Mellanox Deal Terms, Israel-1 and NVIDIA’s Interim Leadership
NVIDIA agreed to buy Mellanox in March 2019 at a $6.9 billion enterprise value and completed the deal in April 2020 at a $7 billion transaction value. Eyal Waldman led Mellanox; Colette Kress, a finance executive, is NVIDIA’s interim CEO in 2026.
NVIDIA completed its purchase of Mellanox Technologies on April 27, 2020, at a transaction value of $7 billion, following a definitive agreement announced March 11, 2019, that put Mellanox’s enterprise value at about $6.9 billion. Those dates and figures are the fixed points in a longer technology history. The record shows a commercial acquisition of an interconnect supplier, named executives on both sides, and later networking products. It does not describe a change in corporate control by a government, and it does not make Mellanox’s founder or NVIDIA’s 2026 interim chief executive interchangeable figures. [1] [2] [4] [7]
The March 11, 2019, announcement set the original terms: $125 in cash for each Mellanox share and an enterprise value of approximately $6.9 billion. NVIDIA said it expected to close by the end of calendar 2019, subject to regulatory approvals, Mellanox shareholder approval and other customary conditions. The April 27, 2020, completion release used a different label, reporting a $7 billion transaction value and recalling the earlier announcement. The documents do not reconcile the $6.9 billion enterprise value with the $7 billion transaction value, so the two numbers should be kept separate rather than blended into one price. [1] [7]
NVIDIA’s stated rationale was technical. It said its computing platform and Mellanox’s interconnects powered more than 250 of the world’s TOP500 supercomputers, and that their customers included every major cloud service provider and computer maker. Mellanox was described as an early InfiniBand innovator whose InfiniBand and high-speed Ethernet products were used in more than half of the world’s fastest supercomputers and in many leading hyperscale data centers. NVIDIA argued that AI, scientific computing and analytics were pushing data centers toward designs with tens of thousands of compute nodes, where compute and networking have to be planned together. Those are company descriptions of the market and the deal’s purpose. [7]
In plain terms, Mellanox supplied the plumbing that moves data between servers and storage. Its product list included adapters, switches, software and silicon for Ethernet and InfiniBand networks. The March 2019 release describes InfiniBand as a high-performance interconnect technology and says Mellanox also sold high-speed Ethernet products; it places both in supercomputing and hyperscale data-center settings. The acquisition therefore added a company with both interconnect families to NVIDIA’s compute business. The release says Mellanox had pioneered InfiniBand and that its Ethernet products served supercomputers and hyperscale data centers. It does not show that every later NVIDIA networking feature came from Mellanox. [7]
Advertisement
The announcement named the leaders clearly. Jensen Huang was identified as NVIDIA’s founder and chief executive; Eyal Waldman was identified as Mellanox’s founder and chief executive. Huang described the need to connect many fast computing nodes through intelligent networking fabrics. Waldman described the combination as an extension of an existing partnership and a fit between performance-focused cultures. NVIDIA also said it intended to continue investing in Israeli talent and that customer sales and support would not change because of the transaction. Those were stated intentions at announcement, not evidence of how integration later unfolded. [7]
At completion, NVIDIA repeated the leadership identification: Huang as founder and CEO, Waldman as Mellanox’s founder and CEO. Waldman described complementary cultures, technology and ambitions and said each company would bring products into the other’s markets. Huang said the combined company would have technologies spanning AI computing and networking, with offerings from processors to software. NVIDIA said it expected the purchase to add to non-GAAP gross margin, earnings per share and free cash flow, after interest costs tied to $5 billion in notes it had issued. Those were company expectations, not an independent audit of the deal’s results. [1]
The 2026 governance change is a separate event. On Nov. 12, 2026, NVIDIA’s board said Huang had stepped down from the president, chief executive and director roles, with immediate effect. Kress, already executive vice president and chief financial officer, was named interim chief executive and will keep the finance post while the board searches for a permanent successor. The board appointed Harvey C. Jones as its first chair and said it expected business, customer work and the product plan to proceed without a break. NVIDIA said it would report third-quarter fiscal 2027 results on Nov. 17. [4]
Kress’s documented background is in finance and operations. She came to NVIDIA in 2013 after roughly a quarter-century in technology finance. She spent three years as finance chief for Cisco’s Business Technology and Operations Finance group. At Microsoft she worked for 13 years, including four as finance chief of the Server and Tools division, and she earlier worked in finance at Texas Instruments. She holds a finance bachelor’s from the University of Arizona and an MBA from Southern Methodist University. The record does not identify her as a Mellanox founder, a Mellanox executive or a networking engineer. [5]
Advertisement
The roles should not be blurred. Waldman led Mellanox at both the 2019 agreement and the 2020 completion. Kress is NVIDIA’s interim CEO and CFO in 2026. Gilad Shainer, identified in a later NVIDIA release as senior vice president of networking, has a networking-specific title; the release does not identify him as Mellanox’s founder, as NVIDIA’s chief executive or as a political actor. These are distinctions in corporate role, not matters of nationality or religion. The documents describe a commercial transaction and technical plans; they do not describe a government taking control of NVIDIA. [1] [2] [4] [5] [7]
NVIDIA’s later Ethernet work shows how the networking theme continued. On May 28, 2023, the company announced Spectrum-X, an accelerated Ethernet platform for AI clouds. NVIDIA put its Spectrum-4 switch together with BlueField-3 processors and software and said the combination delivered 1.7 times better overall AI performance and power efficiency. It also claimed consistent performance in multi-tenant environments, where several customers share infrastructure. Those are NVIDIA’s performance claims; the announcement does not present independent test results. [2]
The specifications were concrete. NVIDIA described Spectrum-4 as the first 51-terabit-per-second Ethernet switch built for AI networks. The design uses advanced RoCE features across the switches, processors and optical links to create a 400-gigabit Ethernet network for AI clouds. A single switch can connect 256 ports running at 200 gigabits per second, according to the announcement. The company said a fabric organized in two switching tiers can expand to 16,000 ports. Software includes Cumulus Linux, pure SONiC, NetQ and the DOCA framework for BlueField processors. The release says the platform uses standards-based Ethernet and can interoperate with Ethernet stacks. [2]
Israel-1 was the testbed attached to that platform. NVIDIA said the machine would sit in its Israeli data center and use Dell PowerEdge XE9680 servers, each built around an HGX H100 eight-GPU baseboard, plus BlueField-3 DPUs and Spectrum-4 switches. The company called Israel-1 a blueprint and testbed for Spectrum-X reference designs. The announcement did not give a completion date, a list of workloads, an availability schedule or independent performance results. It described an engineering vehicle for validating a networking design, not a finished public service. [2]
Advertisement
Shainer’s appearance in the Spectrum-X release is precise. NVIDIA identified him as senior vice president of networking and used his title in the announcement. The release also named Dell Technologies, Lenovo and Supermicro among companies offering Spectrum-X and said Spectrum-X, Spectrum-4 switches, BlueField-3 DPUs and 400G LinkX optics were available at the time. Those are product and ecosystem statements from NVIDIA, not independent market-share data. They do not establish how much of the platform originated inside Mellanox or how customers used it after launch. [2]
NVIDIA’s networking blog kept publishing Spectrum-X material years later. A post dated May 6, 2026, described Spectrum-X as an open, AI-native Ethernet fabric for gigascale AI. The author page lists earlier posts on BlueField DPUs, university research systems and high-performance computing. That record shows the platform remained part of NVIDIA’s public technical messaging after the 2023 launch. It remains company communication. The public posts do not provide a product-level audit trail linking a specific Mellanox product to a specific Spectrum-X feature. [2] [3]
The strategic significance is narrower and more concrete than a claim that Mellanox created NVIDIA’s networking business. The transaction added an interconnect supplier with InfiniBand and Ethernet product lines. The later Spectrum-X announcement shows NVIDIA marketing an Ethernet fabric for AI clouds, and the Israel-1 announcement shows a testbed for that design. What the documents do not show is a clean causal line from a Mellanox product to a Spectrum-X feature, or from the acquisition to every subsequent networking decision. NVIDIA’s stated thesis was that AI and data science require compute and networking to be designed together; the announcements are evidence of that thesis being stated and products being marketed, not proof of how each engineering choice was made. [1] [2] [7]
Several questions remain open. What was the integration timetable after the April 2020 close? How many Mellanox networking engineers stayed, and in what roles? Which Spectrum-X features, if any, depend on Mellanox-origin intellectual property rather than later NVIDIA development? What workloads run on Israel-1, and what results has NVIDIA measured beyond its 1.7-times performance and power-efficiency claim? Did the 2026 CEO transition change sponsorship, staffing or timetables for Israel-1 and Spectrum-X? These are questions, not findings. The absence of a public answer does not prove that no internal evidence exists; it means the announcements do not resolve the issue. [1] [2] [4]
Advertisement
Specific markers will show more. NVIDIA has scheduled its third-quarter fiscal 2027 results and an earnings call for Nov. 17. The transition announcement does not say the call will address Mellanox integration or Israel-1. The board’s search for a permanent CEO could confirm, revise or redirect the current roadmap. Readers should watch for any deployment update on Israel-1, customer adoption data for Spectrum-X and independent benchmarks of the 1.7-times claim. NVIDIA’s networking blog may continue to publish Spectrum-X material, but company posts are not independent evidence. The documented dates and values remain the fixed points: March 11, 2019; April 27, 2020; about $6.9 billion in enterprise value; $7 billion at completion. [1] [2] [3] [4] [7]
Sources
NVIDIA completion of Mellanox acquisition
NVIDIA Spectrum-X and Israel-1 announcement
NVIDIA official Gilad Shainer networking biography
NVIDIA CEO transition announcement
NVIDIA official Colette Kress biography
NVIDIA Rebellions SK Telecom infrastructure announcement
NVIDIA March2019 Mellanox acquisition agreement announcement
Prepared with AI assistance from the cited sources.
The Cali Reporter newsletter
The stories worth your time.
California, the country, and what comes next. Join our free newsletter for reporting that keeps you informed.
More in Artificial Intelligence
- Reuters Reports Second Unnamed Employee Relationship Involving Former NVIDIA CEO Huang
Reuters reported Nov. 12, citing two people familiar with the matter, that Jensen Huang, now NVIDIA's former CEO, had an intimate relationship with a second unnamed employee. The account is attributed, not adjudicated; Huang's Nov. 8 message admitted infidelity without naming anyone, and the board's stated investigation concerns reported foreign-intelligence concerns.
- House-Passed Remote Access Bill Awaits Senate Committee Action as Chip Licensing Rules Set Conditions
H.R.2683 remains a proposal, not law; Commerce’s January rule allows conditional case-by-case review for some exports and IaaS use, while senators seek records on diversion allegations.
- NVIDIA Sets Nov. 17 Earnings Release as Kress Leads Interim CEO Role
The fiscal 2027 third-quarter report is scheduled for Nov. 17; August revenue, margin, spending and cash-flow figures plus the company’s outlook provide concrete comparisons for readers.