New application scenarios that may emerge from the combination of AI and chips. 1. Embodied Intelligence Currently, AI exists only on screens; in the future, AI will need to enter the physical world. Such as robots, drones, and autonomous driving. The chip must be able to process vision, decision-making, and control in real time, and its power consumption must be very low. For example, Tesla's Optimus robot, as well as humanoid robot companies like Figure and 1X, are all developing their own AI chips. General-purpose chips cannot meet the real-time requirements of "thinking and doing simultaneously". If embodied smart chips mature in the next 3-5 years... We may see AI nannies, AI workers, and AI delivery drivers; this is a trillion-dollar market. 2. End-device multimodal: large models can run on mobile phones, cars, and glasses. Most large models are now located in the cloud. In the future, it will be integrated into our mobile phones, cars, and AR glasses. The chip is required to process text, images, voice, and video simultaneously with a power consumption of only a few watts. Qualcomm, MediaTek, and Huawei are all betting on this direction. If this scenario comes to fruition, it will spur the development of a new generation of smart hardware. Just like when the iPhone ushered in the era of smartphones. 3. AI-native scientific computing In the past, supercomputers were used to calculate weather and nuclear explosions. AI chips will become the mainstay of scientific computing in the future. Many scientific problems can be solved thousands of times faster using AI than using traditional methods. For example, AlphaFold predicts protein structures using AI chips. In the future, there may be AI chips specifically optimized for scientific computing. A new drug or a new material can be designed in just a few days. There are actually opportunities in this area domestically. There is a large demand for scientific research in China, and it is not significantly affected by international limitations on computing power. 4. Edge intelligent networks enable everything to be intelligent. Currently, IoT devices are all "dumb terminals," and data can only be analyzed after it is transmitted to the cloud. In the future, every device will have an AI chip that can run locally. For example, smart factories may emerge. Each machine has an AI chip, which can determine when maintenance is needed and when parameters need to be adjusted. The machines can also collaborate with each other. Some domestic chip companies are working in this direction, such as Horizon Robotics' Journey series. Predicting the next 3-5 years. The relationship between semiconductors and AI will change from "AI uses chips" to "AI redefines chips". Who can make breakthroughs in areas such as dedicated architecture, in-memory computing, and hardware-software synergy? Whoever can seize these new scenarios, such as embodied intelligence, edge multimodal computing, scientific computing, and edge intelligence. --- I reluctantly agreed to a roundtable discussion about AI and chips, but since I really didn't understand it, I had an AI summarize it.
Loading thread detail
Fetching the original tweets from X for a clean reading view.
Hang tight—this usually only takes a few seconds.