the market. The current market is characterized by a very wide spread, very decentralized, and large scale, which gives us opportunities for growth. In some areas such as smart phones, the market is very large. , The demand is relatively single, it is suitable for giants to fight in the inside. Some areas have requirements for AI chips, need low power consumption, high enough performance, can be embedded in the front, while the market has not been normalized like a mobile phone, giants Input and output in these areas are not directly proportional.
This leads to the creation of a market for GAP.
Many start-up companies come out of the lab, have technical strength, and the team is small enough and fast enough to respond quickly to such market demands. Compared with the giants, there is a speed and cost advantage, under the precondition of technically maintaining advantages. , there is a prerequisite for the growth of such companies. '
Niu Haoyu and Yang Ge also saw this opportunity.
It is understood that Jinyun Technology's game is aimed at a vertical segmentation field, customizing a complete set of end-to-end software and hardware solutions, by customizing the artificial intelligence algorithm model into modules, and configuring them into unified bottom-layer chips to solve the actual industry. Problems, contributed to the chip's tape.
'The chip itself has no value, only the solution runs on the chip, to solve the customer's actual problems can provide value. The AI market is not mature enough to provide only the chip can solve the problem, is still in its infancy. Take out a chip, It's not hard, but startups need to think about how to land quickly and industrialize. The development of the chip industry is following the industrialization. The chip is the result of industrial development, not the reason. ' Niu Yu-Yu said.
In addition to COMAC, a major customer of China Aerospace, Jinyun Technology has focused on smart cities, industrial monitoring, smart manufacturing, and smart finance. Niu Haoyu told Yibang Power Network that Jinyun Technology selected the industry. There are three criteria:
The first is that the market is large enough to reach 100 billion yuan; the second is deep enough in the market; the third is close enough to commercialization.
It seems that smart cities, finance, etc. are almost all industries targeted by AI companies, a single market, but in fact, Jinyun chooses to be a long-tailed, deeply-spread market in this large market. 'The deeper the market, the more we can accumulate. The more customization programs, the more slowly we have accumulated one by one scenario. Iterate over our own chip architecture according to the application scenario. This is the value we can accumulate.
Different from the AI company that sells face recognition in the security field, Jinyun Technology emphasizes providing other capabilities than face recognition. It uses a whole set of more customized solutions to move customers and solve needs. This is at the current stage of AI development. , It is very easy to get the customer's approval.
It is not the ultimate goal of Jinyun Technology to tackle the market and solve the problem in one area. As a chip company, Niu Yuyu's expectation is to personally accumulate a number of industry scenarios, develop algorithms, and deeply understand the needs of each subdivided industry. , Form a set of practical and feasible solutions, so as to continuously optimize the AI open platform of Jiyun Technology, so that users on the platform can develop and design solutions for subdivided industries at a low cost, and finally the solution will be landed on the bottom chip, and the scale will be maximized. Reduce the cost of the chip.
Jinyun AI development platform mainly focuses on the field of artificial intelligence chips, and can provide full automation support from data labeling, hardware compilation to board testing. It can be automated only by providing user data labels, and provides customized AI for specific areas. Front-end products and solutions, without the need for the underlying hardware expertise, greatly reduce the user's use threshold.
'Our development is bound to be more and more open cooperation. At this stage we as a start-up company to call everyone to use our development platform, this is difficult. On the contrary we can attract everyone to do it together.
What value can we provide to the market? Stable chip, low enough price, mature application case, let developer can make money. If an AI chip company can't do these points, it won't be quite big. The opportunity for growth. ' Niu Yu-yu said.
For the upcoming LoT big market, all chip entrepreneurs are expected to be lucky. However, many fields are still using the AI function such as data calculations in the cloud, just because of the price of AI chips. Too high, it is not realistic to popularize in terminals. Individual wealthy buyers only place AI capabilities on the terminal in partial test scenarios. In the future, let end devices have an AI chip. Is this a certain trend?
Niu Yu-yu thinks that the trend of terminal + cloud is basically a pattern of the future. Some algorithm training, a large number of confidential data can only be entered into the cloud, and many need not to store data, private data, data that need fast feedback is more suitable Put on the terminal processing.
Take the mall flow management as an example. If all camera data is transmitted to the cloud for processing, one camera generates half a T of data a day. Many of these data are not related to commercial applications and are transmitted to the cloud. The cost is high, and the cost of back-end server storage is also high. Especially for chain stores, cloud-based deployment cannot support it.
On the contrary, if each camera can handle basic face extraction, tracking and other calculations, and only transmit the extracted data to the cloud aggregate analysis, it can greatly reduce costs and increase efficiency. Here, it needs a sufficiently low cost. The chip is embedded in the camera. This is also the development direction of Jinyun Technology.
The characteristic of the chip is that the higher the cost is, the lower the cost will be. Firstly, it is used in the military industry, and the big companies finally go to personal consumption. Retailing is actually pulling the chip directly to the level equivalent to the C-side consumption. If a camera program is done in hundreds of pieces, Personal consumption is more competitive.
But when this time can come, Niu Yu-yu can not give accurate answers, more driven by market demand.
At present, the AI chip buyer is mainly based on government agencies and large-scale enterprises. The real LoT era is still in the darkness before dawn.
'The entire AI chip ecosystem has just started. Great companies such as Nvidia and Google are also in the exploration stage. However, the ecology that will ultimately win will form an industry monopoly, such as intel in the PC era and ARM in the mobile Internet era.
Who can successfully land before the monopoly? All AI chip entrepreneurs want to be themselves.