Nvidia SVP networking Gilad Shainer, who manages the field of communications chips, the company’s fastest growing segment, has had many triumphs. However, after the praise the Israeli development center received from CEO Jensen Huang at the CES conference, and the approval to build the huge campus in Kiryat Tivon, Shainer, who spoke to "Globes" to explain the surge in Israeli activity, wants to first point out a triumph of a different kind - the return home of Hamas hostage and Nvidia employee Avinan Or - an achievement that the company sought for two years and a week.
"After Or was taken hostage, the entire company rallied," Shainer recalls. "We supported the family, accompanied them to the protests, and CEO Jensen Huang kept in touch with his mother and supported them the entire time." The chapter closed last month, when Nvidia Israel management was at the company's headquarters in Santa Clara, apparently to discuss building the new campus in Kiryat Tivon.
At the time, Avinan Or and his partner Noa Argamani were in Los Angeles, and management seized the opportunity to bring Or and Huang together at Nvidia’s headquarters. "It was a very exciting meeting," recalls Shainer. "Jensen was excited to see him, and he was assured that his job in the office had been kept for him."
"The world's biggest communications company"
On the eve of being taken into captivity, Or worked on the Spectrum development team - a communications card that has become one of Nvidia's main growth engines in the past year, and one of its best-selling products. The chip, which enables direct connection between graphics processors in different servers by combining optical communications directly to the chip, is part of a quartet of products that Shainer oversees as a product manager. These chips are so crucial that CEO Jensen Huang said this week that they have made Nvidia "the world’s biggest communications company."
It is doubtful that Nvidia would be the world’s most valuable company with a market cap of $4.6 trillion, without its 2020 acquisition of Israel’s Mellanox for just $6.9 billion. Connectivity products, largely developed and managed in Yokneam, are the "butter on the bread" of the company’s graphics processors: Nvidia’s four networking chips are the highway that prevents bottlenecks and allows thousands of processors to work as one unit. Without this infrastructure, it would not be possible to power the new AI models, which process hundreds of billions of tokens, the basic processing units of AI, to meet global demand for this technology.
Networking chips are currently Nvidia’s fastest growth engine. In the third quarter of this year (according to data from November), revenue in the development segment in Israel jumped 164% compared with last year, totaling $8.2 billion. For comparison, the graphics processing unit (GPU) segment grew by "only" 56% in the corresponding quarter.
These numbers explain the praise that Jensen Huang showered on the Israeli team last week. "Four out of six chips that we develop came from Israel. It's an extraordinary team, dedicated, caring and hardworking," Huang said, stressing the rare stability of the Israeli center: "The employee turnover rate there is only 1%-2%. There are people there who have been working with us for 25 years, which is extraordinary."
Jensen Huang's expansive vision
Shainer oversees the product and is considered responsible for recruiting Nvidia's largest customers in the field - according to estimates, he works with Microsoft and OpenAI, as well as Elon Musk's xAI and Anthropic. The chip giant is expected to significantly expand its operations in Israel and hire 10,000 new employees. What's the secret?
Shainer explains, "A few years ago we focused on a single processor, but today the perspective has shifted to the entire data center, that is, the facility where the servers that these and other chips are embedded are located. The facility is designed from the ground up as a single unit, and this is where the critical importance of connectivity between the chips and between data centers on different sides of the world comes in. Jensen Huang recognized this even then, and that is why he acquired Mellanox."
In the age of AI, the old data centers are no longer enough. AI processing requires not only enormous computing power, but perfect synchronization between thousands of processors: every second that an expensive chip 'waits' for data is a huge financial loss for the customer. This is where Nvidia's chips come into the picture - the world's fastest 'network of roads and traffic lights', which prevents bottlenecks in the movement of data.
Today, Nvidia already supplies server racks that combine 72 graphics processors into a single computing unit, and the company is already preparing for the next generation that will contain 576 processors. "When you connect them all together with very fast communication and processing speeds, you get a rack that becomes a powerful supercomputer," says Shainer.
Scheiner explains: This is how we became the heart of the system
But Nvidia's challenge only begins in the server rack. To enable customers to train huge models, fast communication is needed between thousands of racks and millions of graphics processors around the world. "The goal is optimal parallel computing," he adds. "It's like scheduling a meeting between millions of people at the same time. If one is waiting for the other, it's a waste of resources that can cost billions. We make sure that no processor is starved for data."
Does this explain why Israeli networking product revenue is growing even faster than graphics processors?
"Absolutely. When we started with the previous generation processor, Hopper, we put eight processors in one server, connected it to other servers in the data center and that was it. Today, in the Blackwell generation, we already have 72 graphics processors in one server, and the need for data traffic has risen accordingly, as has the total number of processors in the entire data center. In Hopper, the connection rate was 400 gigabits per processor. Today it is 800 gigabits, and in the next generation, Rubin, the rate will double again to 1.6 terabits. For customers to be able to build training and inference systems on this scale, we must increase bandwidth in three dimensions simultaneously: within the server, between the servers, and between the entire data centers. This is the engine behind the extraordinary growth that we are seeing."
"This status gives the Israeli center a tremendous special weight within Nvidia," explains Shainer. "We build here the entire infrastructure that surrounds the graphics processor, everything that makes them a single supercomputer. The heart of Nvidia is in Israel, and it is very clear to the US side that we are not just building a few components here. The market already understands today that Nvidia's 'edge' is not in the single chip, but in the ability to make an entire data center play together in perfect synchronization. The company's weight has shifted to parallel computing capabilities, software libraries and system management and all this connectivity is born in Israel."
Israel cannot be left behind
Shainer describes a unique work pace at Nvidia, which has increased the pace of launches in recent years relative to rivals and is now launching new chips in extremely short development and production cycles: "At Mellanox, we would work on one generation, bring it out, and wait three years for the next generation, but today we are on an annual roadmap," he says. "As we launch, we’re already working on next year’s networks and the one after that. While we’re launching a new GPU, we’re developing five more supporting chips here at the same time. It’s an extremely high pace, and that’s why we’re growing so much." Nvidia’s development center now has over 5,000 employees, up from 4,500 last year.
Your gross profit margin is exceptionally high at 73%, which means you’re very profitable and also allow yourself to charge a high price to customers.
"The reality has changed, and today and in the future every data center will have to be equipped with graphics processors. It is true that companies like OpenAI and xAI are building huge, expensive systems, but there are also companies like CoreWeave and similar ones that are building AI clouds for organizations that need smaller models - whether it is for training an autonomous vehicle, financial models or robotics. A huge model is not always required; sometimes the model is trained in one place, and the inference is run at smaller sites."
Already today there are alternatives to your graphics processors from companies like Google and Broadcom, and replacements for communication processors, as offered by the Israeli company Drivenets.
"I don't know of a solution that comes close to the quality of what we have built. Building accelerated computing with a GPU isn't cheap. The CPU is the hub, but managing the communications in the data center also has a huge impact on the end result. If a customer tries to save 10% on the price and buys a replacement CPU, they could lose 20% of their processing power and that doesn't make economic sense. Huang always says that even if someone gives you their networking solution for free, it won't work well enough. In order not to lose millions and lose about 40% of the CPU's performance, you have to have perfect synchronization. That's why Nvidia is the world's largest communications company."
There are reports that a lack of electrical infrastructure and regulatory difficulties in setting up data centers are creating a bottleneck that is slowing down many of your efforts.
"I don’t see a halt, just more construction and data center systems that are only growing in size. When we place orders, they are destined for places where the data center is already ready to receive them. We make sure that they don’t sit in boxes but go live immediately."
Israel may be the "startup nsation," but it suffers from a lack of AI infrastructure. Leading local companies like AI21 or Decart are forced to train their models on servers abroad.
"Being at the forefront of AI is a strategic interest of every country, and it must be important for Israel too. We are making every effort to push this field here; We built the Israel-1 supercomputer, and we will continue to set up additional systems to allow access to startups and partners."
Are these closed labs, data centers for internal use only, like the ones you are establishing in Mevo Carmel?
"These are not closed labs, but open to collaborations with authorized partners such as startups and research institutions. We are happy to see companies rising that want to build AI systems in Israel. It is important to be at the forefront, and Israel simply cannot afford to be left behind."
Published by Globes, Israel business news - en.globes.co.il - on January 11, 2026.
© Copyright of Globes Publisher Itonut (1983) Ltd., 2026.