Why NVIDIA Israel Became a Global AI Power Center

The global AI economy runs on NVIDIA chips. A meaningful share of the infrastructure connecting those chips was designed in Israel.

The most valuable technology company in the world, with a market capitalization above $4 trillion, anchors its global hardware roadmap in Yokneam — a small town in northern Israel that most non-Israelis cannot find on a map. NVIDIA Israel employs more than 5,000 engineers. Three of the four major product lines on the company’s three-year roadmap are being developed under Israeli leadership. The networking technology that connects NVIDIA’s GPUs into the training clusters powering every frontier AI model in the world was designed in Israel.

The next AI arms race may be constrained less by chips than by the systems connecting them. That sentence is the thesis of this piece. NVIDIA understood it in 2019. The result is one of the most strategically important national-corporate relationships in the global technology economy.

The Mellanox Deal That Created an AI Empire

On March 11, 2019, NVIDIA announced a definitive agreement to acquire Mellanox Technologies for $125 per share in cash — a total enterprise value of approximately $6.9 billion. The deal closed in 2020. At the time, it was NVIDIA’s largest acquisition. It would also prove to be one of the most consequential in modern technology history.

Mellanox was an Israeli company headquartered in Yokneam, founded in 1999 by Eyal Waldman and a team of veterans from the Israeli semiconductor industry. The company pioneered InfiniBand interconnect technology — the high-speed networking fabric that connects servers in supercomputers. By 2019, Mellanox interconnects were used in more than 250 of the world’s TOP500 supercomputers.

The strategic logic: as AI workloads scale, data centers stop being collections of independent servers and become single, giant compute engines. In that architecture, the network between chips matters as much as the chips themselves. NVIDIA had the chips. Mellanox had the network.

What distinguished the acquisition from typical M&A was what happened next. NVIDIA did not absorb and dissolve Mellanox. It kept the Yokneam team intact, elevated it, and aggressively expanded headcount. Calcalist has estimated that the Israeli operation now contributes close to 20 percent of NVIDIA’s revenue. Whether that figure is precise or directional, the scale is unmistakable.

Why Israel?

Several countries have strong engineering bases. Few have all five conditions that produced NVIDIA Israel.

First, a deep semiconductor talent pool. Israel has been designing chips for forty years, since Intel established its first non-US fabrication facility in Haifa in 1985. Generations of Israeli engineers have built careers in chip design and hardware engineering. The country produces approximately 10,000 computer-science and electrical-engineering graduates annually from Technion, Tel Aviv University, the Weizmann Institute of Science, and a growing list of other institutions.

Second, the military pipeline. Unit 8200, the IDF’s signals intelligence corps, produces engineers who spent their formative years working on hardware-software integration at national-security scale. Units 9900 (visual intelligence) and the various Mamram programs add complementary talent. These are not theoretical engineers. They are operators.

Third, hardware-software integration culture. Israeli engineering is pragmatic, fast-moving, and unusually comfortable with the metal — the layer where firmware meets silicon. Networking is that layer. The cultural fit between Mellanox and NVIDIA was as important as the product fit.

Fourth, the Mellanox legacy itself. Twenty years of InfiniBand and Ethernet development built deep institutional knowledge in high-performance networking that NVIDIA bought, intact, in 2019.

Fifth, density of chip-design centers. Intel’s Habana, Amazon’s Annapurna, Apple, Google, and Microsoft all run substantial Israeli silicon operations within a 60-kilometer radius of Yokneam. The ecosystem is concentrated, the talent flows between companies, and the knowledge compounds.

Few other countries combine all five. Taiwan has the fabrication. The United States has the design depth. China has the scale. Israel has the integration of design talent, military hardware culture, and networking heritage in a small geographic footprint — exactly the combination NVIDIA needed.

What NVIDIA Israel Actually Builds

The Yokneam team’s product responsibility is networking infrastructure — the high-speed interconnects that link GPUs into AI training clusters. This is the part of the AI stack most casual observers ignore. It is also the part that has become NVIDIA’s strongest competitive moat.

Several flagship NVIDIA products are developed primarily by Israeli teams:

• The Spectrum-X networking platform — Ethernet networking optimized for AI workloads, used to interconnect thousands of GPUs in a single training cluster.
• The Spectrum-4 Ethernet switch, delivering 51.2 terabits per second of throughput.
• The BlueField-3 DPU and SuperNIC — network accelerators that offload data movement from the main CPU and GPU.
• The Spectrum-XGS architecture — networking that connects physically distant data centers into a single “super-factory,” announced in August 2025 from Yokneam.

Beyond networking, Yokneam teams develop central processors for data centers, systems-on-chip for robotics and automotive applications, autonomous-vehicle algorithms, and applied AI research. According to Israeli press accounts, Israeli teams contributed to the design of four of the six critical chips on NVIDIA’s Rubin system, the company’s next-generation AI computing platform.

NVIDIA Israel is to networking what NVIDIA Santa Clara is to GPUs. Both are essential. Neither is replaceable.

Why NVIDIA’s Israeli Networking Stack Matters for AI

Most public conversation about AI infrastructure focuses on GPUs. This is partial. Modern AI models are trained on clusters of thousands of GPUs operating as a single system. The networking that connects those GPUs determines the speed, efficiency, and cost of training.

A frontier-AI training run that costs hundreds of millions of dollars in compute time can complete days faster — or burn days slower — depending on networking efficiency. Networking translates directly into business strategy. It is no longer the side of the AI stack. It is increasingly the constraint.

In August 2025, NVIDIA Israel announced Spectrum-XGS — technology that extends NVIDIA’s networking beyond a single facility to connect physically distant data centers into a single “super-factory.” The announcement reflects a structural change: today’s largest data centers are approaching the physical limits of what a single facility can provide. Energy supply, cooling, and chip density all cap out. The frontier-AI labs training $10-billion-compute-budget models need capacity no single building can house.

Spectrum-XGS allows multiple data centers to function as a single computing system. In that architecture, networking is the binding constraint and the source of competitive advantage. Both belong to NVIDIA Israel.

The strategic significance: every major foundation-model training cluster currently in operation uses NVIDIA networking — Spectrum-X, BlueField, or InfiniBand. All three trace back to Israeli engineering teams in Yokneam. The next generation of AI compute architecture is being designed in Israel.

The Physical Footprint

Behind the product portfolio sits a physical infrastructure NVIDIA is building out at speed. Four components matter most.

Israel-1 supercomputer

In November 2023, NVIDIA Israel brought the Israel-1 supercomputer online from Yokneam, completing the first phase in under 20 weeks — two months ahead of schedule. The full configuration: 2,048 NVIDIA H100 GPUs supplied by Dell Technologies, 2,560 BlueField-3 DPUs, 80 Spectrum-4 switches, 8 exaflops of peak AI performance. Israel-1 is currently ranked #34 on the TOP500 list of the world’s most powerful supercomputers. A follow-on Blackwell-based cluster, reportedly a ~$500 million investment, is in development.

Mevo Carmel data center

A 30,000-square-meter facility planned for the Mevo Carmel industrial zone, with potential electrical capacity of up to 100 megawatts — Israel’s most powerful AI compute cluster. Combined with NVIDIA’s existing server-farm capacity in Ramot Menashe, the company will have Israeli compute resources sufficient to train and benchmark large AI models without depending on third-party cloud capacity.

Kiryat Tivon campus

Announced December 2025: a 160,000-square-meter campus on 90 dunams of state-owned land granted with exemption from public tender. Projected to employ 10,000 workers within five to six years — making NVIDIA the largest private-sector employer in Israel. NVIDIA has described the project as a “long-term, multibillion shekel investment.”

Beersheba expansion

Tripled R&D presence at the Gav Yam tech park in southern Israel, approximately 3,000 square meters, expected to be fully operational by the first half of 2026.

All four projects were announced or accelerated during the period of sustained conflict that began in October 2023. The decision to expand in Israel through active war is itself a strategic signal: NVIDIA’s investment thesis on Israel does not depend on regional stability in any narrow sense.

NVIDIA Israel in the Global AI Race

NVIDIA operates research centers in many countries. None matches the strategic weight of the Israeli operation. A quick comparison clarifies why.

Vs Santa Clara (headquarters): GPU architecture, AI research, business strategy. Israel handles a separate strategic layer — networking and interconnect — without which the GPUs cannot scale.

Vs Austin and other US sites: product engineering, software stacks, customer engineering. Important but downstream of the chip and networking design Israel leads.

Vs Taiwan: fabrication, packaging, supply chain. Critical — but Taiwan manufactures what others design. Israel designs.

Vs India: large engineering teams in Bangalore and Hyderabad, primarily software, customer support, and applied AI. NVIDIA India is scaling rapidly but does not yet drive flagship hardware product lines.

Vs China: no NVIDIA R&D footprint of consequence after US export controls reshaped the relationship. China is a customer market, not a development location.

Vs the Gulf AI buildout: Saudi Arabia, UAE, and Qatar are building gigawatt-class AI compute capacity. They are NVIDIA customers — large, important ones — but not NVIDIA development sites. The Gulf is buying the infrastructure that Israel designs.

Across the global map of where NVIDIA’s most strategically important work happens, two locations dominate: Santa Clara and Israel. That ranking has structural staying power.

The Israeli AI Infrastructure Cluster

NVIDIA Israel is the largest, most visible, and fastest-growing AI infrastructure operation in the country. It is not the only one.

Within roughly the same engineering ecosystem — drawing from the same universities, the same Unit 8200 talent, and the same vendor base — sit several of the most important chip operations in the world:

• Intel acquired Habana Labs for $2 billion in 2019. The Israeli team in Yokneam designs the Gaudi AI accelerator — Intel’s primary competitive answer to NVIDIA in AI training silicon.
• Amazon acquired Annapurna Labs in 2015. The Israeli team now drives Amazon’s custom silicon roadmap, including the Trainium and Inferentia AI chips that power large portions of AWS’s AI infrastructure.
• Apple runs substantial Israeli silicon operations, primarily focused on processor design and modem development.
• Google operates significant Israeli engineering with overlap into chip and AI infrastructure work.
• Microsoft maintains a large Israeli development presence that includes hardware and AI-infrastructure-adjacent work.

The cumulative effect: a disproportionate share of the world’s AI compute infrastructure — chips, interconnects, and the software that orchestrates both — is designed in Israel. This is not a coincidence. It is the result of a forty-year concentration of semiconductor expertise, a unique military-civilian talent pipeline, and the cultural fit between Israeli engineering and high-performance hardware work.

Israel became one of the core physical operating layers of the global AI economy.

The Talent Pipeline

NVIDIA Israel’s expansion depends on continued access to the Israeli engineering pipeline. The military pipeline — Unit 8200, 9900, Mamram — feeds the company at every level. The university pipeline produces approximately 10,000 graduates annually in computer science and electrical engineering across Technion, Tel Aviv University, the Weizmann Institute of Science, and other institutions.

NVIDIA’s Israeli development center is led by Amit Krig, senior vice president of software engineering. The AI research group is led by Professor Gal Chechik, who joined NVIDIA from Bar-Ilan University, where he remains a faculty member. Senior Israeli NVIDIA engineers now command total compensation packages competitive with their US counterparts — an inflection that has compressed mid-level wages across the broader Israeli market.

In December 2024, NVIDIA acquired Run:AI for approximately $700 million, adding Israeli AI infrastructure software to its portfolio. Around the same time, NVIDIA acquired Deci, contributing model optimization technology. Both teams now operate from within NVIDIA’s Israeli development centers. The pattern: NVIDIA is buying the Israeli AI ecosystem company by company.

What Comes Next

Three questions shape the next five years.

First, compute scaling. The Mevo Carmel data center and the planned Blackwell cluster move NVIDIA’s Israeli compute capacity into a new tier. Whether Israel can power that infrastructure at scale, and whether the Kiryat Tivon land-allocation model extends to additional sites, will determine how much of NVIDIA’s next-generation training infrastructure lives in Israel versus elsewhere.

Second, the Kiryat Tivon timeline. A 160,000-square-meter campus with 10,000 employees is a five-to-six-year project. Construction pace and recruitment ramp will be the most-watched indicators of NVIDIA’s long-term Israel commitment.

Third, the broader chip-design ecosystem. NVIDIA is not alone. The next decade of AI compute will be shaped by the concentration of NVIDIA, Intel, Amazon, Apple, Google, and Microsoft chip-design operations all running in the same small country.

NVIDIA Israel is not an isolated R&D center. It is part of a broader concentration of AI infrastructure engineering that increasingly shapes the global compute economy.

The chips most of the AI world depends on are designed in Santa Clara. The systems that connect them are designed in Yokneam. The next AI arms race may be constrained less by chips than by the systems connecting them — and the systems are Israeli.

Frequently Asked Questions

Why did NVIDIA acquire Mellanox?

NVIDIA acquired Mellanox in 2019 for $6.9 billion because the company’s high-speed interconnect technology — InfiniBand and Ethernet networking — was essential to AI infrastructure. As AI workloads scale, data centers become single compute engines where the network between chips matters as much as the chips themselves. The acquisition closed in 2020 and transformed Mellanox’s Yokneam headquarters into NVIDIA’s primary Israeli site.

What is Mellanox?

Mellanox Technologies was an Israeli company headquartered in Yokneam, founded in 1999 by Eyal Waldman and a team of Israeli semiconductor industry veterans. Mellanox pioneered InfiniBand interconnect technology and grew into the world’s leading provider of high-performance networking infrastructure for supercomputers and data centers. By 2019, Mellanox interconnects were used in more than 250 of the world’s TOP500 supercomputers. NVIDIA acquired Mellanox in 2020. The Yokneam headquarters became NVIDIA’s primary Israeli site.

Why does AI networking matter?

Modern frontier AI models are trained on clusters of thousands of GPUs operating as a single system. The networking that connects those GPUs determines the speed, efficiency, and cost of training. A training run costing hundreds of millions of dollars in compute time can complete days faster — or burn days slower — depending on networking efficiency. As AI training scales beyond what a single data center can support, networking becomes the binding constraint on AI growth. NVIDIA Israel leads the company’s networking development.

What is Spectrum-X?

Spectrum-X is NVIDIA’s Ethernet networking platform optimized for AI workloads. Developed primarily by NVIDIA’s Israeli teams in Yokneam, Spectrum-X is used to interconnect thousands of GPUs in a single training cluster. The platform pairs the Spectrum-4 Ethernet switch with the BlueField-3 SuperNIC network accelerator. Spectrum-X is currently used in the Israel-1 supercomputer and in NVIDIA’s enterprise AI infrastructure offerings globally.

Why is Yokneam important?

Yokneam is a town of approximately 25,000 residents in northern Israel that serves as the operational center of NVIDIA’s networking development. Approximately 3,000 NVIDIA employees work in Yokneam, designing the Spectrum-X platform, Spectrum-4 switches, BlueField-3 DPUs, and Spectrum-XGS architecture. Three of NVIDIA’s four major product lines for the next three years are being developed under the leadership of the Yokneam-based team.

How many people does NVIDIA employ in Israel?

NVIDIA Israel employs more than 5,000 engineers across seven research centers, with approximately 3,000 in Yokneam alone. The planned Kiryat Tivon campus is projected to employ 10,000 workers when complete, which would make NVIDIA the largest private-sector employer in Israel.

What is the Israel-1 supercomputer?

Israel-1 is NVIDIA’s first Israeli supercomputer, built by the company’s Israeli team and operational from Yokneam since November 2023. It features 2,048 NVIDIA H100 GPUs, more than 34 million CUDA cores, 2,560 BlueField-3 DPUs, and 80 Spectrum-4 switches. Peak performance reaches 8 exaflops for AI workloads. It is currently ranked #34 on the TOP500 list of global supercomputers. A follow-on Blackwell-based cluster, reportedly a ~$500 million investment, is in development.

How big is NVIDIA’s planned Kiryat Tivon campus?

The Kiryat Tivon campus is planned to span approximately 160,000 square meters on 90 dunams of state-owned land granted to NVIDIA with an exemption from public tender. It is projected to employ 10,000 workers and represents what NVIDIA has described as a “long-term, multibillion shekel investment.” When completed within five to six years, it will be Israel’s largest high-tech campus.

Has NVIDIA continued expanding in Israel during the war?

Yes. NVIDIA’s Israeli expansion has continued through the post-October 2023 period of sustained conflict. The Israel-1 supercomputer came online ahead of schedule in late 2023. The company acquired Run:AI for approximately $700 million in December 2024 and Deci shortly afterward. In 2025, NVIDIA announced the Beersheba expansion, the Kiryat Tivon campus, the Yokneam building expansion, and the Mevo Carmel data center. CEO Jensen Huang has publicly described Israel as the company’s “second home.”