Focus: artificial intelligence, chips and other industries
Welcome all guests to follow and forward
Daily Chip News
Issue
0622
❶Singapore
semiconductor company Silicon Box will invest $3.4 billion to produce small chips in Novara, Italy
Recently, Silicon Box, a semiconductor company based in Singapore, plans to choose Novara, located in the industrialized Piedmont region of northwest Italy, as the site of its new multi-billion euro chip factory in Italy. The three-year-old startup was founded by the founder of the US chipmaker Marvell. Under a government-backed agreement, Silicon Box will invest 3.2 billion euros ($3.4 billion) to produce so-called "chiplets" in Italy, which are as tiny as grains of sand. The chips are combined through advanced packaging, a cost-effective way to combine small semiconductors into a processor that can power everything from data centers to home appliances.
❷The
AI search company founded by former Baidu executive Jing Kun in the United States received 435 million in first round financing, with the latest valuation of nearly 1.9 billion
Recently, MainFunc, an AI innovation product company co-founded by Eric Jing, former vice president of Baidu Group and former CEO of Xiaodu Technology, announced the completion of its first seed round of financing of US$60 million (approximately RMB 435 million), led by Singapore fund Lanchi Ventures, with a company valuation of US$260 million (approximately RMB 1.887 billion). At the same time, Genspark, MainFunc's first AI Agent search project (company), was officially released and launched in Beta, aiming to use AI to provide a better search experience. Jui Tan, the lead investor in this round and managing partner of Lanchi Ventures, said that Genspark's approach was "really eye-catching" and expressed his confidence in the technical direction of Eric Jing and Zhu Kaihua, noting that the two had extensive experience in building AI and search products before.
❸Hanwei
Technology's joint-stock company Tairui Digital Technology announced the open source of its self-developed semantic entity model data format
Recently, Terry Digital, a company in the digital twin field of Hanwei Technology, announced that it will officially open source the Semantic Entity Model (SEM) data format developed independently to promote the management and sharing of spatial semantic entity data. SEM is a format for storing and exchanging spatial entity model data based on semantic descriptions. It defines the types and relationships of spatial entity objects, and also supports the attachment of unstructured existing data to spatial entity objects, such as artificial models, point cloud models, video files, etc.
❹
Huawei's Ma Jindou: New products will be launched in the second half of the year to address shortcomings in vehicle-road-cloud, roadside and perception
On June 20, Ma Jindou, president of Huawei's wireless network product line, said at the CICV 2024 Technology Week International Side Event yesterday that Huawei will launch new products in the second half of this year to solve the shortcomings of roadside and perception of vehicle-road-cloud. "We have found that a lot of shortcomings are in roadside and perception. We have made the communication latency very low, but we now realize that the latency of our edge computing and platform computing in the perception part, including radar, cameras, etc., is still very high. We now hope to contribute to this area, so that the information on the roadside can be seen more clearly, the calculation is faster, and the service is more comprehensive. In this regard, we will launch new products in the second half of this year to help integrate vehicle-road-cloud and help our partners do this well.
❶
Nvidia spends $1.3 billion to buy HBM3E to seize the "monopoly computing power" business opportunity in H1 this year
Recently, in order to ensure the smooth shipment of GH200 and H200 this year, AI chip giant NVIDIA has opened a budget of US$1.3 billion to reserve part of the new high-bandwidth storage HBM3e production capacity from Micron and SK Hynix. The credibility of this budget figure remains to be confirmed, but even if it is impossible to book the global HBM production capacity this year, it can help NVIDIA seize the business opportunity of "monopolizing computing power" in the first half of this year, because if the product is launched first, it can grab the market share first. According to packaging data, NVIDIA has booked CoWoS production capacity of more than 140,000 wafers this year, of which TSMC has won 120,000 orders and Amkor has been allocated 20,000 to 30,000 pieces, corresponding to a total GPU production capacity of nearly 4.5 million. According to the 1:6 ratio of each GPU logic chip and storage particle, NVIDIA needs about 27 million HBMs this year. Based on the cost of $250 per chip, NVIDIA's annual procurement of HBM chips will cost $6.8 billion, not just $1.3 billion.
❷
AI startup Anthropic launches a new generation of large models
On June 20, three months after the launch of the Claude 3 series of AI models, Anthropic, a startup backed by Google and Amazon, said it would release an updated large model, the Claude 3.5 Sonnet. Compared with the Claude 3 Opus, Anthropic's latest product scored higher in benchmark tests, runs about twice as fast as the former, and costs only one-fifth of the price for software developers. Anthropic said it plans to release more AI models this year, including the Claude 3.5 Opus. Its CEO Amodei said: "We hope to speed up the release cycle as much as possible, but the premise is that it is in line with our security values." Anthropic launched the Claude 3 series of AI models and new chatbots in early March this year, including three models: Opus, Sonnet and Haiku.
❸Tiantong
Fulian Intelligent Driving Domain Controller Project Landed in Lingang New Area
Recently, Tiantong Fulian plans to invest in the construction of a domain controller production base in the Lingang New Area, including smart driving domain controllers and other automotive electronic products. Tiantong Fulian will integrate the advantages of shareholder cloud edge computing and smart manufacturing, as well as self-developed autonomous driving software algorithms, efficient engineering delivery capabilities, and customer resources to build the first software and hardware integrated smart driving domain controller production base in the Lingang New Area. According to Lingang Group, the annual output value of the project will exceed 6 billion yuan after reaching full production. The domain controller products independently developed, designed and manufactured by Tiantong Fulian can be adapted to multiple mature chip platforms, including NVIDIA, Horizon, Texas Instruments, NXP and Qualcomm.
The articles and pictures published on this public account are from the Internet and are for communication purposes only. If there is any infringement, please contact us and reply. We will deal with it within 24 hours after receiving the information.
京公网安备 11010802033920号