Nvidia Israel to benefit from huge Saudi deal

Nvidia CEO Jensen Huang made a brief visit to Saudi Arabia last week alongside US President Donald Trump to sign one of the world’s largest chip deals ever with a single country. Nvidia will market 18,000 of the latest Blackwell Ultra graphics processors this year, at a total cost of $700 million, according to Bank of America, and subsequently about 200,000 chips over 5 years at a total cost of $7 billion.

Israeli industry got left behind when the US did not condition arms and technology supply agreements on the signing of the Abraham Accords, but Nvidia’s development and production centers in Israel will benefit from the huge deal and will be able to sell their wares in Saudi Arabia, even though it has decided not to recognize Israel.

At its development centers in Yokneam and Tel Aviv, Nvidia produces the communication chips that manage data traffic between graphics processors and core processors in data centers and accelerates processing of AI and cloud services. These communication chips and switches enable both communication within the server racks between the different servers and communication between the different racks, which are sometimes located a long distance from each other.

Since Huang signed a supply deal for the company’s newest chips, “Blackwell Ultra” (GB300) supplied in the configuration of server racks with 72 graphics processors (NVLink72), this means that Nvidia’s development center in Yokneam will play a central role in the production and supply of components to Saudi Arabia.

The NVLink server rack switch chips were first developed by Nvidia in Santa Clara, California, but with the acquisition of Mellanox in 2020, responsibility for product development was transferred to Israel because of the expertise the local development center has in designing communications between the various processors.

Nvidia is not alone in the huge deal with Saudi Arabia. Competitor AMD also promised to supply Crown Prince Mohammed bin Salman with its own graphics processors, although the terms of the deal were not published and in any case their supply will not begin before 2026, about a year after the start of work with Nvidia.

Trying to soften the image

In an interesting twist, this morning (Monday) at the Computex chip exhibition in Taiwan, Nvidia launched an NVLink server rack switch chip that for the first time allows the integration of chips from other companies that are not Nvidia’s. In doing so, Nvidia seeks to soften the monopolistic image it has created by allowing its customers to combine chips from competing companies in two possible configurations.

The first will allow the integration of AMD graphics processors such as the Mi400, Google graphics processors (TPU) and Amazon processors (Trinium and Infernasia) provided that they are managed by Nvidia core processors (Grace). The second configuration is the opposite – that is, it allows the integration of non-Nvidia core processors, such as those from Qualcomm and Fujitsu, with Nvidia graphics processors.

In both configurations, Nvidia communication processors will be responsible for communication within the server rack, meaning that NVIDIA will be able to profit at least from selling them, when they are packaged in the expensive server rack of its design. Nvidia understands that the industry takes graphics processors from anywhere, and if the server racks can be sold without graphics processors, but with a gallery of communications processors, the company will be able to profit both from the sale of communications chips and from its presence in AI processing centers, known as “AI factories.”

Nvidia servers can only be equipped with chips from companies that have specifically partnered with it, so AMD or Intel are still out of the picture. Those who rushed to join are MediaTek and Marvell in the graphics processor field, which are developing the core of Google and Amazon’s graphics processors, as well as companies such as Astra Labs and Alchip; Fujitsu and Qualcomm are partnering in the core processor field (CPU). Nvidia in Israel will also manage this project, because, as mentioned, it is responsible for the NVLink switch chip server racks, under the NVLink Fusion brand.

Huang’s U-turn

Huang is known for his skeptical approach to quantum technology, having estimated last January that it would take at least 15 years for the launch of the first effective quantum computer, and after repeating last March that many technologies from different physical worlds are competing for the development of the quantum computer – and hinting that it is a gamble.

Now, according to “The Information” website, it seems Nvidia is approaching its first venture capital investment in a quantum company: PsiQuantum, a US company with Australian roots, which has developed quantum computers based on shooting photons at a constant and predetermined rate.

PsiQuantum is known for its huge computers, the size of a football field, but it is considered in the quantum industry as one of the most advanced computing companies in the field. The company is currently raising $750 million, and Nvidia is estimated to be one of the investors. PsiQuantum has an Israeli connection: one of its first investors is the Pitango First Fund, alongside larger investors such as BlackRock and Microsoft.

Published by Globes, Israel business news – en.globes.co.il – on May 19, 2025.

© Copyright of Globes Publisher Itonut (1983) Ltd., 2025.