Nvidia to Build Israeli Supercomputer as AI Demand Soars

SOURCE: REUTERS

By Steven Scheer

JERUSALEM, May 29 (Reuters) - Nvidia Corp said on Monday it was building Israel's most powerful artificial intelligence (AI) supercomputer to meet soaring customer demand for AI applications.

Nvidia, the world's most valuable listed chip company, said the cloud-based system would cost hundreds of millions of dollars and be partly operational by the end of 2023.

Gilad Shainer, a senior vice president at Nvidia, said Nvidia worked with 800 startups in Israel and tens of thousands of software engineers.

The system, called Israel-1, is expected to deliver performance of up to eight exaflops of AI computing to make it one of the world's fastest AI supercomputers. One exaflop has the ability to perform 1 quintillion - or 1,000,000,000,000,000,000 - calculations per second.

Shainer said AI was the "most important technology in our lifetime" and that to develop AI and generative AI applications large graphics processing units (GPUs) were needed.

"Generative AI is going everywhere nowadays. You need to be able to run training on large datasets," he told Reuters, noting companies in Israel will have access to a supercomputer they don't have today.

"This system is a large scale system that actually will enable them to do training much quicker, to build frameworks and build solutions that can tackle more complex problems."

OpenAI's ChatGPT, for example, was created with thousands of Nvidia GPUs.

The system was developed by the former Mellanox team. Nvidia bought Israeli chip designer Mellanox Technologies in 2019 for nearly $7 billion, outbidding Intel Corp.

Shainer said Nvidia's first priority for the supercomputer was its Israeli partners. "We may use this system to work with partners outside of Israel down the road," he said.

Last week, Nvidia said it had worked with Britain's University of Bristol to build a new supercomputer using a new Nvidia chip that would compete with Intel and Advanced Micro Devices Inc. (Reporting by Steven Scheer Editing by Mark Potter)