Broadcom launches Jericho4, the company’s latest network chip. This chip is specifically designed to support the growing demand for scalable AI infrastructure.
Where previous generations of chips focused primarily on speed and efficiency within data centers, Jericho4 raises the bar by also enabling the interconnection of multiple data centers into a single integrated AI system. This is reported by AInvest, based on multiple sources.
The Jericho4 chip makes it possible to connect more than a million processors across multiple data centers, even if they are far apart. This allows cloud providers to position smaller data centers geographically closer to customers while still functioning together as one large data center for training and running AI models.
Broadcom is thus responding directly to the trend toward distributed infrastructure for artificial intelligence, in which computing power must be brought closer to the user to reduce latency and increase efficiency.
Extensive security measures
The new chip processes approximately four times as much data as its predecessor and includes improvements aimed at managing high volumes of network traffic. Like AI processors from Nvidia and AMD, Jericho4 uses high-bandwidth memory (HBM) to temporarily store data during network congestion. This keeps the performance of AI environments stable, even during intensive use or under heavy loads.
In addition, the chip includes extensive security measures, including data encryption, which is crucial when sending sensitive information between data centers.
Jericho4 was developed using TSMC’s advanced three-nanometer process, which not only contributes to better performance but also to energy efficiency. The scale at which this chip can be deployed is impressive: a single system can contain approximately 4,500 Jericho4 chips. This provides hyperscalers with the building blocks to scale up AI capacity quickly without making significant changes to their physical infrastructure.
Many new data centers in the US
The introduction of the Jericho4 chip coincides with a sharp increase in the construction of data centers in the United States. According to available information, the number of data centers in the country will almost quadruple by the end of 2024, particularly in regions such as Northern Virginia and Maricopa County, Arizona.
The rapid expansion is primarily driven by the need to process huge amounts of data and run complex AI models. At the same time, this growth is raising concerns about the high electricity and water consumption of such facilities.