On March 15th, the GTIC2019 Global AI Chip Innovation Summit, which was an AI Chip Industry Summit hosted by Zhishang, was successfully held in Shanghai. Academic authority, Professor Wei Shaojun of Tsinghua University delivered a keynote speech titled "Vision and Implementation Path of AI Chip 2.0". Walden International, a well-known venture capital firm in the global semiconductor investment industry, and investors of Northern Light Venture Capital shared the investment logic of the AI chip track. Famous AI chip startups and semiconductor giants such as Qualcomm, Huawei, Horizon Robotics, Cambricon, Bitmain gathered at the summit to discuss the topic of AI chip industry ecology.
As a research and consulting institution that have been paying close attention to AI area for a long time, iResearch Consulting Group also attended the summit. The following are the summary of the summit.
Summary of the Guests’ Sharing--Investment Opportunities and Challenges in the AI Chip Track
Professor Wei Shaojun, Director of Institute of Microelectronics, Tsinghua University
The AI chip industry enters a critical period in 2019. The industry is now facing issues such as "What development path should the AI chip industry take?", "Which are the directions of AI chip innovation?" and "How to design a competitive AI chip."
At present, most of the AI services are provided on the cloud, such as smart homes, image recognition, smart medical care, AI translation, etc. The trend from the cloud to the terminals will be inevitable in the future. According to a report released by Tractica in May 2018, the revenue of the deep learning chip market will reach $66.3 billion in 2025.
The AI chip industry is now facing two major problems. First, the algorithm is still evolving. The continuously emerging new algorithms bring challenges to the development. Second, one algorithm correspond with one application and there is no unified algorithm. A general deep learning engine with high energy efficiency is essential to AI chips.
The development of AI chips can be divided into four stages: AI Chip 0.5, 1.0, 1.5 and 2.0. In the stage of AI chip 0.5, people focused on the computing power of chips. In the stage of AI chip 1.0, they have higher requirements for power consumption and exclusive usage in the domain besides computing power. In the stage of AI chip 1.5, performance, energy efficiency and flexibility are all important. Configuration and refactoring are achieved. In the stage of AI chip 2.0, intelligence will become the mainstream.
Implementation Path of AI Chip 2.0: The development direction of energy-efficient neural network computing requires collaborative design, optimization and verification of algorithms and hardware.
Huang Qing, Managing Director of Walden International
The first task of AI chip startups is to find the application scenarios and commercial purposes of the chips. An AI chip startup can only transform a research into a business if it finds a specific purpose of the chips. For example, Google develops TPU since the company has huge demand for deep learning computing.
With the development of technologies and the increasing demand for applications, AI will be implemented on a single-chip SoC in the near future, and the market holdings will show explosive growth. In the future, companies who can closely match technologies, and products with actual applications will occupy the market.
A development direction for AI Chip Startups which has great potential is the data center processor——Now the throughput of the data center is growing and will be too big for GPU. A large scale of processors like TPU will emerge and small companies will have more opportunities since they are flexible and have innovative architecture.
Yang Lei, Managing Director of Northern Light Venture Capital
The core contradiction in China's chip industry: Excessive low-end supply and insufficient high-end supply lead to serious imbalance.
Is the valuation of AI companies too high? This question can be answered by a famous sentence quoted from Bill Gates: "We always overestimate the change that will occur in the next two years and underestimate the change that will occur in the next ten." The number of venture capitals which invest in AI chips is several dozen times higher than that of chip companies. The scarcity of quality projects has further pushed up valuations.
The chip industry is a turning point in history because of the stop of Moore's law, power wall barrier, and the complexity of application scenarios, which leads to the golden age of chip architecture innovation. For chip startups, architectural innovation and software and hardware integration bring them opportunities to compete with large companies.
When you meet the founder of the company, the key questions are: 1. Is this person a remarkable person with special experience? 2. Does his/her company have a very large market? Does he/she enters the market in an interesting way? 3. Does he/she understand the technology that he/she is talking about perfectly? Among them, the market is the starting point and technology of the product is closely related to the person's experience.
Industry pattern: the demand of China's chip market is like a rugby ball. The high and low end demand is small while the medium demand is large. The supply structure of China’s chip industry is like a pyramid. Currently, a large number of various companies are growing and competing with each other.
For further development, the chip industry in China needs to be based on real scenarios and achieve breakthroughs in the mid and high-end markets. The industry needs to avoid repeating the mistakes of the solar power industry and the LED industry.