Nvidia's Israeli development: The dramatic move officially revealed
The chip giant presents the next generation of the artificial intelligence revolution with breakthrough technologies developed in Israel, and marks its next destination that will reach space.
Nvidia continues to prepare for the next phase of the artificial intelligence (AI) revolution, this time with an emphasis on AI agents capable of not only answering questions, but also planning, executing tasks, and acting in real-time. At the Hot Chips conference, the company presented a series of hardware and infrastructure developments designed to meet growing computing demands, centered on technologies originating from the company's development centers in Israel.
One of the key moves is the launch of NVIDIA Scale In, a communication infrastructure designed to support large AI factories and manage data traffic, security, and access to computing resources. Behind a significant part of this technology are the BlueField 4 and Spectrum X chip families, linked to Nvidia's development activity in Israel and the legacy of the Mellanox company.
BlueField 4 serves as a data processing unit and works together with DOCA software to offload infrastructure, security, and management tasks from the computing resources dedicated to AI. At the same time, Spectrum X provides the communication layer that connects computing systems and enables data transfer at a large scale.
As AI factories grow, it is not enough to add more processors. There is a need for an infrastructure that will allow systems to communicate with each other quickly, while maintaining security and efficient management of resources. This is where the Israeli angle comes in, as the knowledge and technologies developed in Nvidia's centers in Israel become part of the next-generation computing infrastructure.
Simultaneously, Nvidia introduced the Groq 3 LPX system, designed to accelerate inference tasks and capable of reaching a rate of 3,400 tokens per second in a test presented by the company. Vera CPU processors also took center stage, with 88 cores and a memory bandwidth of up to 1.2 terabytes per second.
One of the most prominent moves is the integration of the Vera Rubin platform into the Starmind satellite of SpaceXAI, a move that illustrates how AI infrastructures may move beyond Earth in the future.
These announcements illustrate the change the field is undergoing. If the first phase of AI focused mainly on models that manage a conversation with the user, the next phase requires systems capable of performing a sequence of actions independently. For this to happen, not only powerful AI chips are required, but also advanced communication, security, and data management systems. In this part, Nvidia's Israeli development continues to play a significant role.
